chore: 升级版本至 0.7.0 并启用 regex unicode-perl 特性

- 新增 `--no-generate` 参数，强制使用模板生成而非 AI
- 调整 `--yes` 参数仅跳过交互提示，不改变生成行为
- 统一使用工具函数替代 println 输出

CHANGELOG.md

@@ -9,6 +9,32 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 暂无。
 
+## [0.6.1] - 2026-08-18
+
+### 🔧 其他变更
+- 依赖瘦身：移除未使用的直接依赖 `config`、`handlebars`、`shell-words`、`walkdir`、`console`、`lazy_static`、`atty`，crate 总数由 341 降至 296，release 二进制体积由 7.38 MB 降至 7.00 MB（-5.2%）
+- `lazy_static!` 宏替换为标准库 `LazyLock`；`atty` 替换为标准库 `std::io::IsTerminal`
+- `tokio` 特性由 `full` 精简为 `macros` + `rt-multi-thread`
+- `regex` 关闭默认 unicode 特性（保留 `std` + `perf`），行为变化：`\s`/`\d`/`\w` 字符类仅匹配 ASCII，对提交信息、版本号、邮箱、GPG Key ID 校验无实质影响
+
+## [0.6.0] - 2026-08-17
+
+### ✨ 新功能
+- LLM 层整体迁移至 rig-core 0.41：六家 provider（Ollama / OpenAI / Anthropic / Kimi / DeepSeek / OpenRouter）统一由 rig 原生客户端驱动，删除约 2100 行手写 HTTP/SSE/重试代码
+- 新 LLM 门面（`src/llm/rig/`）：provider 枚举 + 单一生成入口 + 凭据校验（rig VerifyClient）+ 错误映射层；提示词与提交解析拆分至独立模块
+- Anthropic 支持自定义 `base_url`（此前被静默忽略，本次修复）
+- 新增基于 rig mock 后端的 LLM 层离线测试（请求形状、流式事件、错误映射），并附 6 家 provider 的 `#[ignore]` 实网冒烟用例
+
+### 🐞 错误修复
+- 各 provider 超时统一由 `llm.timeout` 配置控制（此前默认超时 60/120/300 秒不一致）
+- 消除 Ollama 客户端构造时的 panic（`expect` 式构造）
+
+### 🔧 其他变更（行为变化）
+- Ollama 请求端点由 `/api/generate` 改为 `/api/chat`（功能等价；自定义 Ollama 代理需兼容 chat 端点）
+- 非流式请求不再做应用层重试（此前按错误文案字符串匹配重试 3 次）；流式请求由 rig 的 SSE 机制自动重连
+- 可用性校验端点变化：DeepSeek 改为 `/user/balance`、OpenRouter 改为 `/key`、Anthropic 改为 `/v1/models`（对外表现为 `LLM provider 'xxx' is not available` 语义不变）
+- 依赖瘦身：移除 `reqwest`(0.12) 与 `async-trait` 直接依赖，不引入 rig agent/fastembed/lancedb 等组件
+
 ## [0.5.0] - 2026-07-24
 
 ### ✨ 新功能

Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "quicommit"
-version = "0.5.5"
+version = "0.7.0"
 edition = "2024"
 authors = ["Sidney Zhang <zly@lyzhang.me>"]
 description = "A powerful Git assistant tool with AI-powered commit/tag/changelog generation"
@@ -19,11 +19,9 @@ path = "src/main.rs"
 clap = { version = "4.5", features = ["derive", "env", "wrap_help"] }
 clap_complete = "4.5"
 dialoguer = "0.11"
-console = "0.15"
 indicatif = "0.17"
 
 # Configuration management
-config = "0.14"
 serde = { version = "1.0", features = ["derive"] }
 toml = "0.8"
 dirs = "5.0"
@@ -32,9 +30,7 @@ dirs = "5.0"
 git2 = "0.20.3"
 which = "6.0"
 
-# HTTP client for LLM APIs
-reqwest = { version = "0.12", features = ["json", "rustls-tls", "stream"], default-features = false }
-tokio = { version = "1.35", features = ["full"] }
+tokio = { version = "1.35", features = ["macros", "rt-multi-thread"] }
 
 # Error handling
 thiserror = "1.0"
@@ -46,21 +42,16 @@ tracing-subscriber = { version = "0.3", features = ["env-filter", "fmt"] }
 
 # Utilities
 chrono = { version = "0.4", features = ["serde"] }
-regex = "1.10"
+regex = { version = "1.10", default-features = false, features = ["std", "perf", "unicode-perl"] }
 roxmltree = "0.20"
-lazy_static = "1.4"
 colored = "2.1"
-handlebars = "5.1"
 semver = "1.0"
-walkdir = "2.4"
 tempfile = "3.9"
 sha2 = "0.10"
 hex = "0.4"
 textwrap = "0.16"
-async-trait = "0.1"
 futures-util = "0.3"
 serde_json = "1.0"
-atty = "0.2"
 
 # Encryption for sensitive data (SSH keys, GPG, etc.)
 aes-gcm = "0.10"
@@ -74,8 +65,8 @@ keyring = { version = "3.6.3", features = ["apple-native", "windows-native", "sy
 # Interactive editor
 edit = "0.1"
 
-# Shell completion generation
-shell-words = "1.1"
+# LLM integration (rig-core only: HTTP backend + rustls; no agent/derive)
+rig-core = { version = "0.41", default-features = false, features = ["reqwest", "rustls"] }
 
 [dev-dependencies]
 assert_cmd = "2.0"
@@ -83,6 +74,9 @@ predicates = "3.1"
 tempfile = "3.9"
 mockall = "0.12"
 wiremock = "0.6"
+# rig mock HTTP backends for LLM layer tests
+rig-core = { version = "0.41", default-features = false, features = ["test-utils"] }
+http = "1"
 
 [profile.release]
 opt-level = "s"

README.md

@@ -236,6 +236,10 @@ quicommit profile token
 quicommit config set-llm ollama
 quicommit config set-llm ollama --base-url http://localhost:11434 --model llama2
 
+> ⚠️ **Security note**: passing your API key as a CLI argument puts it in your shell
+> history and process list. Prefer running `quicommit config set-api-key` (or
+> `config set-llm`) without the key to enter it via a hidden prompt instead.
+
 # Configure OpenAI
 quicommit config set-llm openai
 quicommit config set-api-key YOUR_API_KEY
@@ -329,7 +333,7 @@ quicommit config reset --force
 | `--commitlint` | Use commitlint format |
 | `--no-verify` | Skip commit message verification |
 | `-t, --think` | Enable LLM thinking/reasoning mode (overrides config) |
-| `-y, --yes` | Skip confirmation |
+| `-y, --yes` | Skip confirmation prompts only (generation behavior unchanged) |
 | `--push` | Push after committing |
 | `--remote` | Specify remote repository (default: origin) |
 
@@ -349,7 +353,7 @@ quicommit config reset --force
 | `-r, --remote` | Specify remote repository (default: origin) |
 | `--dry-run` | Dry run |
 | `-t, --think` | Enable LLM thinking/reasoning mode (overrides config) |
-| `-y, --yes` | Skip confirmation |
+| `-y, --yes` | Skip confirmation prompts only (generation behavior unchanged) |
 
 ### Changelog Options
 
@@ -366,7 +370,8 @@ quicommit config reset --force
 | `--format` | Format (keep-a-changelog, github-releases) |
 | `--dry-run` | Dry run (output to stdout) |
 | `--think` | Enable LLM thinking/reasoning mode (overrides config) |
-| `-y, --yes` | Skip confirmation |
+| `--no-generate` | Generate deterministically from template (no AI call) |
+| `-y, --yes` | Skip confirmation prompts only (generation behavior unchanged) |
 
 ## Configuration File
 
@@ -425,6 +430,9 @@ auto_generate = true
 path = "CHANGELOG.md"
 auto_generate = true
 
+[output]
+emoji = true
+
 [repo_profiles]
 "/path/to/work/project" = "work"
 "/path/to/personal/project" = "personal"
@@ -436,7 +444,15 @@ auto_generate = true
 |----------|-------------|
 | `QUICOMMIT_CONFIG` | Configuration file path |
 | `EDITOR` | Default editor |
-| `NO_COLOR` | Disable colored output |
+| `NO_COLOR` | Disable colored output and emoji decorations (any value) |
+| `RUST_LOG` | Overrides `-v` log filtering (e.g. `debug`) |
+
+## Global Options
+
+- `-v, --verbose` (repeatable): `-v` shows info logs, `-vv` debug, `-vvv` trace.
+- `--no-color`: disable colors **and** emoji decorations (highest priority).
+- `--emoji` / `--no-emoji`: force emoji decorations on/off, overriding the `output.emoji` config value.
+- Emoji decorations can also be configured via `config set output.emoji true|false`.
 
 ## Troubleshooting
 
@@ -453,7 +469,7 @@ quicommit config set llm.provider ollama
 # Get configuration value
 quicommit config get llm.provider
 
-# Set API key (stored in system keyring)
+# Set API key (stored in system keyring; omit the value for hidden prompt input)
 quicommit config set-api-key YOUR_API_KEY
 
 # Delete API key from keyring
@@ -542,13 +558,12 @@ src/
 │   ├── commit.rs
 │   ├── tag.rs
 │   └── changelog.rs
-├── llm/         # LLM provider implementations
-│   ├── ollama.rs
-│   ├── openai.rs
-│   ├── anthropic.rs
-│   ├── kimi.rs
-│   ├── deepseek.rs
-│   └── openrouter.rs
+├── llm/         # LLM integration (rig-core based)
+│   ├── mod.rs
+│   ├── prompts.rs
+│   ├── parsing.rs
+│   ├── thinking.rs
+│   └── rig/      # rig provider facade
 ├── i18n/        # Internationalization support
 │   ├── messages.rs
 │   └── translator.rs

readme_zh.md

@@ -230,6 +230,10 @@ quicommit profile token
 quicommit config set-llm ollama
 quicommit config set-llm ollama --base-url http://localhost:11434 --model llama2
 
+> ⚠️ **安全提示**：把 API 密钥作为命令行参数传入会留在 shell history 与进程列表中。
+> 建议直接运行 `quicommit config set-api-key`（或 `config set-llm`）不携带密钥，
+> 以隐藏方式输入。
+
 # 配置OpenAI
 quicommit config set-llm openai
 quicommit config set-api-key YOUR_API_KEY
@@ -323,7 +327,7 @@ quicommit config reset --force
 | `--commitlint` | 使用commitlint格式 |
 | `--no-verify` | 不验证提交信息 |
 | `-t, --think` | 启用 LLM 思考/推理模式（覆盖配置） |
-| `-y, --yes` | 跳过确认提示 |
+| `-y, --yes` | 仅跳过确认提示（生成行为不变） |
 | `--push` | 提交后推送到远程 |
 | `--remote` | 指定远程仓库（默认：origin） |
 
@@ -343,7 +347,7 @@ quicommit config reset --force
 | `-r, --remote` | 指定远程仓库（默认：origin） |
 | `--dry-run` | 试运行 |
 | `-t, --think` | 启用 LLM 思考/推理模式（覆盖配置） |
-| `-y, --yes` | 跳过确认提示 |
+| `-y, --yes` | 仅跳过确认提示（生成行为不变） |
 
 ### changelog命令选项
 
@@ -360,7 +364,8 @@ quicommit config reset --force
 | `--format` | 格式（keep-a-changelog、github-releases） |
 | `--dry-run` | 试运行（输出到stdout） |
 | `--think` | 启用 LLM 思考/推理模式（覆盖配置） |
-| `-y, --yes` | 跳过确认提示 |
+| `--no-generate` | 使用模板确定性生成（不调用 AI） |
+| `-y, --yes` | 仅跳过确认提示（生成行为不变） |
 
 ## 配置文件
 
@@ -419,6 +424,9 @@ auto_generate = true
 path = "CHANGELOG.md"
 auto_generate = true
 
+[output]
+emoji = true
+
 [repo_profiles]
 "/path/to/work/project" = "work"
 "/path/to/personal/project" = "personal"
@@ -430,7 +438,15 @@ auto_generate = true
 |--------|------|
 | `QUICOMMIT_CONFIG` | 配置文件路径 |
 | `EDITOR` | 默认编辑器 |
-| `NO_COLOR` | 禁用彩色输出 |
+| `NO_COLOR` | 禁用彩色与 Emoji 装饰输出（任意值即生效） |
+| `RUST_LOG` | 覆盖 `-v` 的日志过滤（例如 `debug`） |
+
+## 全局选项
+
+- `-v, --verbose`（可叠加）：`-v` 显示 info 日志、`-vv` debug、`-vvv` trace。
+- `--no-color`：禁用彩色**与** Emoji 装饰（最高优先级）。
+- `--emoji` / `--no-emoji`：强制开启/关闭 Emoji 装饰，覆盖 `output.emoji` 配置。
+- Emoji 装饰也可通过 `quicommit config set output.emoji true|false` 配置。
 
 ## 故障排除
 
@@ -447,7 +463,7 @@ quicommit config set llm.provider ollama
 # 获取配置值
 quicommit config get llm.provider
 
-# 设置API密钥（存储在系统密钥环中）
+# 设置API密钥（存储在系统密钥环中；省略密钥值可通过隐藏提示输入）
 quicommit config set-api-key YOUR_API_KEY
 
 # 从密钥环删除API密钥
@@ -536,13 +552,12 @@ src/
 │   ├── commit.rs
 │   ├── tag.rs
 │   └── changelog.rs
-├── llm/         # LLM提供商实现
-│   ├── ollama.rs
-│   ├── openai.rs
-│   ├── anthropic.rs
-│   ├── kimi.rs
-│   ├── deepseek.rs
-│   └── openrouter.rs
+├── llm/         # LLM 集成（基于 rig-core）
+│   ├── mod.rs
+│   ├── prompts.rs
+│   ├── parsing.rs
+│   ├── thinking.rs
+│   └── rig/      # rig provider 门面
 ├── i18n/        # 国际化支持
 │   ├── messages.rs
 │   └── translator.rs

src/commands/changelog.rs

@@ -11,6 +11,7 @@ use crate::git::GitRepo;
 use crate::git::find_repo;
 use crate::git::{CommitInfo, changelog::*};
 use crate::i18n::{Messages, translate_changelog_category};
+use crate::utils::{print_progress, print_success, print_warning};
 
 /// Generate changelog
 #[derive(Parser)]
@@ -60,7 +61,11 @@ pub struct ChangelogCommand {
     #[arg(long)]
     think: bool,
 
-    /// Skip interactive prompts
+    /// Generate deterministically from template (no AI call)
+    #[arg(long, conflicts_with = "generate")]
+    no_generate: bool,
+
+    /// Skip interactive prompts only (generation behavior unchanged)
     #[arg(short = 'y', long)]
     yes: bool,
 }
@@ -85,7 +90,7 @@ impl ChangelogCommand {
                 .unwrap_or_else(|| PathBuf::from(&config.changelog.path));
 
             init_changelog(&path)?;
-            println!("{}", messages.initialized_changelog(&format!("{:?}", path)));
+            print_success(&messages.initialized_changelog(&path.display().to_string()));
             return Ok(());
         }
 
@@ -101,10 +106,7 @@ impl ChangelogCommand {
             Some("keep") | Some("keep-a-changelog") => ChangelogFormat::KeepAChangelog,
             Some("custom") => ChangelogFormat::Custom,
             None => ChangelogFormat::KeepAChangelog,
-            Some(f) => bail!(
-                "Unknown format: {}. Use: keep-a-changelog, github-releases",
-                f
-            ),
+            Some(f) => bail!("{}", messages.unknown_changelog_format(f)),
         };
 
         // Get version
@@ -120,7 +122,7 @@ impl ChangelogCommand {
         };
 
         // Get commits
-        println!("{}", messages.fetching_commits());
+        print_progress(messages.fetching_commits());
 
         // Determine from_tag: use explicit --from, or auto-detect from changelog
         let from_tag = self.resolve_from_tag(&repo, &output_path, &messages);
@@ -130,10 +132,10 @@ impl ChangelogCommand {
             bail!("{}", messages.no_commits_found());
         }
 
-        println!("{}", messages.found_commits(commits.len()));
+        print_success(&messages.found_commits(commits.len()));
 
         // Generate changelog
-        let changelog = if self.generate || (config.changelog.auto_generate && !self.yes) {
+        let changelog = if self.generate || (config.changelog.auto_generate && !self.no_generate) {
             self.generate_with_ai(&version, &commits, &messages).await?
         } else {
             self.generate_with_template(format, &version, &commits, language)?
@@ -161,12 +163,12 @@ impl ChangelogCommand {
             println!("{}", "─".repeat(60));
 
             let confirm = Confirm::new()
-                .with_prompt(messages.write_to_file(&format!("{:?}", output_path)))
+                .with_prompt(messages.write_to_file(&output_path.display().to_string()))
                 .default(true)
                 .interact()?;
 
             if !confirm {
-                println!("{}", messages.cancelled().yellow());
+                print_warning(messages.cancelled());
                 return Ok(());
             }
         }
@@ -185,7 +187,11 @@ impl ChangelogCommand {
             std::fs::write(&output_path, content)?;
         }
 
-        println!("{} {:?}", messages.changelog_written(), output_path);
+        print_success(&format!(
+            "{} {}",
+            messages.changelog_written(),
+            output_path.display()
+        ));
 
         Ok(())
     }
@@ -221,12 +227,13 @@ impl ChangelogCommand {
         let manager = ConfigManager::new()?;
         let language = manager.get_language().unwrap_or(Language::English);
 
-        println!("{}", messages.ai_generating_changelog());
-
         let generator = ContentGenerator::new_with_think(&manager, self.think, None).await?;
-        generator
+        let spinner = crate::utils::Spinner::start(messages.ai_generating_changelog());
+        let result = generator
             .generate_changelog_entry(version, commits, language)
-            .await
+            .await;
+        spinner.finish_clear();
+        result
     }
 
     fn generate_with_template(

src/commands/commit.rs

@@ -11,6 +11,7 @@ use crate::git::commit::{CommitBuilder, create_date_commit_message};
 use crate::git::{GitRepo, find_repo};
 use crate::i18n::Messages;
 use crate::utils::validators::get_commit_types;
+use crate::utils::{print_progress, print_success, print_warning};
 
 /// Generate and execute conventional commits
 #[derive(Parser)]
@@ -75,7 +76,7 @@ pub struct CommitCommand {
     #[arg(short = 't', long)]
     think: bool,
 
-    /// Skip interactive prompts
+    /// Skip interactive prompts only (generation behavior unchanged)
     #[arg(short = 'y', long)]
     yes: bool,
 
@@ -121,13 +122,13 @@ impl CommitCommand {
         // Auto-add if no files are staged and there are unstaged/untracked changes
         if status.staged == 0 && (status.unstaged > 0 || status.untracked > 0) && !self.all {
             println!("{}", messages.auto_stage_changes().yellow());
-            let removed = repo.stage_all()?;
+            let removed = repo.stage_all(&messages)?;
             println!("{}", messages.staged_all().green());
             if !removed.is_empty() {
-                println!(
-                    "{}",
-                    format!("Removed {} ignored files from staging:", removed.len()).yellow()
-                );
+                print_warning(&format!(
+                    "Removed {} ignored files from staging:",
+                    removed.len()
+                ));
                 for file in &removed {
                     println!("  • {}", file);
                 }
@@ -136,19 +137,19 @@ impl CommitCommand {
             // Re-check status after staging to ensure changes are detected
             let new_status = repo.status_summary()?;
             if new_status.staged == 0 {
-                bail!("Failed to stage changes. Please try running 'git add -A' manually.");
+                bail!("{}", messages.failed_to_stage());
             }
         }
 
         // Stage all if requested
         if self.all {
-            let removed = repo.stage_all()?;
+            let removed = repo.stage_all(&messages)?;
             println!("{}", messages.staged_all().green());
             if !removed.is_empty() {
-                println!(
-                    "{}",
-                    format!("Removed {} ignored files from staging:", removed.len()).yellow()
-                );
+                print_warning(&format!(
+                    "Removed {} ignored files from staging:",
+                    removed.len()
+                ));
                 for file in &removed {
                     println!("  • {}", file);
                 }
@@ -194,29 +195,21 @@ impl CommitCommand {
                 .interact()?;
 
             if !confirm {
-                println!("{}", messages.commit_cancelled().yellow());
+                print_warning(messages.commit_cancelled());
                 return Ok(());
             }
         }
 
         let result = if self.amend {
             if self.dry_run {
-                println!(
-                    "\n{} {}",
-                    messages.dry_run(),
-                    "- commit not amended.".yellow()
-                );
+                println!("\n{}", messages.dry_run_commit_not_amended().yellow());
                 return Ok(());
             }
             self.amend_commit(&repo, &commit_message)?;
             None
         } else {
             if self.dry_run {
-                println!(
-                    "\n{} {}",
-                    messages.dry_run(),
-                    "- commit not created.".yellow()
-                );
+                println!("\n{}", messages.dry_run_commit_not_created().yellow());
                 return Ok(());
             }
             CommitBuilder::new()
@@ -226,30 +219,34 @@ impl CommitCommand {
         };
 
         if let Some(commit_oid) = result {
-            println!(
+            print_success(&format!(
                 "{} {}",
                 messages.commit_created().green().bold(),
                 commit_oid.to_string()[..8].to_string().cyan()
-            );
+            ));
         } else {
-            println!("{} successfully", messages.commit_amended().green().bold());
+            print_success(messages.commit_amended_successfully());
         }
 
         // Push after commit if requested or ask user
-        if self.push {
-            println!("\n{}", messages.pushing_commit(&self.remote));
-            repo.push(&self.remote, "HEAD")?;
-            println!("{}", messages.pushed_commit(&self.remote));
-        } else if !self.yes && !self.dry_run {
-            let should_push = Confirm::new()
-                .with_prompt(messages.push_after_commit())
-                .default(false)
-                .interact()?;
+        if self.push || (!self.yes && !self.dry_run) {
+            let branch = repo
+                .current_branch()
+                .unwrap_or_else(|_| "HEAD (detached)".to_string());
+
+            let should_push = if self.push {
+                true
+            } else {
+                Confirm::new()
+                    .with_prompt(messages.push_after_commit(&branch))
+                    .default(false)
+                    .interact()?
+            };
 
             if should_push {
-                println!("\n{}", messages.pushing_commit(&self.remote));
+                print_progress(&messages.pushing_commit(&self.remote, &branch));
                 repo.push(&self.remote, "HEAD")?;
-                println!("{}", messages.pushed_commit(&self.remote));
+                print_success(&messages.pushed_commit(&self.remote, &branch));
             }
         }
 
@@ -306,21 +303,22 @@ impl CommitCommand {
             .await
             .context("Failed to initialize LLM. Use --manual for manual commit.")?;
 
-        println!("{}", messages.ai_analyzing());
+        let spinner = crate::utils::Spinner::start(messages.ai_analyzing());
 
         let language = manager.get_language().unwrap_or(Language::English);
 
         let generated = if self.yes {
             generator
                 .generate_commit_from_repo(repo, format, language)
-                .await?
+                .await
         } else {
             generator
-                .generate_commit_interactive(repo, format, language)
-                .await?
+                .generate_commit_interactive(repo, format, language, messages)
+                .await
         };
+        spinner.finish_clear();
 
-        Ok(generated.to_conventional())
+        Ok(generated?.to_conventional())
     }
 
     async fn create_interactive_commit(

src/commands/credential.rs

@@ -1,458 +1,457 @@
-use anyhow::{Context, Result};
-use clap::{Parser, Subcommand};
-use std::io::{self, BufRead, Write};
-use std::path::PathBuf;
-
-use crate::config::manager::ConfigManager;
-use crate::config::{TokenConfig, TokenType};
-
-/// Git credential helper command.
-///
-/// Implements the git credential helper protocol
-/// (https://git-scm.com/docs/gitcredentials). Intended to be invoked by git
-/// via `credential.helper` configuration, not by end users. Hidden from the
-/// main help output.
-#[derive(Parser)]
-#[command(hide = true)]
-pub struct CredentialCommand {
-    #[command(subcommand)]
-    command: CredentialSubcommand,
-}
-
-#[derive(Subcommand)]
-enum CredentialSubcommand {
-    /// Read attributes on stdin and output credentials on stdout.
-    #[command(hide = true)]
-    Get,
-    /// Read attributes (including password) on stdin and store them.
-    #[command(hide = true)]
-    Store,
-    /// Read attributes on stdin and erase any matching stored credentials.
-    #[command(hide = true)]
-    Erase,
-}
-
-impl CredentialCommand {
-    pub async fn execute(&self, config_path: Option<PathBuf>) -> Result<()> {
-        match &self.command {
-            CredentialSubcommand::Get => Self::get(config_path),
-            CredentialSubcommand::Store => Self::store(config_path),
-            CredentialSubcommand::Erase => Self::erase(config_path),
-        }
-    }
-
-    fn get_manager(config_path: &Option<PathBuf>) -> Result<ConfigManager> {
-        match config_path {
-            Some(path) => ConfigManager::with_path(path),
-            None => ConfigManager::new(),
-        }
-    }
-
-    /// `git credential get`: look up a PAT for the requested host and emit it
-    /// on stdout following the git credential helper protocol.
-    fn get(config_path: Option<PathBuf>) -> Result<()> {
-        let attrs = CredentialAttributes::from_stdin()?;
-
-        let host = match attrs.host.as_deref() {
-            Some(h) if !h.is_empty() => h,
-            _ => return Ok(()),
-        };
-        let service = host_to_service(host);
-
-        let manager = match Self::get_manager(&config_path) {
-            Ok(m) => m,
-            Err(_) => return Ok(()),
-        };
-
-        let (profile_name, pat) = match find_pat_for_service(&manager, &service) {
-            Some(tuple) => tuple,
-            None => return Ok(()),
-        };
-
-        // Prefer a git-supplied username; fall back to the profile's user_name.
-        let username = attrs.username.clone().or_else(|| {
-            manager
-                .get_profile(&profile_name)
-                .map(|p| p.user_name.clone())
-        });
-
-        let output = CredentialAttributes {
-            protocol: attrs.protocol.clone(),
-            host: attrs.host.clone(),
-            path: attrs.path.clone(),
-            username,
-            password: Some(pat),
-        };
-        output.to_stdout()?;
-        Ok(())
-    }
-
-    /// `git credential store`: persist the PAT provided by git using the
-    /// existing keyring-backed storage, associated with a matching profile.
-    fn store(config_path: Option<PathBuf>) -> Result<()> {
-        let attrs = CredentialAttributes::from_stdin()?;
-
-        let host = match attrs.host.as_deref() {
-            Some(h) if !h.is_empty() => h,
-            _ => return Ok(()),
-        };
-        let service = host_to_service(host);
-
-        let password = match attrs.password.as_deref() {
-            Some(p) if !p.is_empty() => p,
-            _ => return Ok(()),
-        };
-
-        let mut manager = Self::get_manager(&config_path)?;
-
-        let profile_name = match find_profile_for_store(&manager, attrs.username.as_deref()) {
-            Some(name) => name,
-            None => return Ok(()),
-        };
-
-        // Store PAT in keyring using the existing profile-bound logic.
-        if let Err(e) = manager.store_pat_for_profile(&profile_name, &service, password) {
-            eprintln!("[quicommit credential] failed to store PAT: {}", e);
-            return Ok(());
-        }
-
-        // Register the token in the profile config if not already present.
-        let already_has = manager
-            .get_profile(&profile_name)
-            .map(|p| p.tokens.contains_key(&service))
-            .unwrap_or(false);
-        if !already_has {
-            let _ = manager.add_token_to_profile(
-                &profile_name,
-                service.clone(),
-                TokenConfig::new(TokenType::Personal),
-            );
-        }
-
-        manager.save()?;
-        Ok(())
-    }
-
-    /// `git credential erase`: remove any PAT stored for the requested host
-    /// from the keyring and the associated profile config entries.
-    fn erase(config_path: Option<PathBuf>) -> Result<()> {
-        let attrs = CredentialAttributes::from_stdin()?;
-
-        let host = match attrs.host.as_deref() {
-            Some(h) if !h.is_empty() => h,
-            _ => return Ok(()),
-        };
-        let service = host_to_service(host);
-
-        let mut manager = Self::get_manager(&config_path)?;
-
-        let profile_names: Vec<String> = manager
-            .list_profiles()
-            .into_iter()
-            .filter(|name| {
-                manager
-                    .get_profile(name)
-                    .map(|p| p.tokens.contains_key(&service))
-                    .unwrap_or(false)
-            })
-            .cloned()
-            .collect();
-
-        for name in &profile_names {
-            if let Err(e) = manager.remove_token_from_profile(name, &service) {
-                eprintln!(
-                    "[quicommit credential] failed to erase PAT for '{}': {}",
-                    name, e
-                );
-            }
-        }
-
-        manager.save()?;
-        Ok(())
-    }
-}
-
-/// Credential attributes exchanged with git via stdin/stdout following the
-/// git credential helper protocol (`key=value`, one per line, blank line ends).
-#[derive(Debug, Default, Clone, PartialEq, Eq)]
-pub struct CredentialAttributes {
-    pub protocol: Option<String>,
-    pub host: Option<String>,
-    pub path: Option<String>,
-    pub username: Option<String>,
-    pub password: Option<String>,
-}
-
-impl CredentialAttributes {
-    /// Read attributes from stdin following the git credential helper protocol.
-    pub fn from_stdin() -> Result<Self> {
-        let stdin = io::stdin();
-        let mut attrs = Self::default();
-        for line in stdin.lock().lines() {
-            let line =
-                line.context("Failed to read credential attributes from stdin")?;
-            if line.is_empty() {
-                break;
-            }
-            if let Some((key, value)) = line.split_once('=') {
-                match key {
-                    "protocol" => attrs.protocol = Some(value.to_string()),
-                    "host" => attrs.host = Some(value.to_string()),
-                    "path" => attrs.path = Some(value.to_string()),
-                    "username" => attrs.username = Some(value.to_string()),
-                    "password" => attrs.password = Some(value.to_string()),
-                    _ => {} // ignore unknown keys
-                }
-            }
-        }
-        Ok(attrs)
-    }
-
-    /// Write attributes to stdout following the git credential helper protocol.
-    pub fn to_stdout(&self) -> Result<()> {
-        let stdout = io::stdout();
-        let mut handle = stdout.lock();
-        if let Some(ref v) = self.protocol {
-            writeln!(handle, "protocol={}", v)?;
-        }
-        if let Some(ref v) = self.host {
-            writeln!(handle, "host={}", v)?;
-        }
-        if let Some(ref v) = self.path {
-            writeln!(handle, "path={}", v)?;
-        }
-        if let Some(ref v) = self.username {
-            writeln!(handle, "username={}", v)?;
-        }
-        if let Some(ref v) = self.password {
-            writeln!(handle, "password={}", v)?;
-        }
-        writeln!(handle)?; // blank line terminates the attribute list
-        handle.flush()?;
-        Ok(())
-    }
-
-    /// Parse attributes from a text block following the git credential helper
-    /// protocol. Intended for testing and non-stdin input handling.
-    pub fn parse_str(input: &str) -> Self {
-        let mut attrs = Self::default();
-        for line in input.lines() {
-            if line.is_empty() {
-                break;
-            }
-            if let Some((key, value)) = line.split_once('=') {
-                match key {
-                    "protocol" => attrs.protocol = Some(value.to_string()),
-                    "host" => attrs.host = Some(value.to_string()),
-                    "path" => attrs.path = Some(value.to_string()),
-                    "username" => attrs.username = Some(value.to_string()),
-                    "password" => attrs.password = Some(value.to_string()),
-                    _ => {}
-                }
-            }
-        }
-        attrs
-    }
-
-    /// Serialize attributes to a string following the git credential helper
-    /// protocol. Intended for testing and non-stdout output handling.
-    pub fn serialize(&self) -> String {
-        let mut out = String::new();
-        if let Some(ref v) = self.protocol {
-            out.push_str(&format!("protocol={}\n", v));
-        }
-        if let Some(ref v) = self.host {
-            out.push_str(&format!("host={}\n", v));
-        }
-        if let Some(ref v) = self.path {
-            out.push_str(&format!("path={}\n", v));
-        }
-        if let Some(ref v) = self.username {
-            out.push_str(&format!("username={}\n", v));
-        }
-        if let Some(ref v) = self.password {
-            out.push_str(&format!("password={}\n", v));
-        }
-        out.push('\n');
-        out
-    }
-}
-
-/// Map a git host to a service name used by the keyring-backed PAT storage.
-///
-/// Common git hosting services are mapped to short canonical names. Unknown
-/// hosts are used as-is (lowercased, trailing slash trimmed).
-pub fn host_to_service(host: &str) -> String {
-    let host = host.to_lowercase();
-    let host = host.trim_end_matches('/');
-    match host {
-        "github.com" | "www.github.com" => "github".to_string(),
-        "gitlab.com" | "www.gitlab.com" => "gitlab".to_string(),
-        "bitbucket.org" | "www.bitbucket.org" => "bitbucket".to_string(),
-        "codeberg.org" | "www.codeberg.org" => "codeberg".to_string(),
-        "gitea.com" | "www.gitea.com" => "gitea".to_string(),
-        "gitee.com" | "www.gitee.com" => "gitee".to_string(),
-        other => other.to_string(),
-    }
-}
-
-/// Search all profiles for one that has a PAT stored for the given service.
-/// Returns the profile name and the PAT value.
-fn find_pat_for_service(manager: &ConfigManager, service: &str) -> Option<(String, String)> {
-    for profile_name in manager.list_profiles() {
-        if manager.has_pat_for_profile(profile_name, service) {
-            if let Ok(Some(pat)) = manager.get_pat_for_profile(profile_name, service) {
-                return Some((profile_name.clone(), pat));
-            }
-        }
-    }
-    None
-}
-
-/// Determine which profile to use when storing a credential.
-///
-/// Prefers a profile whose `user_name` or `user_email` matches the
-/// git-supplied username; otherwise falls back to the default profile.
-fn find_profile_for_store(
-    manager: &ConfigManager,
-    username: Option<&str>,
-) -> Option<String> {
-    if let Some(username) = username {
-        for name in manager.list_profiles() {
-            if let Some(profile) = manager.get_profile(name) {
-                if profile.user_name == username || profile.user_email == username {
-                    return Some(name.clone());
-                }
-            }
-        }
-    }
-    manager.default_profile_name().cloned()
-}
-
-/// Extract a PAT for the given host from saved credentials across all profiles.
-///
-/// This is intended for use by other parts of the application (e.g. when
-/// verifying access to a git hosting service) and searches every configured
-/// profile for a stored PAT matching the host.
-pub fn get_pat_for_host(host: &str) -> Result<Option<String>> {
-    let manager = ConfigManager::new()?;
-    let service = host_to_service(host);
-    Ok(find_pat_for_service(&manager, &service).map(|(_, pat)| pat))
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-
-    #[test]
-    fn test_host_to_service_known_hosts() {
-        assert_eq!(host_to_service("github.com"), "github");
-        assert_eq!(host_to_service("www.github.com"), "github");
-        assert_eq!(host_to_service("gitlab.com"), "gitlab");
-        assert_eq!(host_to_service("bitbucket.org"), "bitbucket");
-        assert_eq!(host_to_service("codeberg.org"), "codeberg");
-        assert_eq!(host_to_service("gitea.com"), "gitea");
-        assert_eq!(host_to_service("gitee.com"), "gitee");
-    }
-
-    #[test]
-    fn test_host_to_service_case_insensitive() {
-        assert_eq!(host_to_service("GitHub.Com"), "github");
-        assert_eq!(host_to_service("GITLAB.COM"), "gitlab");
-    }
-
-    #[test]
-    fn test_host_to_service_trailing_slash() {
-        assert_eq!(host_to_service("github.com/"), "github");
-        assert_eq!(host_to_service("gitlab.com//"), "gitlab");
-    }
-
-    #[test]
-    fn test_host_to_service_unknown_host() {
-        assert_eq!(host_to_service("example.com"), "example.com");
-        assert_eq!(host_to_service("git.internal.corp"), "git.internal.corp");
-    }
-
-    #[test]
-    fn test_credential_attributes_parse_basic() {
-        let input = "protocol=https\nhost=github.com\nusername=octocat\npassword=ghp_token123\n\n";
-        let attrs = CredentialAttributes::parse_str(input);
-        assert_eq!(attrs.protocol.as_deref(), Some("https"));
-        assert_eq!(attrs.host.as_deref(), Some("github.com"));
-        assert_eq!(attrs.username.as_deref(), Some("octocat"));
-        assert_eq!(attrs.password.as_deref(), Some("ghp_token123"));
-        assert!(attrs.path.is_none());
-    }
-
-    #[test]
-    fn test_credential_attributes_parse_with_path() {
-        let input = "protocol=https\nhost=github.com\npath=owner/repo.git\n\n";
-        let attrs = CredentialAttributes::parse_str(input);
-        assert_eq!(attrs.path.as_deref(), Some("owner/repo.git"));
-    }
-
-    #[test]
-    fn test_credential_attributes_parse_empty() {
-        let attrs = CredentialAttributes::parse_str("");
-        assert!(attrs.protocol.is_none());
-        assert!(attrs.host.is_none());
-        assert!(attrs.username.is_none());
-        assert!(attrs.password.is_none());
-    }
-
-    #[test]
-    fn test_credential_attributes_parse_ignores_unknown_keys() {
-        let input = "protocol=https\nhost=github.com\nunknown=value\nfoo=bar\n\n";
-        let attrs = CredentialAttributes::parse_str(input);
-        assert_eq!(attrs.protocol.as_deref(), Some("https"));
-        assert_eq!(attrs.host.as_deref(), Some("github.com"));
-    }
-
-    #[test]
-    fn test_credential_attributes_parse_stops_at_blank_line() {
-        let input = "protocol=https\nhost=github.com\n\npassword=should_be_ignored\n";
-        let attrs = CredentialAttributes::parse_str(input);
-        assert_eq!(attrs.protocol.as_deref(), Some("https"));
-        assert_eq!(attrs.host.as_deref(), Some("github.com"));
-        assert!(attrs.password.is_none());
-    }
-
-    #[test]
-    fn test_credential_attributes_serialize_roundtrip() {
-        let attrs = CredentialAttributes {
-            protocol: Some("https".to_string()),
-            host: Some("github.com".to_string()),
-            path: None,
-            username: Some("octocat".to_string()),
-            password: Some("ghp_token".to_string()),
-        };
-        let serialized = attrs.serialize();
-        let reparsed = CredentialAttributes::parse_str(&serialized);
-        assert_eq!(attrs, reparsed);
-    }
-
-    #[test]
-    fn test_credential_attributes_serialize_includes_blank_line() {
-        let attrs = CredentialAttributes {
-            protocol: Some("https".to_string()),
-            host: Some("github.com".to_string()),
-            path: None,
-            username: None,
-            password: None,
-        };
-        let serialized = attrs.serialize();
-        assert!(serialized.ends_with("\n\n"));
-        assert!(serialized.contains("protocol=https\n"));
-        assert!(serialized.contains("host=github.com\n"));
-    }
-
-    #[test]
-    fn test_credential_attributes_default() {
-        let attrs = CredentialAttributes::default();
-        assert!(attrs.protocol.is_none());
-        assert!(attrs.host.is_none());
-        assert!(attrs.path.is_none());
-        assert!(attrs.username.is_none());
-        assert!(attrs.password.is_none());
-    }
-}
+use anyhow::{Context, Result};
+use clap::{Parser, Subcommand};
+use std::io::{self, BufRead, Write};
+use std::path::PathBuf;
+
+use crate::config::manager::ConfigManager;
+use crate::config::{TokenConfig, TokenType};
+
+/// Git credential helper command.
+///
+/// Implements the git credential helper protocol
+/// (https://git-scm.com/docs/gitcredentials). Intended to be invoked by git
+/// via `credential.helper` configuration, not by end users. Hidden from the
+/// main help output.
+#[derive(Parser)]
+#[command(hide = true)]
+pub struct CredentialCommand {
+    #[command(subcommand)]
+    command: CredentialSubcommand,
+}
+
+#[derive(Subcommand)]
+enum CredentialSubcommand {
+    /// Read attributes on stdin and output credentials on stdout.
+    #[command(hide = true)]
+    Get,
+    /// Read attributes (including password) on stdin and store them.
+    #[command(hide = true)]
+    Store,
+    /// Read attributes on stdin and erase any matching stored credentials.
+    #[command(hide = true)]
+    Erase,
+}
+
+impl CredentialCommand {
+    pub async fn execute(&self, config_path: Option<PathBuf>) -> Result<()> {
+        match &self.command {
+            CredentialSubcommand::Get => Self::get(config_path),
+            CredentialSubcommand::Store => Self::store(config_path),
+            CredentialSubcommand::Erase => Self::erase(config_path),
+        }
+    }
+
+    fn get_manager(config_path: &Option<PathBuf>) -> Result<ConfigManager> {
+        match config_path {
+            Some(path) => ConfigManager::with_path(path),
+            None => ConfigManager::new(),
+        }
+    }
+
+    /// `git credential get`: look up a PAT for the requested host and emit it
+    /// on stdout following the git credential helper protocol.
+    fn get(config_path: Option<PathBuf>) -> Result<()> {
+        let attrs = CredentialAttributes::from_stdin()?;
+
+        let host = match attrs.host.as_deref() {
+            Some(h) if !h.is_empty() => h,
+            _ => return Ok(()),
+        };
+        let service = host_to_service(host);
+
+        let manager = match Self::get_manager(&config_path) {
+            Ok(m) => m,
+            Err(_) => return Ok(()),
+        };
+
+        let (profile_name, pat) = match find_pat_for_service(&manager, &service) {
+            Some(tuple) => tuple,
+            None => return Ok(()),
+        };
+
+        // Prefer a git-supplied username; fall back to the profile's user_name.
+        let username = attrs.username.clone().or_else(|| {
+            manager
+                .get_profile(&profile_name)
+                .map(|p| p.user_name.clone())
+        });
+
+        let output = CredentialAttributes {
+            protocol: attrs.protocol.clone(),
+            host: attrs.host.clone(),
+            path: attrs.path.clone(),
+            username,
+            password: Some(pat),
+        };
+        output.to_stdout()?;
+        Ok(())
+    }
+
+    /// `git credential store`: persist the PAT provided by git using the
+    /// existing keyring-backed storage, associated with a matching profile.
+    fn store(config_path: Option<PathBuf>) -> Result<()> {
+        let attrs = CredentialAttributes::from_stdin()?;
+
+        let host = match attrs.host.as_deref() {
+            Some(h) if !h.is_empty() => h,
+            _ => return Ok(()),
+        };
+        let service = host_to_service(host);
+
+        let password = match attrs.password.as_deref() {
+            Some(p) if !p.is_empty() => p,
+            _ => return Ok(()),
+        };
+
+        let mut manager = Self::get_manager(&config_path)?;
+
+        let profile_name = match find_profile_for_store(&manager, attrs.username.as_deref()) {
+            Some(name) => name,
+            None => return Ok(()),
+        };
+
+        // Store PAT in keyring using the existing profile-bound logic.
+        if let Err(e) = manager.store_pat_for_profile(&profile_name, &service, password) {
+            eprintln!("[quicommit credential] failed to store PAT: {}", e);
+            return Ok(());
+        }
+
+        // Register the token in the profile config if not already present.
+        let already_has = manager
+            .get_profile(&profile_name)
+            .map(|p| p.tokens.contains_key(&service))
+            .unwrap_or(false);
+        if !already_has {
+            let _ = manager.add_token_to_profile(
+                &profile_name,
+                service.clone(),
+                TokenConfig::new(TokenType::Personal),
+            );
+        }
+
+        manager.save()?;
+        Ok(())
+    }
+
+    /// `git credential erase`: remove any PAT stored for the requested host
+    /// from the keyring and the associated profile config entries.
+    fn erase(config_path: Option<PathBuf>) -> Result<()> {
+        let attrs = CredentialAttributes::from_stdin()?;
+
+        let host = match attrs.host.as_deref() {
+            Some(h) if !h.is_empty() => h,
+            _ => return Ok(()),
+        };
+        let service = host_to_service(host);
+
+        let mut manager = Self::get_manager(&config_path)?;
+
+        let profile_names: Vec<String> = manager.list_profiles().into_iter().cloned().collect();
+
+        for name in &profile_names {
+            let registered = manager
+                .get_profile(name)
+                .map(|p| p.tokens.contains_key(&service))
+                .unwrap_or(false);
+
+            if registered {
+                if let Err(e) = manager.remove_token_from_profile(name, &service) {
+                    eprintln!(
+                        "[quicommit credential] failed to erase PAT for '{}': {}",
+                        name, e
+                    );
+                }
+            } else if let Ok(true) = manager.delete_orphan_pat(name, &service) {
+                // Keyring PAT without a config token entry (issue 24).
+                eprintln!(
+                    "[quicommit credential] erased unregistered PAT for '{}' (service {}) from keyring",
+                    name, service
+                );
+            }
+        }
+
+        manager.save()?;
+        Ok(())
+    }
+}
+
+/// Credential attributes exchanged with git via stdin/stdout following the
+/// git credential helper protocol (`key=value`, one per line, blank line ends).
+#[derive(Debug, Default, Clone, PartialEq, Eq)]
+pub struct CredentialAttributes {
+    pub protocol: Option<String>,
+    pub host: Option<String>,
+    pub path: Option<String>,
+    pub username: Option<String>,
+    pub password: Option<String>,
+}
+
+impl CredentialAttributes {
+    /// Read attributes from stdin following the git credential helper protocol.
+    pub fn from_stdin() -> Result<Self> {
+        let stdin = io::stdin();
+        let mut attrs = Self::default();
+        for line in stdin.lock().lines() {
+            let line = line.context("Failed to read credential attributes from stdin")?;
+            if line.is_empty() {
+                break;
+            }
+            if let Some((key, value)) = line.split_once('=') {
+                match key {
+                    "protocol" => attrs.protocol = Some(value.to_string()),
+                    "host" => attrs.host = Some(value.to_string()),
+                    "path" => attrs.path = Some(value.to_string()),
+                    "username" => attrs.username = Some(value.to_string()),
+                    "password" => attrs.password = Some(value.to_string()),
+                    _ => {} // ignore unknown keys
+                }
+            }
+        }
+        Ok(attrs)
+    }
+
+    /// Write attributes to stdout following the git credential helper protocol.
+    pub fn to_stdout(&self) -> Result<()> {
+        let stdout = io::stdout();
+        let mut handle = stdout.lock();
+        if let Some(ref v) = self.protocol {
+            writeln!(handle, "protocol={}", v)?;
+        }
+        if let Some(ref v) = self.host {
+            writeln!(handle, "host={}", v)?;
+        }
+        if let Some(ref v) = self.path {
+            writeln!(handle, "path={}", v)?;
+        }
+        if let Some(ref v) = self.username {
+            writeln!(handle, "username={}", v)?;
+        }
+        if let Some(ref v) = self.password {
+            writeln!(handle, "password={}", v)?;
+        }
+        writeln!(handle)?; // blank line terminates the attribute list
+        handle.flush()?;
+        Ok(())
+    }
+
+    /// Parse attributes from a text block following the git credential helper
+    /// protocol. Intended for testing and non-stdin input handling.
+    pub fn parse_str(input: &str) -> Self {
+        let mut attrs = Self::default();
+        for line in input.lines() {
+            if line.is_empty() {
+                break;
+            }
+            if let Some((key, value)) = line.split_once('=') {
+                match key {
+                    "protocol" => attrs.protocol = Some(value.to_string()),
+                    "host" => attrs.host = Some(value.to_string()),
+                    "path" => attrs.path = Some(value.to_string()),
+                    "username" => attrs.username = Some(value.to_string()),
+                    "password" => attrs.password = Some(value.to_string()),
+                    _ => {}
+                }
+            }
+        }
+        attrs
+    }
+
+    /// Serialize attributes to a string following the git credential helper
+    /// protocol. Intended for testing and non-stdout output handling.
+    pub fn serialize(&self) -> String {
+        let mut out = String::new();
+        if let Some(ref v) = self.protocol {
+            out.push_str(&format!("protocol={}\n", v));
+        }
+        if let Some(ref v) = self.host {
+            out.push_str(&format!("host={}\n", v));
+        }
+        if let Some(ref v) = self.path {
+            out.push_str(&format!("path={}\n", v));
+        }
+        if let Some(ref v) = self.username {
+            out.push_str(&format!("username={}\n", v));
+        }
+        if let Some(ref v) = self.password {
+            out.push_str(&format!("password={}\n", v));
+        }
+        out.push('\n');
+        out
+    }
+}
+
+/// Map a git host to a service name used by the keyring-backed PAT storage.
+///
+/// Common git hosting services are mapped to short canonical names. Unknown
+/// hosts are used as-is (lowercased, trailing slash trimmed).
+pub fn host_to_service(host: &str) -> String {
+    let host = host.to_lowercase();
+    let host = host.trim_end_matches('/');
+    match host {
+        "github.com" | "www.github.com" => "github".to_string(),
+        "gitlab.com" | "www.gitlab.com" => "gitlab".to_string(),
+        "bitbucket.org" | "www.bitbucket.org" => "bitbucket".to_string(),
+        "codeberg.org" | "www.codeberg.org" => "codeberg".to_string(),
+        "gitea.com" | "www.gitea.com" => "gitea".to_string(),
+        "gitee.com" | "www.gitee.com" => "gitee".to_string(),
+        other => other.to_string(),
+    }
+}
+
+/// Search all profiles for one that has a PAT stored for the given service.
+/// Returns the profile name and the PAT value.
+fn find_pat_for_service(manager: &ConfigManager, service: &str) -> Option<(String, String)> {
+    for profile_name in manager.list_profiles() {
+        if manager.has_pat_for_profile(profile_name, service) {
+            if let Ok(Some(pat)) = manager.get_pat_for_profile(profile_name, service) {
+                return Some((profile_name.clone(), pat));
+            }
+        }
+    }
+    None
+}
+
+/// Determine which profile to use when storing a credential.
+///
+/// Prefers a profile whose `user_name` or `user_email` matches the
+/// git-supplied username; otherwise falls back to the default profile.
+fn find_profile_for_store(manager: &ConfigManager, username: Option<&str>) -> Option<String> {
+    if let Some(username) = username {
+        for name in manager.list_profiles() {
+            if let Some(profile) = manager.get_profile(name) {
+                if profile.user_name == username || profile.user_email == username {
+                    return Some(name.clone());
+                }
+            }
+        }
+    }
+    manager.default_profile_name().cloned()
+}
+
+/// Extract a PAT for the given host from saved credentials across all profiles.
+///
+/// This is intended for use by other parts of the application (e.g. when
+/// verifying access to a git hosting service) and searches every configured
+/// profile for a stored PAT matching the host.
+pub fn get_pat_for_host(host: &str) -> Result<Option<String>> {
+    let manager = ConfigManager::new()?;
+    let service = host_to_service(host);
+    Ok(find_pat_for_service(&manager, &service).map(|(_, pat)| pat))
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn test_host_to_service_known_hosts() {
+        assert_eq!(host_to_service("github.com"), "github");
+        assert_eq!(host_to_service("www.github.com"), "github");
+        assert_eq!(host_to_service("gitlab.com"), "gitlab");
+        assert_eq!(host_to_service("bitbucket.org"), "bitbucket");
+        assert_eq!(host_to_service("codeberg.org"), "codeberg");
+        assert_eq!(host_to_service("gitea.com"), "gitea");
+        assert_eq!(host_to_service("gitee.com"), "gitee");
+    }
+
+    #[test]
+    fn test_host_to_service_case_insensitive() {
+        assert_eq!(host_to_service("GitHub.Com"), "github");
+        assert_eq!(host_to_service("GITLAB.COM"), "gitlab");
+    }
+
+    #[test]
+    fn test_host_to_service_trailing_slash() {
+        assert_eq!(host_to_service("github.com/"), "github");
+        assert_eq!(host_to_service("gitlab.com//"), "gitlab");
+    }
+
+    #[test]
+    fn test_host_to_service_unknown_host() {
+        assert_eq!(host_to_service("example.com"), "example.com");
+        assert_eq!(host_to_service("git.internal.corp"), "git.internal.corp");
+    }
+
+    #[test]
+    fn test_credential_attributes_parse_basic() {
+        let input = "protocol=https\nhost=github.com\nusername=octocat\npassword=ghp_token123\n\n";
+        let attrs = CredentialAttributes::parse_str(input);
+        assert_eq!(attrs.protocol.as_deref(), Some("https"));
+        assert_eq!(attrs.host.as_deref(), Some("github.com"));
+        assert_eq!(attrs.username.as_deref(), Some("octocat"));
+        assert_eq!(attrs.password.as_deref(), Some("ghp_token123"));
+        assert!(attrs.path.is_none());
+    }
+
+    #[test]
+    fn test_credential_attributes_parse_with_path() {
+        let input = "protocol=https\nhost=github.com\npath=owner/repo.git\n\n";
+        let attrs = CredentialAttributes::parse_str(input);
+        assert_eq!(attrs.path.as_deref(), Some("owner/repo.git"));
+    }
+
+    #[test]
+    fn test_credential_attributes_parse_empty() {
+        let attrs = CredentialAttributes::parse_str("");
+        assert!(attrs.protocol.is_none());
+        assert!(attrs.host.is_none());
+        assert!(attrs.username.is_none());
+        assert!(attrs.password.is_none());
+    }
+
+    #[test]
+    fn test_credential_attributes_parse_ignores_unknown_keys() {
+        let input = "protocol=https\nhost=github.com\nunknown=value\nfoo=bar\n\n";
+        let attrs = CredentialAttributes::parse_str(input);
+        assert_eq!(attrs.protocol.as_deref(), Some("https"));
+        assert_eq!(attrs.host.as_deref(), Some("github.com"));
+    }
+
+    #[test]
+    fn test_credential_attributes_parse_stops_at_blank_line() {
+        let input = "protocol=https\nhost=github.com\n\npassword=should_be_ignored\n";
+        let attrs = CredentialAttributes::parse_str(input);
+        assert_eq!(attrs.protocol.as_deref(), Some("https"));
+        assert_eq!(attrs.host.as_deref(), Some("github.com"));
+        assert!(attrs.password.is_none());
+    }
+
+    #[test]
+    fn test_credential_attributes_serialize_roundtrip() {
+        let attrs = CredentialAttributes {
+            protocol: Some("https".to_string()),
+            host: Some("github.com".to_string()),
+            path: None,
+            username: Some("octocat".to_string()),
+            password: Some("ghp_token".to_string()),
+        };
+        let serialized = attrs.serialize();
+        let reparsed = CredentialAttributes::parse_str(&serialized);
+        assert_eq!(attrs, reparsed);
+    }
+
+    #[test]
+    fn test_credential_attributes_serialize_includes_blank_line() {
+        let attrs = CredentialAttributes {
+            protocol: Some("https".to_string()),
+            host: Some("github.com".to_string()),
+            path: None,
+            username: None,
+            password: None,
+        };
+        let serialized = attrs.serialize();
+        assert!(serialized.ends_with("\n\n"));
+        assert!(serialized.contains("protocol=https\n"));
+        assert!(serialized.contains("host=github.com\n"));
+    }
+
+    #[test]
+    fn test_credential_attributes_default() {
+        let attrs = CredentialAttributes::default();
+        assert!(attrs.protocol.is_none());
+        assert!(attrs.host.is_none());
+        assert!(attrs.path.is_none());
+        assert!(attrs.username.is_none());
+        assert!(attrs.password.is_none());
+    }
+}

src/commands/init.rs

@@ -10,6 +10,7 @@ use crate::config::{GitProfile, Language};
 use crate::i18n::Messages;
 use crate::utils::keyring::{get_default_model, get_supported_providers, provider_needs_api_key};
 use crate::utils::validators::validate_email;
+use crate::utils::{print_success, print_warning};
 
 /// Initialize quicommit configuration
 #[derive(Parser)]
@@ -34,19 +35,16 @@ impl InitCommand {
         if config_path.exists() && !self.reset {
             if !self.yes {
                 let overwrite = Confirm::new()
-                    .with_prompt("Configuration already exists. Overwrite?")
+                    .with_prompt(messages.config_exists_overwrite())
                     .default(false)
                     .interact()?;
 
                 if !overwrite {
-                    println!("{}", "Initialization cancelled.".yellow());
+                    print_warning(messages.init_cancelled());
                     return Ok(());
                 }
             } else {
-                println!(
-                    "{}",
-                    "Configuration already exists. Use --reset to overwrite.".yellow()
-                );
+                print_warning(messages.config_exists_use_reset());
                 return Ok(());
             }
         }
@@ -59,7 +57,7 @@ impl InitCommand {
         let mut manager = ConfigManager::with_path_fresh(&config_path)?;
 
         if self.yes {
-            self.quick_setup(&mut manager).await?;
+            self.quick_setup(&mut manager, &messages).await?;
         } else {
             self.interactive_setup(&mut manager).await?;
         }
@@ -69,27 +67,52 @@ impl InitCommand {
         let language = manager.get_language().unwrap_or(Language::English);
         let messages = Messages::new(language);
 
-        println!("{}", messages.init_success().bold().green());
+        print_success(messages.init_success());
         println!("\n{}: {}", messages.config_file(), config_path.display());
         println!("\n{}:", messages.next_steps());
-        println!("  1. Create a profile: {}", "quicommit profile add".cyan());
-        println!("  2. Configure LLM: {}", "quicommit config set-llm".cyan());
-        println!("  3. Start committing: {}", "quicommit commit".cyan());
+        println!(
+            "  1. {}: {}",
+            messages.next_steps_create_profile(),
+            "quicommit profile add".cyan()
+        );
+        println!(
+            "  2. {}: {}",
+            messages.next_steps_configure_llm(),
+            "quicommit config set-llm".cyan()
+        );
+        println!(
+            "  3. {}: {}",
+            messages.next_steps_start_committing(),
+            "quicommit commit".cyan()
+        );
 
         Ok(())
     }
 
-    async fn quick_setup(&self, manager: &mut ConfigManager) -> Result<()> {
+    async fn quick_setup(&self, manager: &mut ConfigManager, messages: &Messages) -> Result<()> {
         let git_config = git2::Config::open_default()?;
 
-        let user_name = git_config
-            .get_string("user.name")
-            .unwrap_or_else(|_| "User".to_string());
-        let user_email = git_config
-            .get_string("user.email")
-            .unwrap_or_else(|_| "user@example.com".to_string());
+        let name_result = git_config.get_string("user.name");
+        let email_result = git_config.get_string("user.email");
+        let name_missing = name_result.is_err();
+        let email_missing = email_result.is_err();
+        let user_name = name_result.unwrap_or_else(|_| "User".to_string());
+        let user_email = email_result.unwrap_or_else(|_| "user@example.com".to_string());
 
-        let profile = GitProfile::new("default".to_string(), user_name, user_email);
+        let profile = GitProfile::new("default".to_string(), user_name.clone(), user_email.clone());
+
+        println!("\n{}", messages.quick_setup_summary().bold());
+        println!("  {}: {}", messages.profile_label(), "default".cyan());
+        println!(
+            "  {}: {} <{}>",
+            messages.identity_label(),
+            user_name,
+            user_email
+        );
+        if name_missing || email_missing {
+            print_warning(messages.identity_placeholder_warning());
+        }
+        println!("  {}: {}", messages.llm_provider_label(), "ollama".cyan());
 
         manager.add_profile("default".to_string(), profile)?;
         manager.set_default_profile(Some("default".to_string()))?;
@@ -231,11 +254,8 @@ impl InitCommand {
         let keyring_available = keyring.is_available();
 
         if !keyring_available {
-            println!(
-                "\n{}",
-                "⚠ Keyring is not available on this system.".yellow()
-            );
-            println!("{}", keyring.get_status_message().yellow());
+            print_warning(messages.keyring_unavailable());
+            print_warning(&keyring.get_status_message());
         }
 
         let api_key = if provider_needs_api_key(&provider) {
@@ -246,11 +266,7 @@ impl InitCommand {
                 .ok();
 
             if let Some(_key) = env_key {
-                println!(
-                    "\n{} {}",
-                    "✓".green(),
-                    "Found API key in environment variable.".green()
-                );
+                print_success(messages.api_key_found_env());
                 None
             } else if keyring_available {
                 let prompt = match provider.as_str() {
@@ -259,16 +275,13 @@ impl InitCommand {
                     "kimi" => messages.kimi_api_key(),
                     "deepseek" => messages.deepseek_api_key(),
                     "openrouter" => messages.openrouter_api_key(),
-                    _ => "API Key",
+                    _ => messages.api_key_prompt_other(),
                 };
 
-                let key: String = Input::new().with_prompt(prompt).interact_text()?;
+                let key: String = crate::utils::password_input(prompt)?;
                 Some(key)
             } else {
-                println!(
-                    "\n{}",
-                    "Please set the QUICOMMIT_API_KEY environment variable.".yellow()
-                );
+                print_warning(messages.please_set_api_key_env());
                 None
             }
         } else {
@@ -277,24 +290,26 @@ impl InitCommand {
 
         let default_model = get_default_model(&provider);
         let model: String = Input::new()
-            .with_prompt("Model name")
+            .with_prompt(messages.model_name())
             .default(default_model.to_string())
             .interact_text()?;
 
         let base_url: Option<String> = if provider == "ollama" {
             let url: String = Input::new()
-                .with_prompt("Ollama server URL")
+                .with_prompt(messages.ollama_server_url())
                 .default("http://localhost:11434".to_string())
                 .interact_text()?;
             Some(url)
         } else {
             let use_custom_url = Confirm::new()
-                .with_prompt("Use custom API base URL?")
+                .with_prompt(messages.use_custom_base_url())
                 .default(false)
                 .interact()?;
 
             if use_custom_url {
-                let url: String = Input::new().with_prompt("Base URL").interact_text()?;
+                let url: String = Input::new()
+                    .with_prompt(messages.base_url_plain())
+                    .interact_text()?;
                 Some(url)
             } else {
                 None
@@ -309,11 +324,7 @@ impl InitCommand {
             && provider_needs_api_key(&provider)
         {
             manager.set_api_key(&key)?;
-            println!(
-                "\n{} {}",
-                "✓".green(),
-                "API key stored securely in system keyring.".green()
-            );
+            print_success(messages.api_key_stored_keyring());
         }
 
         Ok(())
@@ -332,7 +343,7 @@ impl InitCommand {
             .interact_text()?;
 
         let pub_key_path: String = Input::new()
-            .with_prompt("SSH public key path (optional, leave empty to auto-detect)")
+            .with_prompt(messages.ssh_public_key_path())
             .default(ssh_dir.join("id_rsa.pub").display().to_string())
             .allow_empty(true)
             .interact_text()?;
@@ -354,12 +365,12 @@ impl InitCommand {
         };
 
         let agent_forwarding = Confirm::new()
-            .with_prompt("Enable SSH agent forwarding (-A)?")
+            .with_prompt(messages.ssh_agent_forwarding())
             .default(false)
             .interact()?;
 
         let known_hosts: String = Input::new()
-            .with_prompt("Custom known_hosts file path (optional)")
+            .with_prompt(messages.known_hosts_path())
             .allow_empty(true)
             .interact_text()?;
         let known_hosts_file = if known_hosts.is_empty() {
@@ -369,7 +380,7 @@ impl InitCommand {
         };
 
         let custom_cmd: String = Input::new()
-            .with_prompt("Custom SSH command (optional, overrides all other SSH settings)")
+            .with_prompt(messages.custom_ssh_command())
             .allow_empty(true)
             .interact_text()?;
         let ssh_command = if custom_cmd.is_empty() {

src/commands/tag.rs

@@ -13,6 +13,7 @@ use crate::git::tag::{
 };
 use crate::git::{GitRepo, find_repo};
 use crate::i18n::Messages;
+use crate::utils::{print_progress, print_success, print_warning};
 
 /// Generate and create Git tags
 #[derive(Parser)]
@@ -61,7 +62,7 @@ pub struct TagCommand {
     #[arg(short = 't', long)]
     think: bool,
 
-    /// Skip interactive prompts
+    /// Skip interactive prompts only (generation behavior unchanged)
     #[arg(short = 'y', long)]
     yes: bool,
 
@@ -115,11 +116,11 @@ impl TagCommand {
         {
             let version_str = tag_name.trim_start_matches('v');
             if let Err(e) = crate::utils::validators::validate_semver(version_str) {
-                println!("{}: {}", "Warning".yellow(), e);
+                print_warning(&format!("{}: {}", messages.warning(), e));
 
                 if !self.yes {
                     let proceed = Confirm::new()
-                        .with_prompt("Proceed with this tag name anyway?")
+                        .with_prompt(messages.proceed_invalid_tag_name())
                         .default(true)
                         .interact()?;
 
@@ -135,7 +136,7 @@ impl TagCommand {
             None
         } else if let Some(msg) = &self.message {
             Some(msg.clone())
-        } else if self.generate || (config.tag.auto_generate && !self.yes) {
+        } else if self.generate || config.tag.auto_generate {
             Some(
                 self.generate_tag_message(&repo, &tag_name, &messages)
                     .await?,
@@ -143,7 +144,7 @@ impl TagCommand {
         } else if !self.yes {
             Some(self.input_message_interactive(&tag_name, &messages)?)
         } else {
-            Some(format!("Release {}", tag_name))
+            Some(messages.release_default(&tag_name))
         };
 
         // Show preview
@@ -165,13 +166,13 @@ impl TagCommand {
                 .interact()?;
 
             if !confirm {
-                println!("{}", messages.tag_cancelled().yellow());
+                print_warning(messages.tag_cancelled());
                 return Ok(());
             }
         }
 
         if self.dry_run {
-            println!("\n{} {}", messages.dry_run(), "- tag not created.".yellow());
+            println!("\n{}", messages.dry_run_tag_not_created().yellow());
             return Ok(());
         }
 
@@ -184,13 +185,13 @@ impl TagCommand {
 
         builder.execute(&repo)?;
 
-        println!("{} {}", messages.tag_created().green(), tag_name.cyan());
+        print_success(&format!("{} {}", messages.tag_created(), tag_name.cyan()));
 
         // Push if requested or ask user
         if self.push {
-            println!("{}", messages.pushing_tag(&self.remote));
+            print_progress(&messages.pushing_tag(&self.remote));
             repo.push(&self.remote, &format!("refs/tags/{}", tag_name))?;
-            println!("{}", messages.pushed_tag(&self.remote));
+            print_success(&messages.pushed_tag(&self.remote));
         } else if !self.yes && !self.dry_run {
             let should_push = Confirm::new()
                 .with_prompt(messages.push_after_tag())
@@ -198,9 +199,9 @@ impl TagCommand {
                 .interact()?;
 
             if should_push {
-                println!("{}", messages.pushing_tag(&self.remote));
+                print_progress(&messages.pushing_tag(&self.remote));
                 repo.push(&self.remote, &format!("refs/tags/{}", tag_name))?;
-                println!("{}", messages.pushed_tag(&self.remote));
+                print_success(&messages.pushed_tag(&self.remote));
             }
         }
 
@@ -323,15 +324,16 @@ impl TagCommand {
         };
 
         if commits.is_empty() {
-            return Ok(format!("Release {}", version));
+            return Ok(messages.release_default(version));
         }
 
-        println!("{}", messages.ai_generating_tag(commits.len()));
-
         let generator = ContentGenerator::new_with_think(&manager, self.think, None).await?;
-        generator
+        let spinner = crate::utils::Spinner::start(&messages.ai_generating_tag(commits.len()));
+        let result = generator
             .generate_tag_message(version, &commits, language)
-            .await
+            .await;
+        spinner.finish_clear();
+        result
     }
 
     async fn auto_detect_version(
@@ -418,7 +420,7 @@ impl TagCommand {
     }
 
     fn input_message_interactive(&self, version: &str, messages: &Messages) -> Result<String> {
-        let default_msg = format!("Release {}", version);
+        let default_msg = messages.release_default(version);
 
         let use_editor = Confirm::new()
             .with_prompt(messages.open_editor())

src/config/manager.rs

@@ -278,6 +278,22 @@ impl ConfigManager {
         Ok(())
     }
 
+    /// Delete a PAT from the keyring even when no config token entry exists.
+    /// Returns true if a keyring entry was deleted (issue 24).
+    pub fn delete_orphan_pat(&self, profile_name: &str, service: &str) -> Result<bool> {
+        if let Some(profile) = self.get_profile(profile_name) {
+            let user_email = &profile.user_email;
+            if self.keyring.has_pat(profile_name, user_email, service) {
+                self.keyring.delete_pat(profile_name, user_email, service)?;
+                Ok(true)
+            } else {
+                Ok(false)
+            }
+        } else {
+            Ok(false)
+        }
+    }
+
     /// Delete all PAT tokens for a profile (used when removing a profile)
     pub fn delete_all_pats_for_profile(&self, profile_name: &str) -> Result<()> {
         if let Some(profile) = self.get_profile(profile_name) {
@@ -658,6 +674,17 @@ impl ConfigManager {
         self.modified = true;
     }
 
+    /// Check if emoji decorations are enabled
+    pub fn emoji_enabled(&self) -> bool {
+        self.config.output.emoji
+    }
+
+    /// Set emoji decorations flag
+    pub fn set_emoji_enabled(&mut self, enabled: bool) {
+        self.config.output.emoji = enabled;
+        self.modified = true;
+    }
+
     /// Export configuration to TOML string
     pub fn export(&self) -> Result<String> {
         toml::to_string_pretty(&self.config).context("Failed to serialize config")

src/config/mod.rs

@@ -47,6 +47,10 @@ pub struct AppConfig {
     /// Language settings
     #[serde(default)]
     pub language: LanguageConfig,
+
+    /// Output settings
+    #[serde(default)]
+    pub output: OutputConfig,
 }
 
 impl Default for AppConfig {
@@ -61,6 +65,7 @@ impl Default for AppConfig {
             changelog: ChangelogConfig::default(),
             repo_profiles: HashMap::new(),
             language: LanguageConfig::default(),
+            output: OutputConfig::default(),
         }
     }
 }
@@ -236,6 +241,20 @@ impl Default for LanguageConfig {
     }
 }
 
+/// Output configuration
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct OutputConfig {
+    /// Show emoji/symbol decorations (✓/✗/⚠/→)
+    #[serde(default = "default_true")]
+    pub emoji: bool,
+}
+
+impl Default for OutputConfig {
+    fn default() -> Self {
+        Self { emoji: true }
+    }
+}
+
 /// Supported languages
 #[derive(Debug, Clone, Copy, PartialEq, Eq)]
 pub enum Language {

src/generator/mod.rs

@@ -1,7 +1,10 @@
 use crate::config::manager::ConfigManager;
 use crate::config::{CommitFormat, Language};
 use crate::git::{CommitInfo, GitRepo};
-use crate::llm::{GeneratedCommit, LlmClient};
+use crate::i18n::Messages;
+use crate::llm::parsing::GeneratedCommit;
+use crate::llm::rig::LlmClient;
+use crate::utils::{eprint_warning, success_prefix};
 use anyhow::{Context, Result};
 
 /// Content generator using LLM
@@ -32,11 +35,9 @@ impl ContentGenerator {
         if thinking_enabled {
             let provider = manager.llm_provider();
             if !Self::supports_thinking(provider) {
-                eprintln!(
-                    "Warning: Provider '{}' does not support thinking mode. \
-                     Disabling thinking for this invocation.",
-                    provider
-                );
+                let language = manager.get_language().unwrap_or(Language::English);
+                let messages = Messages::new(language);
+                eprint_warning(&messages.thinking_unsupported(provider));
                 thinking_enabled = false;
             }
         }
@@ -155,6 +156,7 @@ impl ContentGenerator {
         repo: &GitRepo,
         format: CommitFormat,
         language: Language,
+        messages: &Messages,
     ) -> Result<GeneratedCommit> {
         use dialoguer::Select;
 
@@ -166,33 +168,33 @@ impl ContentGenerator {
 
         // Show diff summary
         let files = repo.get_staged_files()?;
-        println!("\nStaged files ({}):", files.len());
+        println!("\n{}", messages.staged_files(files.len()));
         for file in &files {
             println!("  • {}", file);
         }
 
         // Generate initial commit
-        println!("\nGenerating commit message...");
+        println!("\n{}", messages.generating_commit_message());
         let mut generated = self
             .generate_commit_message(&diff, format, language)
             .await?;
 
         loop {
             println!("\n{}", "─".repeat(60));
-            println!("Generated commit message:");
+            println!("{}", messages.generated_commit_message());
             println!("{}", "─".repeat(60));
             println!("{}", generated.to_conventional());
             println!("{}", "─".repeat(60));
 
             let options = vec![
-                "✓ Accept and commit",
-                "🔄 Regenerate",
-                "✏️  Edit",
-                "❌ Cancel",
+                format!("{} {}", success_prefix(), messages.accept_and_commit()),
+                messages.regenerate().to_string(),
+                messages.edit().to_string(),
+                messages.cancel().to_string(),
             ];
 
             let selection = Select::new()
-                .with_prompt("What would you like to do?")
+                .with_prompt(messages.what_would_you_like_to_do())
                 .items(&options)
                 .default(0)
                 .interact()?;
@@ -200,7 +202,7 @@ impl ContentGenerator {
             match selection {
                 0 => return Ok(generated),
                 1 => {
-                    println!("Regenerating...");
+                    println!("{}", messages.regenerating());
                     generated = self
                         .generate_commit_message(&diff, format, language)
                         .await?;
@@ -209,7 +211,7 @@ impl ContentGenerator {
                     let edited = crate::utils::editor::edit_content(&generated.to_conventional())?;
                     generated = self.parse_edited_commit(&edited, format)?;
                 }
-                3 => anyhow::bail!("Cancelled by user"),
+                3 => anyhow::bail!("{}", messages.cancelled_by_user()),
                 _ => {}
             }
         }

src/git/mod.rs

@@ -76,7 +76,7 @@ fn try_open_repo_with_git2(path: &Path) -> Result<Repository> {
         .or_else(|_| Repository::discover(&normalized))
         .or_else(|_| Repository::open(&normalized));
 
-    repo.map_err(|e| anyhow::anyhow!("git2 failed: {}", e))
+    repo.map_err(|e| anyhow::anyhow!("Failed to open repository: {}", e))
 }
 
 fn try_open_repo_with_git_cli(path: &Path) -> Result<Repository> {
@@ -642,7 +642,7 @@ impl GitRepo {
     }
 
     /// Stage all changes including subdirectories, then remove ignored tracked files
-    pub fn stage_all(&self) -> Result<Vec<String>> {
+    pub fn stage_all(&self, messages: &crate::i18n::Messages) -> Result<Vec<String>> {
         // Use git command for reliable staging (handles all edge cases)
         let output = std::process::Command::new("git")
             .args(["add", "-A"])
@@ -662,7 +662,11 @@ impl GitRepo {
         match self.remove_ignored_from_index() {
             Ok(removed) => Ok(removed),
             Err(e) => {
-                eprintln!("Warning: failed to clean ignored files from index: {}", e);
+                crate::utils::eprint_warning(&format!(
+                    "{}: {}",
+                    messages.failed_clean_ignored(),
+                    e
+                ));
                 Ok(Vec::new())
             }
         }
@@ -937,16 +941,6 @@ impl GitRepo {
         Ok(())
     }
 
-    /// Create GPG signature for arbitrary content
-    fn create_gpg_signature_for_content(
-        &self,
-        _content: &str,
-        _gpg_program: &str,
-        _signing_key: &str,
-    ) -> Result<String> {
-        Ok(String::new())
-    }
-
     /// Delete a tag
     pub fn delete_tag(&self, name: &str) -> Result<()> {
         self.repo.tag_delete(name)?;
@@ -1515,7 +1509,14 @@ mod tests {
         let head = repo.repo.head().unwrap().peel_to_commit().unwrap();
 
         // Create a lightweight tag
-        repo.repo.tag("v0.1.0", head.as_object(), &repo.repo.signature().unwrap(), "", false)
+        repo.repo
+            .tag(
+                "v0.1.0",
+                head.as_object(),
+                &repo.repo.signature().unwrap(),
+                "",
+                false,
+            )
             .unwrap();
 
         let tags = repo.get_tags().unwrap();
@@ -1541,7 +1542,13 @@ mod tests {
 
         // Create lightweight tag
         repo.repo
-            .tag("v0.1.0", head.as_object(), &repo.repo.signature().unwrap(), "", false)
+            .tag(
+                "v0.1.0",
+                head.as_object(),
+                &repo.repo.signature().unwrap(),
+                "",
+                false,
+            )
             .unwrap();
 
         let tags = repo.get_tags().unwrap();

src/lib.rs

@@ -1,5 +1,3 @@
-#![allow(dead_code)]
-
 pub mod commands;
 pub mod config;
 pub mod generator;

src/llm/anthropic.rs

@@ -1,655 +0,0 @@
-use super::thinking::ThinkingStateManager;
-use super::{LlmProvider, create_http_client};
-use anyhow::{Context, Result, bail};
-use async_trait::async_trait;
-use serde::{Deserialize, Serialize};
-use std::sync::Arc;
-use std::time::Duration;
-
-/// Anthropic Claude API client
-pub struct AnthropicClient {
-    api_key: String,
-    model: String,
-    client: reqwest::Client,
-    thinking_enabled: bool,
-    thinking_budget_tokens: u32,
-    max_tokens: u32,
-    temperature: f32,
-    top_p: Option<f32>,
-    thinking_state: Option<Arc<ThinkingStateManager>>,
-}
-
-#[derive(Debug, Serialize)]
-struct MessagesRequest {
-    model: String,
-    max_tokens: u32,
-    #[serde(skip_serializing_if = "Option::is_none")]
-    temperature: Option<f32>,
-    #[serde(skip_serializing_if = "Option::is_none")]
-    top_p: Option<f32>,
-    messages: Vec<AnthropicMessage>,
-    #[serde(skip_serializing_if = "Option::is_none")]
-    system: Option<Vec<SystemContent>>,
-    #[serde(skip_serializing_if = "Option::is_none")]
-    thinking: Option<ThinkingConfig>,
-    stream: bool,
-}
-
-#[derive(Debug, Serialize, Clone)]
-struct SystemContent {
-    #[serde(rename = "type")]
-    content_type: String,
-    text: String,
-}
-
-#[derive(Debug, Serialize)]
-struct ThinkingConfig {
-    #[serde(rename = "type")]
-    thinking_type: String,
-    #[serde(skip_serializing_if = "Option::is_none")]
-    budget_tokens: Option<u32>,
-}
-
-#[derive(Debug, Serialize, Deserialize, Clone)]
-struct AnthropicMessage {
-    role: String,
-    content: AnthropicContent,
-}
-
-#[derive(Debug, Serialize, Deserialize, Clone)]
-#[serde(untagged)]
-enum AnthropicContent {
-    Text(String),
-    Blocks(Vec<ContentBlock>),
-}
-
-#[derive(Debug, Serialize, Deserialize, Clone)]
-struct ContentBlock {
-    #[serde(rename = "type")]
-    content_type: String,
-    #[serde(skip_serializing_if = "Option::is_none")]
-    text: Option<String>,
-}
-
-#[derive(Debug, Deserialize)]
-struct MessagesResponse {
-    content: Vec<ResponseContentBlock>,
-}
-
-#[derive(Debug, Deserialize)]
-struct ResponseContentBlock {
-    #[serde(rename = "type")]
-    content_type: String,
-    text: String,
-}
-
-#[derive(Debug, Deserialize)]
-struct ErrorResponse {
-    error: AnthropicError,
-}
-
-#[derive(Debug, Deserialize)]
-struct AnthropicError {
-    #[serde(rename = "type")]
-    error_type: String,
-    message: String,
-}
-
-// --- Streaming SSE event structures ---
-
-#[derive(Debug, Deserialize)]
-struct SseEvent {
-    #[serde(rename = "type")]
-    event_type: String,
-    #[serde(default)]
-    message: Option<SseMessage>,
-    #[serde(default)]
-    index: Option<u32>,
-    #[serde(default)]
-    content_block: Option<SseContentBlock>,
-    #[serde(default)]
-    delta: Option<SseDelta>,
-    #[serde(default)]
-    usage: Option<SseUsage>,
-}
-
-#[derive(Debug, Deserialize)]
-struct SseMessage {
-    #[serde(default)]
-    content: Option<Vec<SseContentBlock>>,
-}
-
-#[derive(Debug, Deserialize)]
-struct SseContentBlock {
-    #[serde(rename = "type")]
-    content_type: String,
-    #[serde(default)]
-    thinking: Option<String>,
-    #[serde(default)]
-    text: Option<String>,
-}
-
-#[derive(Debug, Deserialize)]
-struct SseDelta {
-    #[serde(rename = "type")]
-    delta_type: Option<String>,
-    #[serde(default)]
-    thinking: Option<String>,
-    #[serde(default)]
-    text: Option<String>,
-}
-
-#[derive(Debug, Deserialize)]
-struct SseUsage {
-    #[serde(default)]
-    output_tokens: Option<u32>,
-}
-
-impl AnthropicClient {
-    pub fn new(api_key: &str, model: &str) -> Result<Self> {
-        let client = create_http_client(Duration::from_secs(60))?;
-
-        Ok(Self {
-            api_key: api_key.to_string(),
-            model: model.to_string(),
-            client,
-            thinking_enabled: false,
-            thinking_budget_tokens: 1024,
-            max_tokens: 500,
-            temperature: 0.7,
-            top_p: None,
-            thinking_state: None,
-        })
-    }
-
-    pub fn with_timeout(mut self, timeout: Duration) -> Result<Self> {
-        self.client = create_http_client(timeout)?;
-        Ok(self)
-    }
-
-    pub fn with_thinking(mut self, enabled: bool) -> Self {
-        self.thinking_enabled = enabled;
-        self
-    }
-
-    pub fn with_thinking_budget_tokens(mut self, budget_tokens: u32) -> Self {
-        self.thinking_budget_tokens = budget_tokens;
-        self
-    }
-
-    pub fn with_max_tokens(mut self, max_tokens: u32) -> Self {
-        self.max_tokens = max_tokens;
-        self
-    }
-
-    pub fn with_temperature(mut self, temperature: f32) -> Self {
-        self.temperature = temperature;
-        self
-    }
-
-    pub fn with_top_p(mut self, top_p: f32) -> Self {
-        self.top_p = Some(top_p);
-        self
-    }
-
-    pub fn with_thinking_state(mut self, state: Arc<ThinkingStateManager>) -> Self {
-        self.thinking_state = Some(state);
-        self
-    }
-
-    pub async fn list_models(&self) -> Result<Vec<String>> {
-        Ok(ANTHROPIC_MODELS.iter().map(|&m| m.to_string()).collect())
-    }
-
-    pub async fn validate_key(&self) -> Result<bool> {
-        let url = "https://api.anthropic.com/v1/messages";
-
-        let request = MessagesRequest {
-            model: self.model.clone(),
-            max_tokens: 5,
-            temperature: Some(0.0),
-            top_p: None,
-            messages: vec![AnthropicMessage {
-                role: "user".to_string(),
-                content: AnthropicContent::Text("Hi".to_string()),
-            }],
-            system: None,
-            thinking: None,
-            stream: false,
-        };
-
-        let response = self
-            .client
-            .post(url)
-            .header("x-api-key", &self.api_key)
-            .header("anthropic-version", "2023-06-01")
-            .header("Content-Type", "application/json")
-            .json(&request)
-            .send()
-            .await;
-
-        match response {
-            Ok(resp) => {
-                if resp.status().is_success() {
-                    Ok(true)
-                } else {
-                    let status = resp.status();
-                    if status.as_u16() == 401 {
-                        Ok(false)
-                    } else {
-                        let text = resp.text().await.unwrap_or_default();
-                        bail!("Anthropic API error: {} - {}", status, text)
-                    }
-                }
-            }
-            Err(e) => Err(e.into()),
-        }
-    }
-}
-
-#[async_trait]
-impl LlmProvider for AnthropicClient {
-    async fn generate(&self, prompt: &str) -> Result<String> {
-        let messages = vec![AnthropicMessage {
-            role: "user".to_string(),
-            content: AnthropicContent::Text(prompt.to_string()),
-        }];
-
-        self.messages_request_with_retry(messages, None).await
-    }
-
-    async fn generate_with_system(&self, system: &str, user: &str) -> Result<String> {
-        let messages = vec![AnthropicMessage {
-            role: "user".to_string(),
-            content: AnthropicContent::Text(user.to_string()),
-        }];
-
-        let system = if system.is_empty() {
-            None
-        } else {
-            Some(vec![SystemContent {
-                content_type: "text".to_string(),
-                text: system.to_string(),
-            }])
-        };
-
-        self.messages_request_with_retry(messages, system).await
-    }
-
-    async fn is_available(&self) -> bool {
-        self.validate_key().await.unwrap_or(false)
-    }
-
-    fn name(&self) -> &str {
-        "anthropic"
-    }
-}
-
-impl AnthropicClient {
-    async fn messages_request_with_retry(
-        &self,
-        messages: Vec<AnthropicMessage>,
-        system: Option<Vec<SystemContent>>,
-    ) -> Result<String> {
-        let mut last_error = None;
-
-        for attempt in 1..=3 {
-            match self
-                .messages_request(messages.clone(), system.clone())
-                .await
-            {
-                Ok(result) => return Ok(result),
-                Err(e) => {
-                    let err_msg = e.to_string();
-                    let is_retryable = err_msg.contains("timeout")
-                        || err_msg.contains("connection")
-                        || err_msg.contains("temporary")
-                        || err_msg.contains("5")
-                            && (err_msg.contains("500")
-                                || err_msg.contains("502")
-                                || err_msg.contains("503")
-                                || err_msg.contains("504"));
-
-                    if !is_retryable || attempt == 3 {
-                        last_error = Some(e);
-                        break;
-                    }
-
-                    tokio::time::sleep(Duration::from_millis(500 * 2u64.pow(attempt - 1))).await;
-                }
-            }
-        }
-
-        Err(last_error.unwrap_or_else(|| anyhow::anyhow!("Request failed after retries")))
-    }
-
-    async fn messages_request(
-        &self,
-        messages: Vec<AnthropicMessage>,
-        system: Option<Vec<SystemContent>>,
-    ) -> Result<String> {
-        if self.thinking_enabled {
-            self.streaming_messages_request(messages, system).await
-        } else {
-            self.non_streaming_messages_request(messages, system).await
-        }
-    }
-
-    async fn non_streaming_messages_request(
-        &self,
-        messages: Vec<AnthropicMessage>,
-        system: Option<Vec<SystemContent>>,
-    ) -> Result<String> {
-        let url = "https://api.anthropic.com/v1/messages";
-
-        let temperature = if self.temperature == 0.0 {
-            None
-        } else {
-            Some(self.temperature)
-        };
-
-        let request = MessagesRequest {
-            model: self.model.clone(),
-            max_tokens: self.max_tokens,
-            temperature,
-            top_p: self.top_p,
-            messages,
-            system,
-            thinking: Some(ThinkingConfig {
-                thinking_type: "disabled".to_string(),
-                budget_tokens: None,
-            }),
-            stream: false,
-        };
-
-        let response = self
-            .client
-            .post(url)
-            .header("x-api-key", &self.api_key)
-            .header("anthropic-version", "2023-06-01")
-            .header("Content-Type", "application/json")
-            .json(&request)
-            .send()
-            .await
-            .context("Failed to send request to Anthropic")?;
-
-        let status = response.status();
-
-        if !status.is_success() {
-            let text = response.text().await.unwrap_or_default();
-
-            if let Ok(error) = serde_json::from_str::<ErrorResponse>(&text) {
-                bail!(
-                    "Anthropic API error: {} ({})",
-                    error.error.message,
-                    error.error.error_type
-                );
-            }
-
-            bail!("Anthropic API error: {} - {}", status, text);
-        }
-
-        let result: MessagesResponse = response
-            .json()
-            .await
-            .context("Failed to parse Anthropic response")?;
-
-        result
-            .content
-            .into_iter()
-            .find(|c| c.content_type == "text")
-            .map(|c| c.text.trim().to_string())
-            .filter(|s| !s.is_empty())
-            .ok_or_else(|| anyhow::anyhow!("No text response from Anthropic"))
-    }
-
-    /// Streaming request for thinking mode, filters thinking content blocks
-    async fn streaming_messages_request(
-        &self,
-        messages: Vec<AnthropicMessage>,
-        system: Option<Vec<SystemContent>>,
-    ) -> Result<String> {
-        let url = "https://api.anthropic.com/v1/messages";
-
-        let thinking = ThinkingConfig {
-            thinking_type: "enabled".to_string(),
-            budget_tokens: Some(self.thinking_budget_tokens),
-        };
-
-        // max_tokens must exceed budget_tokens
-        let max_tokens = (self.max_tokens).max(self.thinking_budget_tokens + 100);
-
-        let request = MessagesRequest {
-            model: self.model.clone(),
-            max_tokens,
-            temperature: None, // must be omitted for thinking mode
-            top_p: None,
-            messages,
-            system,
-            thinking: Some(thinking),
-            stream: true,
-        };
-
-        let response = self
-            .client
-            .post(url)
-            .header("x-api-key", &self.api_key)
-            .header("anthropic-version", "2023-06-01")
-            .header("Content-Type", "application/json")
-            .header("Accept", "text/event-stream")
-            .json(&request)
-            .send()
-            .await
-            .context("Failed to send streaming request to Anthropic")?;
-
-        let status = response.status();
-
-        if !status.is_success() {
-            let text = response.text().await.unwrap_or_default();
-
-            if let Ok(error) = serde_json::from_str::<ErrorResponse>(&text) {
-                bail!(
-                    "Anthropic API error: {} ({})",
-                    error.error.message,
-                    error.error.error_type
-                );
-            }
-
-            bail!("Anthropic API error: {} - {}", status, text);
-        }
-
-        let mut content_buffer = String::new();
-        let mut in_thinking = false;
-        let mut has_reasoning = false;
-        let mut has_content = false;
-
-        let thinking_state = self.thinking_state.as_ref();
-
-        let mut byte_stream = response.bytes_stream();
-        let mut line_buffer = String::new();
-
-        use futures_util::StreamExt;
-
-        while let Some(chunk) = byte_stream.next().await {
-            let chunk = chunk.context("Failed to read streaming response chunk")?;
-            let chunk_str =
-                String::from_utf8(chunk.to_vec()).context("Invalid UTF-8 in stream chunk")?;
-
-            line_buffer.push_str(&chunk_str);
-
-            while let Some(line_end) = line_buffer.find('\n') {
-                let line = line_buffer[..line_end].trim().to_string();
-                line_buffer = line_buffer[line_end + 1..].to_string();
-
-                if line.is_empty() {
-                    continue;
-                }
-
-                // Parse SSE event line
-                if let Some(data) = line.strip_prefix("data: ") {
-                    if let Ok(event) = serde_json::from_str::<SseEvent>(data) {
-                        match event.event_type.as_str() {
-                            "content_block_start" => {
-                                if let Some(ref block) = event.content_block {
-                                    if block.content_type == "thinking" {
-                                        in_thinking = true;
-                                        if !has_reasoning {
-                                            has_reasoning = true;
-                                            if let Some(state) = thinking_state {
-                                                state.start_thinking();
-                                            }
-                                        }
-                                    }
-                                }
-                            }
-                            "content_block_delta" => {
-                                if let Some(ref delta) = event.delta {
-                                    // Thinking delta - ignore content but track state
-                                    if delta.thinking.is_some() {
-                                        continue;
-                                    }
-
-                                    // Text delta - collect
-                                    if in_thinking && delta.text.is_some() {
-                                        // Transition from thinking to text
-                                        if let Some(state) = thinking_state {
-                                            state.end_thinking();
-                                        }
-                                        in_thinking = false;
-                                    }
-                                    if let Some(ref text) = delta.text
-                                        && !text.is_empty()
-                                    {
-                                        has_content = true;
-                                        content_buffer.push_str(text);
-                                    }
-                                }
-                            }
-                            "content_block_stop" => {
-                                if in_thinking {
-                                    if let Some(state) = thinking_state {
-                                        state.end_thinking();
-                                    }
-                                    in_thinking = false;
-                                }
-                            }
-                            _ => {}
-                        }
-                    }
-                }
-            }
-        }
-
-        // Ensure thinking state is ended
-        if let Some(state) = thinking_state {
-            state.end_thinking();
-        }
-
-        let result = content_buffer.trim().to_string();
-
-        if result.is_empty() {
-            if has_reasoning && !has_content {
-                bail!(
-                    "Anthropic returned thinking content but no final answer. \
-                     The model may have entered an incomplete thinking state. \
-                     Please try again or disable thinking mode."
-                );
-            }
-            bail!(
-                "No response from Anthropic. \
-                 If thinking mode is enabled, try disabling it or ensure the model supports it."
-            );
-        }
-
-        Ok(result)
-    }
-}
-
-/// Available Anthropic models (Claude 4 series with extended thinking)
-pub const ANTHROPIC_MODELS: &[&str] = &[
-    "claude-opus-4-7",
-    "claude-sonnet-4-6",
-    "claude-haiku-4-5",
-    // Legacy models
-    "claude-3-opus-20240229",
-    "claude-3-sonnet-20240229",
-    "claude-3-haiku-20240307",
-    "claude-2.1",
-    "claude-2.0",
-    "claude-instant-1.2",
-];
-
-pub fn is_valid_model(model: &str) -> bool {
-    ANTHROPIC_MODELS.contains(&model)
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-
-    #[test]
-    fn test_model_validation_claude4() {
-        assert!(is_valid_model("claude-opus-4-7"));
-        assert!(is_valid_model("claude-sonnet-4-6"));
-        assert!(is_valid_model("claude-haiku-4-5"));
-        assert!(is_valid_model("claude-3-sonnet-20240229"));
-        assert!(!is_valid_model("invalid-model"));
-    }
-
-    #[test]
-    fn test_thinking_config_serialization() {
-        let config = ThinkingConfig {
-            thinking_type: "enabled".to_string(),
-            budget_tokens: Some(2048),
-        };
-        let json = serde_json::to_string(&config).unwrap();
-        assert!(json.contains(r#""type":"enabled""#));
-        assert!(json.contains(r#""budget_tokens":2048"#));
-    }
-
-    #[test]
-    fn test_thinking_config_disabled_serialization() {
-        let config = ThinkingConfig {
-            thinking_type: "disabled".to_string(),
-            budget_tokens: None,
-        };
-        let json = serde_json::to_string(&config).unwrap();
-        assert_eq!(json, r#"{"type":"disabled"}"#);
-    }
-
-    #[test]
-    fn test_system_content_serialization() {
-        let content = SystemContent {
-            content_type: "text".to_string(),
-            text: "You are helpful.".to_string(),
-        };
-        let json = serde_json::to_string(&content).unwrap();
-        assert!(json.contains(r#""type":"text""#));
-    }
-
-    #[test]
-    fn test_sse_event_parsing_content_block_start() {
-        let json = r#"{"type":"content_block_start","index":0,"content_block":{"type":"thinking","thinking":""}}"#;
-        let event: SseEvent = serde_json::from_str(json).unwrap();
-        assert_eq!(event.event_type, "content_block_start");
-        assert_eq!(event.content_block.unwrap().content_type, "thinking");
-    }
-
-    #[test]
-    fn test_sse_event_parsing_text_delta() {
-        let json = r#"{"type":"content_block_delta","index":0,"delta":{"type":"text_delta","text":"Hello"}}"#;
-        let event: SseEvent = serde_json::from_str(json).unwrap();
-        assert_eq!(event.event_type, "content_block_delta");
-        assert_eq!(event.delta.unwrap().text, Some("Hello".to_string()));
-    }
-
-    #[test]
-    fn test_anthropic_content_text() {
-        let msg = AnthropicMessage {
-            role: "user".to_string(),
-            content: AnthropicContent::Text("Hello".to_string()),
-        };
-        let json = serde_json::to_string(&msg).unwrap();
-        assert!(json.contains(r#""content":"Hello""#));
-    }
-}

src/llm/deepseek.rs

@@ -1,622 +0,0 @@
-use super::thinking::ThinkingStateManager;
-use super::{LlmProvider, create_http_client};
-use anyhow::{Context, Result, bail};
-use async_trait::async_trait;
-use serde::{Deserialize, Serialize};
-use std::sync::Arc;
-use std::time::Duration;
-
-/// DeepSeek API client
-pub struct DeepSeekClient {
-    base_url: String,
-    api_key: String,
-    model: String,
-    client: reqwest::Client,
-    thinking_enabled: bool,
-    reasoning_effort: Option<String>,
-    max_tokens: u32,
-    temperature: f32,
-    thinking_state: Option<Arc<ThinkingStateManager>>,
-}
-
-#[derive(Debug, Serialize)]
-struct ChatCompletionRequest {
-    model: String,
-    messages: Vec<Message>,
-    #[serde(skip_serializing_if = "Option::is_none")]
-    max_tokens: Option<u32>,
-    #[serde(skip_serializing_if = "Option::is_none")]
-    temperature: Option<f32>,
-    #[serde(skip_serializing_if = "Option::is_none")]
-    top_p: Option<f32>,
-    #[serde(skip_serializing_if = "Option::is_none")]
-    presence_penalty: Option<f32>,
-    #[serde(skip_serializing_if = "Option::is_none")]
-    frequency_penalty: Option<f32>,
-    stream: bool,
-    #[serde(skip_serializing_if = "Option::is_none")]
-    thinking: Option<ThinkingConfig>,
-    #[serde(skip_serializing_if = "Option::is_none")]
-    reasoning_effort: Option<String>,
-}
-
-#[derive(Debug, Serialize)]
-struct ThinkingConfig {
-    #[serde(rename = "type")]
-    thinking_type: String,
-}
-
-#[derive(Debug, Clone, Serialize, Deserialize)]
-struct Message {
-    role: String,
-    content: String,
-    #[serde(skip_serializing_if = "Option::is_none")]
-    reasoning_content: Option<String>,
-}
-
-#[derive(Debug, Deserialize)]
-struct ChatCompletionResponse {
-    choices: Vec<Choice>,
-}
-
-#[derive(Debug, Deserialize)]
-struct Choice {
-    message: Message,
-    #[serde(default)]
-    reasoning_content: Option<String>,
-}
-
-// --- Streaming response structures ---
-
-#[derive(Debug, Deserialize)]
-struct StreamChunk {
-    choices: Vec<StreamChoice>,
-}
-
-#[derive(Debug, Deserialize)]
-struct StreamChoice {
-    delta: StreamDelta,
-    #[serde(default)]
-    finish_reason: Option<String>,
-    index: Option<u32>,
-}
-
-#[derive(Debug, Deserialize, Default)]
-struct StreamDelta {
-    #[serde(default)]
-    content: Option<String>,
-    #[serde(default)]
-    reasoning_content: Option<String>,
-}
-
-#[derive(Debug, Deserialize)]
-struct ErrorResponse {
-    error: ApiError,
-}
-
-#[derive(Debug, Deserialize)]
-struct ApiError {
-    message: String,
-    #[serde(rename = "type")]
-    error_type: String,
-}
-
-impl DeepSeekClient {
-    pub fn new(api_key: &str, model: &str) -> Result<Self> {
-        let client = create_http_client(Duration::from_secs(300))?;
-
-        Ok(Self {
-            base_url: "https://api.deepseek.com".to_string(),
-            api_key: api_key.to_string(),
-            model: model.to_string(),
-            client,
-            thinking_enabled: false,
-            reasoning_effort: None,
-            max_tokens: 500,
-            temperature: 0.7,
-            thinking_state: None,
-        })
-    }
-
-    pub fn with_base_url(api_key: &str, model: &str, base_url: &str) -> Result<Self> {
-        let client = create_http_client(Duration::from_secs(300))?;
-
-        Ok(Self {
-            base_url: base_url.trim_end_matches('/').to_string(),
-            api_key: api_key.to_string(),
-            model: model.to_string(),
-            client,
-            thinking_enabled: false,
-            reasoning_effort: None,
-            max_tokens: 500,
-            temperature: 0.7,
-            thinking_state: None,
-        })
-    }
-
-    pub fn with_timeout(mut self, timeout: Duration) -> Result<Self> {
-        self.client = create_http_client(timeout)?;
-        Ok(self)
-    }
-
-    pub fn with_thinking(mut self, enabled: bool) -> Self {
-        self.thinking_enabled = enabled;
-        self
-    }
-
-    pub fn with_reasoning_effort(mut self, effort: Option<String>) -> Self {
-        self.reasoning_effort = effort;
-        self
-    }
-
-    pub fn with_max_tokens(mut self, max_tokens: u32) -> Self {
-        self.max_tokens = max_tokens;
-        self
-    }
-
-    pub fn with_temperature(mut self, temperature: f32) -> Self {
-        self.temperature = temperature;
-        self
-    }
-
-    pub fn with_thinking_state(mut self, state: Arc<ThinkingStateManager>) -> Self {
-        self.thinking_state = Some(state);
-        self
-    }
-
-    pub async fn list_models(&self) -> Result<Vec<String>> {
-        let url = format!("{}/models", self.base_url);
-
-        let response = self
-            .client
-            .get(&url)
-            .header("Authorization", format!("Bearer {}", self.api_key))
-            .send()
-            .await
-            .context("Failed to list DeepSeek models")?;
-
-        if !response.status().is_success() {
-            let status = response.status();
-            let text = response.text().await.unwrap_or_default();
-            bail!("DeepSeek API error: {} - {}", status, text);
-        }
-
-        #[derive(Deserialize)]
-        struct ModelsResponse {
-            data: Vec<ModelId>,
-        }
-
-        #[derive(Deserialize)]
-        struct ModelId {
-            id: String,
-        }
-
-        let result: ModelsResponse = response
-            .json()
-            .await
-            .context("Failed to parse DeepSeek response")?;
-
-        Ok(result.data.into_iter().map(|m| m.id).collect())
-    }
-
-    pub async fn validate_key(&self) -> Result<bool> {
-        match self.list_models().await {
-            Ok(_) => Ok(true),
-            Err(e) => {
-                let err_str = e.to_string();
-                if err_str.contains("401") || err_str.contains("Unauthorized") {
-                    Ok(false)
-                } else {
-                    Err(e)
-                }
-            }
-        }
-    }
-}
-
-#[async_trait]
-impl LlmProvider for DeepSeekClient {
-    async fn generate(&self, prompt: &str) -> Result<String> {
-        let messages = vec![Message {
-            role: "user".to_string(),
-            content: prompt.to_string(),
-            reasoning_content: None,
-        }];
-
-        self.chat_completion_with_retry(messages).await
-    }
-
-    async fn generate_with_system(&self, system: &str, user: &str) -> Result<String> {
-        let mut messages = vec![];
-
-        if !system.is_empty() {
-            messages.push(Message {
-                role: "system".to_string(),
-                content: system.to_string(),
-                reasoning_content: None,
-            });
-        }
-
-        messages.push(Message {
-            role: "user".to_string(),
-            content: user.to_string(),
-            reasoning_content: None,
-        });
-
-        self.chat_completion_with_retry(messages).await
-    }
-
-    async fn is_available(&self) -> bool {
-        self.validate_key().await.unwrap_or(false)
-    }
-
-    fn name(&self) -> &str {
-        "deepseek"
-    }
-}
-
-impl DeepSeekClient {
-    async fn chat_completion_with_retry(&self, messages: Vec<Message>) -> Result<String> {
-        let mut last_error = None;
-
-        for attempt in 1..=3 {
-            match self.chat_completion(messages.clone()).await {
-                Ok(result) => return Ok(result),
-                Err(e) => {
-                    let err_msg = e.to_string();
-                    // 网络临时错误才重试
-                    let is_retryable = err_msg.contains("timeout")
-                        || err_msg.contains("connection")
-                        || err_msg.contains("temporary")
-                        || err_msg.contains("5")
-                            && (err_msg.contains("500")
-                                || err_msg.contains("502")
-                                || err_msg.contains("503")
-                                || err_msg.contains("504"));
-
-                    if !is_retryable || attempt == 3 {
-                        last_error = Some(e);
-                        break;
-                    }
-
-                    // 指数退避
-                    tokio::time::sleep(Duration::from_millis(500 * 2u64.pow(attempt - 1))).await;
-                }
-            }
-        }
-
-        Err(last_error.unwrap_or_else(|| anyhow::anyhow!("Request failed after retries")))
-    }
-
-    async fn chat_completion(&self, messages: Vec<Message>) -> Result<String> {
-        let url = format!("{}/chat/completions", self.base_url);
-
-        let thinking = Some(ThinkingConfig {
-            thinking_type: if self.thinking_enabled {
-                "enabled".to_string()
-            } else {
-                "disabled".to_string()
-            },
-        });
-
-        // 思考模式下，temperature/top_p 等参数不应传递
-        // 非思考模式下可以正常传递
-        let (temperature, max_tokens, top_p, presence_penalty, frequency_penalty) =
-            if self.thinking_enabled {
-                (None, Some(self.max_tokens), None, None, None)
-            } else {
-                (
-                    Some(self.temperature),
-                    Some(self.max_tokens),
-                    None,
-                    None,
-                    None,
-                )
-            };
-
-        let reasoning_effort = if self.thinking_enabled {
-            self.reasoning_effort.clone()
-        } else {
-            None
-        };
-
-        let request = ChatCompletionRequest {
-            model: self.model.clone(),
-            messages: messages.clone(),
-            max_tokens,
-            temperature,
-            top_p,
-            presence_penalty,
-            frequency_penalty,
-            stream: self.thinking_enabled,
-            thinking,
-            reasoning_effort,
-        };
-
-        if self.thinking_enabled {
-            self.streaming_chat_completion(&url, &request).await
-        } else {
-            self.non_streaming_chat_completion(&url, &request).await
-        }
-    }
-
-    /// 非流式请求（非思考模式）
-    async fn non_streaming_chat_completion(
-        &self,
-        url: &str,
-        request: &ChatCompletionRequest,
-    ) -> Result<String> {
-        let response = self
-            .client
-            .post(url)
-            .header("Authorization", format!("Bearer {}", self.api_key))
-            .header("Content-Type", "application/json")
-            .json(request)
-            .send()
-            .await
-            .context("Failed to send request to DeepSeek")?;
-
-        let status = response.status();
-
-        if !status.is_success() {
-            let text = response.text().await.unwrap_or_default();
-
-            if let Ok(error) = serde_json::from_str::<ErrorResponse>(&text) {
-                bail!(
-                    "DeepSeek API error: {} ({})",
-                    error.error.message,
-                    error.error.error_type
-                );
-            }
-
-            bail!("DeepSeek API error: {} - {}", status, text);
-        }
-
-        let result: ChatCompletionResponse = response
-            .json()
-            .await
-            .context("Failed to parse DeepSeek response")?;
-
-        result
-            .choices
-            .into_iter()
-            .next()
-            .map(|c| c.message.content.trim().to_string())
-            .filter(|s| !s.is_empty())
-            .ok_or_else(|| anyhow::anyhow!("No response from DeepSeek"))
-    }
-
-    /// 流式请求（思考模式），处理 reasoning_content 和 content
-    async fn streaming_chat_completion(
-        &self,
-        url: &str,
-        request: &ChatCompletionRequest,
-    ) -> Result<String> {
-        let response = self
-            .client
-            .post(url)
-            .header("Authorization", format!("Bearer {}", self.api_key))
-            .header("Content-Type", "application/json")
-            .header("Accept", "text/event-stream")
-            .json(request)
-            .send()
-            .await
-            .context("Failed to send streaming request to DeepSeek")?;
-
-        let status = response.status();
-
-        if !status.is_success() {
-            let text = response.text().await.unwrap_or_default();
-
-            if let Ok(error) = serde_json::from_str::<ErrorResponse>(&text) {
-                bail!(
-                    "DeepSeek API error: {} ({})",
-                    error.error.message,
-                    error.error.error_type
-                );
-            }
-
-            bail!("DeepSeek API error: {} - {}", status, text);
-        }
-
-        let mut content_buffer = String::new();
-        let mut has_reasoning = false;
-        let mut has_content = false;
-        let mut stream_ended = false;
-
-        let thinking_state = self.thinking_state.as_ref();
-
-        let mut byte_stream = response.bytes_stream();
-        let mut line_buffer = String::new();
-
-        use futures_util::StreamExt;
-
-        while let Some(chunk) = byte_stream.next().await {
-            let chunk = chunk.context("Failed to read streaming response chunk")?;
-            let chunk_str =
-                String::from_utf8(chunk.to_vec()).context("Invalid UTF-8 in stream chunk")?;
-
-            line_buffer.push_str(&chunk_str);
-
-            // 处理完整行
-            while let Some(line_end) = line_buffer.find('\n') {
-                let line = line_buffer[..line_end].trim().to_string();
-                line_buffer = line_buffer[line_end + 1..].to_string();
-
-                if line.is_empty() {
-                    continue;
-                }
-
-                // SSE 格式：data: {...} 或 data: [DONE]
-                if line == "data: [DONE]" {
-                    stream_ended = true;
-                    break;
-                }
-
-                if let Some(json_str) = line.strip_prefix("data: ") {
-                    match serde_json::from_str::<StreamChunk>(json_str) {
-                        Ok(chunk) => {
-                            for choice in &chunk.choices {
-                                // 处理 reasoning_content
-                                if let Some(ref reasoning) = choice.delta.reasoning_content
-                                    && !reasoning.is_empty()
-                                {
-                                    if !has_reasoning {
-                                        has_reasoning = true;
-                                        if let Some(state) = thinking_state {
-                                            state.start_thinking();
-                                        }
-                                    }
-                                    // reasoning_content 不对外输出，仅用于内部状态判断
-                                    continue;
-                                }
-
-                                // 处理 content
-                                if let Some(ref content) = choice.delta.content
-                                    && !content.is_empty()
-                                {
-                                    // reasoning 结束，content 开始出现时移除 thinking 标识
-                                    if has_reasoning
-                                        && !has_content
-                                        && let Some(state) = thinking_state
-                                    {
-                                        state.end_thinking();
-                                    }
-                                    has_content = true;
-                                    content_buffer.push_str(content);
-                                }
-
-                                // 检查 finish_reason
-                                if let Some(ref reason) = choice.finish_reason
-                                    && reason == "stop"
-                                {
-                                    stream_ended = true;
-                                }
-                            }
-                        }
-                        Err(_) => {
-                            // 忽略无法解析的行（可能是心跳或注释）
-                        }
-                    }
-                }
-            }
-
-            if stream_ended {
-                break;
-            }
-        }
-
-        // 确保思考状态已结束
-        if let Some(state) = thinking_state {
-            state.end_thinking();
-        }
-
-        let result = content_buffer.trim().to_string();
-
-        if result.is_empty() {
-            if has_reasoning && !has_content {
-                bail!(
-                    "DeepSeek returned reasoning content but no final answer. \
-                     The model may have entered an incomplete thinking state. \
-                     Please try again or disable thinking mode."
-                );
-            }
-            bail!(
-                "No response from DeepSeek. \
-                 If thinking mode is enabled, try disabling it or ensure the model supports it."
-            );
-        }
-
-        Ok(result)
-    }
-}
-
-/// 可用 DeepSeek 模型列表
-/// deepseek-chat / deepseek-reasoner 将于 2026-07-24 停用，推荐使用 V4 系列
-pub const DEEPSEEK_MODELS: &[&str] = &[
-    "deepseek-v4-flash",
-    "deepseek-v4-pro",
-    // 兼容旧版模型 ID（将于 2026-07-24 停用）
-    "deepseek-chat",
-    "deepseek-reasoner",
-];
-
-pub fn is_valid_model(model: &str) -> bool {
-    DEEPSEEK_MODELS.contains(&model)
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-
-    #[test]
-    fn test_model_validation_v4() {
-        assert!(is_valid_model("deepseek-v4-flash"));
-        assert!(is_valid_model("deepseek-v4-pro"));
-        assert!(is_valid_model("deepseek-chat"));
-        assert!(is_valid_model("deepseek-reasoner"));
-        assert!(!is_valid_model("invalid-model"));
-        assert!(!is_valid_model("deepseek-v3"));
-    }
-
-    #[test]
-    fn test_client_builder_defaults() {
-        let client = DeepSeekClient::new("test-key", "deepseek-v4-flash").unwrap();
-        assert!(!client.thinking_enabled);
-        assert_eq!(client.max_tokens, 500);
-        assert_eq!(client.temperature, 0.7);
-        assert!(client.reasoning_effort.is_none());
-        assert!(client.thinking_state.is_none());
-    }
-
-    #[test]
-    fn test_client_builder_with_thinking() {
-        let client = DeepSeekClient::new("test-key", "deepseek-v4-flash")
-            .unwrap()
-            .with_thinking(true)
-            .with_reasoning_effort(Some("high".to_string()))
-            .with_max_tokens(1000)
-            .with_temperature(0.5);
-
-        assert!(client.thinking_enabled);
-        assert_eq!(client.reasoning_effort, Some("high".to_string()));
-        assert_eq!(client.max_tokens, 1000);
-        assert_eq!(client.temperature, 0.5);
-    }
-
-    #[test]
-    fn test_thinking_config_serialization() {
-        let config = ThinkingConfig {
-            thinking_type: "enabled".to_string(),
-        };
-        let json = serde_json::to_string(&config).unwrap();
-        assert_eq!(json, r#"{"type":"enabled"}"#);
-    }
-
-    #[test]
-    fn test_message_serialization_without_reasoning() {
-        let msg = Message {
-            role: "user".to_string(),
-            content: "Hello".to_string(),
-            reasoning_content: None,
-        };
-        let json = serde_json::to_string(&msg).unwrap();
-        assert!(!json.contains("reasoning_content"));
-    }
-
-    #[test]
-    fn test_stream_delta_parsing() {
-        let json = r#"{"content":"Hello","reasoning_content":null}"#;
-        let delta: StreamDelta = serde_json::from_str(json).unwrap();
-        assert_eq!(delta.content, Some("Hello".to_string()));
-        assert!(delta.reasoning_content.is_none());
-    }
-
-    #[test]
-    fn test_stream_delta_reasoning_only() {
-        let json = r#"{"content":null,"reasoning_content":"Let me think..."}"#;
-        let delta: StreamDelta = serde_json::from_str(json).unwrap();
-        assert!(delta.content.is_none());
-        assert_eq!(delta.reasoning_content, Some("Let me think...".to_string()));
-    }
-}

src/llm/kimi.rs

@@ -1,587 +0,0 @@
-use super::thinking::ThinkingStateManager;
-use super::{LlmProvider, create_http_client};
-use anyhow::{Context, Result, bail};
-use async_trait::async_trait;
-use serde::{Deserialize, Serialize};
-use std::sync::Arc;
-use std::time::Duration;
-
-/// Kimi API client (Moonshot AI)
-pub struct KimiClient {
-    base_url: String,
-    api_key: String,
-    model: String,
-    client: reqwest::Client,
-    thinking_enabled: bool,
-    max_tokens: u32,
-    temperature: f32,
-    thinking_state: Option<Arc<ThinkingStateManager>>,
-}
-
-#[derive(Debug, Serialize)]
-struct ChatCompletionRequest {
-    model: String,
-    messages: Vec<Message>,
-    #[serde(skip_serializing_if = "Option::is_none")]
-    max_tokens: Option<u32>,
-    #[serde(skip_serializing_if = "Option::is_none")]
-    temperature: Option<f32>,
-    stream: bool,
-    #[serde(skip_serializing_if = "Option::is_none")]
-    thinking: Option<ThinkingConfig>,
-}
-
-#[derive(Debug, Serialize)]
-struct ThinkingConfig {
-    #[serde(rename = "type")]
-    thinking_type: String,
-}
-
-#[derive(Debug, Clone, Serialize, Deserialize)]
-struct Message {
-    role: String,
-    content: String,
-    #[serde(skip_serializing_if = "Option::is_none")]
-    reasoning_content: Option<String>,
-}
-
-#[derive(Debug, Deserialize)]
-struct ChatCompletionResponse {
-    choices: Vec<Choice>,
-}
-
-#[derive(Debug, Deserialize)]
-struct Choice {
-    message: Message,
-    #[serde(default)]
-    reasoning_content: Option<String>,
-}
-
-// --- Streaming response structures ---
-
-#[derive(Debug, Deserialize)]
-struct StreamChunk {
-    choices: Vec<StreamChoice>,
-}
-
-#[derive(Debug, Deserialize)]
-struct StreamChoice {
-    delta: StreamDelta,
-    #[serde(default)]
-    finish_reason: Option<String>,
-    index: Option<u32>,
-}
-
-#[derive(Debug, Deserialize, Default)]
-struct StreamDelta {
-    #[serde(default)]
-    content: Option<String>,
-    #[serde(default)]
-    reasoning_content: Option<String>,
-}
-
-#[derive(Debug, Deserialize)]
-struct ErrorResponse {
-    error: ApiError,
-}
-
-#[derive(Debug, Deserialize)]
-struct ApiError {
-    message: String,
-    #[serde(rename = "type")]
-    error_type: String,
-}
-
-impl KimiClient {
-    pub fn new(api_key: &str, model: &str) -> Result<Self> {
-        let client = create_http_client(Duration::from_secs(300))?;
-
-        Ok(Self {
-            base_url: "https://api.moonshot.cn/v1".to_string(),
-            api_key: api_key.to_string(),
-            model: model.to_string(),
-            client,
-            thinking_enabled: false,
-            max_tokens: 500,
-            temperature: 1.0,
-            thinking_state: None,
-        })
-    }
-
-    pub fn with_base_url(api_key: &str, model: &str, base_url: &str) -> Result<Self> {
-        let client = create_http_client(Duration::from_secs(300))?;
-
-        Ok(Self {
-            base_url: base_url.trim_end_matches('/').to_string(),
-            api_key: api_key.to_string(),
-            model: model.to_string(),
-            client,
-            thinking_enabled: false,
-            max_tokens: 500,
-            temperature: 1.0,
-            thinking_state: None,
-        })
-    }
-
-    pub fn with_timeout(mut self, timeout: Duration) -> Result<Self> {
-        self.client = create_http_client(timeout)?;
-        Ok(self)
-    }
-
-    pub fn with_thinking(mut self, enabled: bool) -> Self {
-        self.thinking_enabled = enabled;
-        self
-    }
-
-    pub fn with_max_tokens(mut self, max_tokens: u32) -> Self {
-        self.max_tokens = max_tokens;
-        self
-    }
-
-    pub fn with_temperature(mut self, temperature: f32) -> Self {
-        self.temperature = temperature;
-        self
-    }
-
-    pub fn with_thinking_state(mut self, state: Arc<ThinkingStateManager>) -> Self {
-        self.thinking_state = Some(state);
-        self
-    }
-
-    pub async fn list_models(&self) -> Result<Vec<String>> {
-        let url = format!("{}/models", self.base_url);
-
-        let response = self
-            .client
-            .get(&url)
-            .header("Authorization", format!("Bearer {}", self.api_key))
-            .send()
-            .await
-            .context("Failed to list Kimi models")?;
-
-        if !response.status().is_success() {
-            let status = response.status();
-            let text = response.text().await.unwrap_or_default();
-            bail!("Kimi API error: {} - {}", status, text);
-        }
-
-        #[derive(Deserialize)]
-        struct ModelsResponse {
-            data: Vec<ModelId>,
-        }
-
-        #[derive(Deserialize)]
-        struct ModelId {
-            id: String,
-        }
-
-        let result: ModelsResponse = response
-            .json()
-            .await
-            .context("Failed to parse Kimi response")?;
-
-        Ok(result.data.into_iter().map(|m| m.id).collect())
-    }
-
-    pub async fn validate_key(&self) -> Result<bool> {
-        match self.list_models().await {
-            Ok(_) => Ok(true),
-            Err(e) => {
-                let err_str = e.to_string();
-                if err_str.contains("401") || err_str.contains("Unauthorized") {
-                    Ok(false)
-                } else {
-                    Err(e)
-                }
-            }
-        }
-    }
-}
-
-#[async_trait]
-impl LlmProvider for KimiClient {
-    async fn generate(&self, prompt: &str) -> Result<String> {
-        let messages = vec![Message {
-            role: "user".to_string(),
-            content: prompt.to_string(),
-            reasoning_content: None,
-        }];
-
-        self.chat_completion_with_retry(messages).await
-    }
-
-    async fn generate_with_system(&self, system: &str, user: &str) -> Result<String> {
-        let mut messages = vec![];
-
-        if !system.is_empty() {
-            messages.push(Message {
-                role: "system".to_string(),
-                content: system.to_string(),
-                reasoning_content: None,
-            });
-        }
-
-        messages.push(Message {
-            role: "user".to_string(),
-            content: user.to_string(),
-            reasoning_content: None,
-        });
-
-        self.chat_completion_with_retry(messages).await
-    }
-
-    async fn is_available(&self) -> bool {
-        self.validate_key().await.unwrap_or(false)
-    }
-
-    fn name(&self) -> &str {
-        "kimi"
-    }
-}
-
-impl KimiClient {
-    async fn chat_completion_with_retry(&self, messages: Vec<Message>) -> Result<String> {
-        let mut last_error = None;
-
-        for attempt in 1..=3 {
-            match self.chat_completion(messages.clone()).await {
-                Ok(result) => return Ok(result),
-                Err(e) => {
-                    let err_msg = e.to_string();
-                    let is_retryable = err_msg.contains("timeout")
-                        || err_msg.contains("connection")
-                        || err_msg.contains("temporary")
-                        || err_msg.contains("5")
-                            && (err_msg.contains("500")
-                                || err_msg.contains("502")
-                                || err_msg.contains("503")
-                                || err_msg.contains("504"));
-
-                    if !is_retryable || attempt == 3 {
-                        last_error = Some(e);
-                        break;
-                    }
-
-                    tokio::time::sleep(Duration::from_millis(500 * 2u64.pow(attempt - 1))).await;
-                }
-            }
-        }
-
-        Err(last_error.unwrap_or_else(|| anyhow::anyhow!("Request failed after retries")))
-    }
-
-    async fn chat_completion(&self, messages: Vec<Message>) -> Result<String> {
-        let url = format!("{}/chat/completions", self.base_url);
-
-        let thinking = Some(ThinkingConfig {
-            thinking_type: if self.thinking_enabled {
-                "enabled".to_string()
-            } else {
-                "disabled".to_string()
-            },
-        });
-
-        // Kimi API temperature 要求:
-        // - 思考模式: temperature 必须为 1.0
-        // - 非思考模式: temperature 必须为 0.6
-        let temperature = if self.thinking_enabled {
-            Some(1.0)
-        } else {
-            Some(0.6)
-        };
-
-        let request = ChatCompletionRequest {
-            model: self.model.clone(),
-            messages: messages.clone(),
-            max_tokens: Some(self.max_tokens),
-            temperature,
-            stream: self.thinking_enabled,
-            thinking,
-        };
-
-        if self.thinking_enabled {
-            self.streaming_chat_completion(&url, &request).await
-        } else {
-            self.non_streaming_chat_completion(&url, &request).await
-        }
-    }
-
-    /// 非流式请求（非思考模式）
-    async fn non_streaming_chat_completion(
-        &self,
-        url: &str,
-        request: &ChatCompletionRequest,
-    ) -> Result<String> {
-        let response = self
-            .client
-            .post(url)
-            .header("Authorization", format!("Bearer {}", self.api_key))
-            .header("Content-Type", "application/json")
-            .json(request)
-            .send()
-            .await
-            .context("Failed to send request to Kimi")?;
-
-        let status = response.status();
-
-        if !status.is_success() {
-            let text = response.text().await.unwrap_or_default();
-
-            if let Ok(error) = serde_json::from_str::<ErrorResponse>(&text) {
-                bail!(
-                    "Kimi API error: {} ({})",
-                    error.error.message,
-                    error.error.error_type
-                );
-            }
-
-            bail!("Kimi API error: {} - {}", status, text);
-        }
-
-        let result: ChatCompletionResponse = response
-            .json()
-            .await
-            .context("Failed to parse Kimi response")?;
-
-        result
-            .choices
-            .into_iter()
-            .next()
-            .map(|c| {
-                let content = c.message.content.trim().to_string();
-                if content.is_empty() {
-                    c.reasoning_content
-                        .or(c.message.reasoning_content)
-                        .map(|r| r.trim().to_string())
-                        .unwrap_or_default()
-                } else {
-                    content
-                }
-            })
-            .filter(|s| !s.is_empty())
-            .ok_or_else(|| anyhow::anyhow!("No response from Kimi"))
-    }
-
-    /// 流式请求（思考模式），处理 reasoning_content 和 content
-    async fn streaming_chat_completion(
-        &self,
-        url: &str,
-        request: &ChatCompletionRequest,
-    ) -> Result<String> {
-        let response = self
-            .client
-            .post(url)
-            .header("Authorization", format!("Bearer {}", self.api_key))
-            .header("Content-Type", "application/json")
-            .header("Accept", "text/event-stream")
-            .json(request)
-            .send()
-            .await
-            .context("Failed to send streaming request to Kimi")?;
-
-        let status = response.status();
-
-        if !status.is_success() {
-            let text = response.text().await.unwrap_or_default();
-
-            if let Ok(error) = serde_json::from_str::<ErrorResponse>(&text) {
-                bail!(
-                    "Kimi API error: {} ({})",
-                    error.error.message,
-                    error.error.error_type
-                );
-            }
-
-            bail!("Kimi API error: {} - {}", status, text);
-        }
-
-        let mut content_buffer = String::new();
-        let mut has_reasoning = false;
-        let mut has_content = false;
-        let mut stream_ended = false;
-
-        let thinking_state = self.thinking_state.as_ref();
-
-        let mut byte_stream = response.bytes_stream();
-        let mut line_buffer = String::new();
-
-        use futures_util::StreamExt;
-
-        while let Some(chunk) = byte_stream.next().await {
-            let chunk = chunk.context("Failed to read streaming response chunk")?;
-            let chunk_str =
-                String::from_utf8(chunk.to_vec()).context("Invalid UTF-8 in stream chunk")?;
-
-            line_buffer.push_str(&chunk_str);
-
-            while let Some(line_end) = line_buffer.find('\n') {
-                let line = line_buffer[..line_end].trim().to_string();
-                line_buffer = line_buffer[line_end + 1..].to_string();
-
-                if line.is_empty() {
-                    continue;
-                }
-
-                if line == "data: [DONE]" {
-                    stream_ended = true;
-                    break;
-                }
-
-                if let Some(json_str) = line.strip_prefix("data: ") {
-                    match serde_json::from_str::<StreamChunk>(json_str) {
-                        Ok(chunk) => {
-                            for choice in &chunk.choices {
-                                if let Some(ref reasoning) = choice.delta.reasoning_content
-                                    && !reasoning.is_empty()
-                                {
-                                    if !has_reasoning {
-                                        has_reasoning = true;
-                                        if let Some(state) = thinking_state {
-                                            state.start_thinking();
-                                        }
-                                    }
-                                    continue;
-                                }
-
-                                if let Some(ref content) = choice.delta.content
-                                    && !content.is_empty()
-                                {
-                                    if has_reasoning
-                                        && !has_content
-                                        && let Some(state) = thinking_state
-                                    {
-                                        state.end_thinking();
-                                    }
-                                    has_content = true;
-                                    content_buffer.push_str(content);
-                                }
-
-                                if let Some(ref reason) = choice.finish_reason
-                                    && reason == "stop"
-                                {
-                                    stream_ended = true;
-                                }
-                            }
-                        }
-                        Err(_) => {
-                            // 忽略无法解析的行
-                        }
-                    }
-                }
-            }
-
-            if stream_ended {
-                break;
-            }
-        }
-
-        // 确保思考状态已结束
-        if let Some(state) = thinking_state {
-            state.end_thinking();
-        }
-
-        let result = content_buffer.trim().to_string();
-
-        if result.is_empty() {
-            if has_reasoning && !has_content {
-                bail!(
-                    "Kimi returned reasoning content but no final answer. \
-                     The model may have entered an incomplete thinking state. \
-                     Please try again or disable thinking mode."
-                );
-            }
-            bail!(
-                "No response from Kimi. \
-                 If thinking mode is enabled, try disabling it or ensure the model supports it."
-            );
-        }
-
-        Ok(result)
-    }
-}
-
-/// 可用 Kimi 模型列表
-pub const KIMI_MODELS: &[&str] = &[
-    // K2 系列（推荐）
-    "kimi-k2.6",
-    "kimi-k2.5",
-    "kimi-k2-thinking",
-    "kimi-k2-thinking-turbo",
-    "kimi-k2-instruct",
-    "kimi-k2-instruct-0905",
-    // 兼容旧版模型 ID
-    "moonshot-v1-8k",
-    "moonshot-v1-32k",
-    "moonshot-v1-128k",
-];
-
-pub fn is_valid_model(model: &str) -> bool {
-    KIMI_MODELS.contains(&model)
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-
-    #[test]
-    fn test_model_validation_k2() {
-        assert!(is_valid_model("kimi-k2.6"));
-        assert!(is_valid_model("kimi-k2.5"));
-        assert!(is_valid_model("kimi-k2-thinking"));
-        assert!(is_valid_model("kimi-k2-thinking-turbo"));
-        assert!(is_valid_model("moonshot-v1-8k"));
-        assert!(is_valid_model("moonshot-v1-32k"));
-        assert!(is_valid_model("moonshot-v1-128k"));
-        assert!(!is_valid_model("invalid-model"));
-        assert!(!is_valid_model("kimi-k1.5"));
-    }
-
-    #[test]
-    fn test_client_builder_defaults() {
-        let client = KimiClient::new("test-key", "kimi-k2.6").unwrap();
-        assert!(!client.thinking_enabled);
-        assert_eq!(client.max_tokens, 500);
-        assert_eq!(client.temperature, 1.0);
-        assert!(client.thinking_state.is_none());
-    }
-
-    #[test]
-    fn test_client_builder_with_thinking() {
-        let client = KimiClient::new("test-key", "kimi-k2.6")
-            .unwrap()
-            .with_thinking(true)
-            .with_max_tokens(1000)
-            .with_temperature(0.5);
-
-        assert!(client.thinking_enabled);
-        assert_eq!(client.max_tokens, 1000);
-        assert_eq!(client.temperature, 0.5);
-    }
-
-    #[test]
-    fn test_thinking_config_serialization() {
-        let config = ThinkingConfig {
-            thinking_type: "enabled".to_string(),
-        };
-        let json = serde_json::to_string(&config).unwrap();
-        assert_eq!(json, r#"{"type":"enabled"}"#);
-    }
-
-    #[test]
-    fn test_client_new_defaults() {
-        let client = KimiClient::new("test-key", "kimi-k2.6").unwrap();
-        assert_eq!(client.name(), "kimi");
-        assert!(!client.thinking_enabled);
-    }
-
-    #[test]
-    fn test_message_serialization() {
-        let msg = Message {
-            role: "user".to_string(),
-            content: "Hello".to_string(),
-            reasoning_content: None,
-        };
-        let json = serde_json::to_string(&msg).unwrap();
-        assert!(!json.contains("reasoning_content"));
-    }
-}

src/llm/ollama.rs

@@ -1,229 +0,0 @@
-use super::{LlmProvider, create_http_client};
-use anyhow::{Context, Result};
-use async_trait::async_trait;
-use serde::{Deserialize, Serialize};
-use std::time::Duration;
-
-/// Ollama API client
-pub struct OllamaClient {
-    base_url: String,
-    model: String,
-    client: reqwest::Client,
-    max_tokens: u32,
-    temperature: f32,
-    top_p: Option<f32>,
-}
-
-#[derive(Debug, Serialize)]
-struct GenerateRequest {
-    model: String,
-    prompt: String,
-    system: Option<String>,
-    stream: bool,
-    options: GenerationOptions,
-}
-
-#[derive(Debug, Serialize, Default)]
-struct GenerationOptions {
-    #[serde(skip_serializing_if = "Option::is_none")]
-    temperature: Option<f32>,
-    #[serde(skip_serializing_if = "Option::is_none")]
-    num_predict: Option<u32>,
-}
-
-#[derive(Debug, Deserialize)]
-struct GenerateResponse {
-    response: String,
-    done: bool,
-}
-
-#[derive(Debug, Deserialize)]
-struct ListModelsResponse {
-    models: Vec<ModelInfo>,
-}
-
-#[derive(Debug, Deserialize)]
-struct ModelInfo {
-    name: String,
-}
-
-impl OllamaClient {
-    /// Create new Ollama client
-    pub fn new(base_url: &str, model: &str) -> Self {
-        let client =
-            create_http_client(Duration::from_secs(120)).expect("Failed to create HTTP client");
-
-        Self {
-            base_url: base_url.trim_end_matches('/').to_string(),
-            model: model.to_string(),
-            client,
-            max_tokens: 500,
-            temperature: 0.7,
-            top_p: None,
-        }
-    }
-
-    /// Set timeout
-    pub fn with_timeout(mut self, timeout: Duration) -> Self {
-        self.client = create_http_client(timeout).expect("Failed to create HTTP client");
-        self
-    }
-
-    pub fn with_max_tokens(mut self, max_tokens: u32) -> Self {
-        self.max_tokens = max_tokens;
-        self
-    }
-
-    pub fn with_temperature(mut self, temperature: f32) -> Self {
-        self.temperature = temperature;
-        self
-    }
-
-    pub fn with_top_p(mut self, top_p: f32) -> Self {
-        self.top_p = Some(top_p);
-        self
-    }
-
-    /// List available models
-    pub async fn list_models(&self) -> Result<Vec<String>> {
-        let url = format!("{}/api/tags", self.base_url);
-
-        let response = self
-            .client
-            .get(&url)
-            .send()
-            .await
-            .context("Failed to list Ollama models")?;
-
-        if !response.status().is_success() {
-            let status = response.status();
-            let text = response.text().await.unwrap_or_default();
-            anyhow::bail!("Ollama API error: {} - {}", status, text);
-        }
-
-        let result: ListModelsResponse = response
-            .json()
-            .await
-            .context("Failed to parse Ollama response")?;
-
-        Ok(result.models.into_iter().map(|m| m.name).collect())
-    }
-
-    /// Pull a model
-    pub async fn pull_model(&self, model: &str) -> Result<()> {
-        let url = format!("{}/api/pull", self.base_url);
-
-        let request = serde_json::json!({
-            "name": model,
-            "stream": false,
-        });
-
-        let response = self
-            .client
-            .post(&url)
-            .json(&request)
-            .send()
-            .await
-            .context("Failed to pull Ollama model")?;
-
-        if !response.status().is_success() {
-            let status = response.status();
-            let text = response.text().await.unwrap_or_default();
-            anyhow::bail!("Ollama pull error: {} - {}", status, text);
-        }
-
-        Ok(())
-    }
-
-    /// Check if model exists
-    pub async fn model_exists(&self, model: &str) -> bool {
-        match self.list_models().await {
-            Ok(models) => models.contains(&model.to_string()),
-            Err(_) => false,
-        }
-    }
-}
-
-#[async_trait]
-impl LlmProvider for OllamaClient {
-    async fn generate(&self, prompt: &str) -> Result<String> {
-        self.generate_with_system("", prompt).await
-    }
-
-    async fn generate_with_system(&self, system: &str, user: &str) -> Result<String> {
-        let url = format!("{}/api/generate", self.base_url);
-
-        let system = if system.is_empty() {
-            None
-        } else {
-            Some(system.to_string())
-        };
-
-        let request = GenerateRequest {
-            model: self.model.clone(),
-            prompt: user.to_string(),
-            system,
-            stream: false,
-            options: GenerationOptions {
-                temperature: Some(self.temperature),
-                num_predict: Some(self.max_tokens),
-            },
-        };
-
-        let response = self
-            .client
-            .post(&url)
-            .json(&request)
-            .send()
-            .await
-            .context("Failed to send request to Ollama")?;
-
-        if !response.status().is_success() {
-            let status = response.status();
-            let text = response.text().await.unwrap_or_default();
-            anyhow::bail!("Ollama API error: {} - {}", status, text);
-        }
-
-        let result: GenerateResponse = response
-            .json()
-            .await
-            .context("Failed to parse Ollama response")?;
-
-        Ok(result.response.trim().to_string())
-    }
-
-    async fn is_available(&self) -> bool {
-        let url = format!("{}/api/tags", self.base_url);
-
-        match self.client.get(&url).send().await {
-            Ok(response) => response.status().is_success(),
-            Err(_) => false,
-        }
-    }
-
-    fn name(&self) -> &str {
-        "ollama"
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-
-    // These tests require a running Ollama server
-    #[tokio::test]
-    #[ignore]
-    async fn test_ollama_connection() {
-        let client = OllamaClient::new("http://localhost:11434", "llama2");
-        assert!(client.is_available().await);
-    }
-
-    #[tokio::test]
-    #[ignore]
-    async fn test_ollama_generate() {
-        let client = OllamaClient::new("http://localhost:11434", "llama2");
-        let response = client.generate("Hello, how are you?").await;
-        assert!(response.is_ok());
-        println!("Response: {}", response.unwrap());
-    }
-}

src/llm/openai.rs

@@ -1,659 +0,0 @@
-use super::thinking::ThinkingStateManager;
-use super::{LlmProvider, create_http_client};
-use anyhow::{Context, Result, bail};
-use async_trait::async_trait;
-use serde::{Deserialize, Serialize};
-use std::sync::Arc;
-use std::time::Duration;
-
-/// OpenAI API client with o-series reasoning support
-pub struct OpenAiClient {
-    base_url: String,
-    api_key: String,
-    model: String,
-    client: reqwest::Client,
-    thinking_enabled: bool,
-    reasoning_effort: Option<String>,
-    max_tokens: u32,
-    temperature: f32,
-    top_p: Option<f32>,
-    thinking_state: Option<Arc<ThinkingStateManager>>,
-}
-
-#[derive(Debug, Serialize)]
-struct ChatCompletionRequest {
-    model: String,
-    messages: Vec<Message>,
-    #[serde(skip_serializing_if = "Option::is_none")]
-    max_tokens: Option<u32>,
-    #[serde(skip_serializing_if = "Option::is_none")]
-    temperature: Option<f32>,
-    #[serde(skip_serializing_if = "Option::is_none")]
-    top_p: Option<f32>,
-    #[serde(skip_serializing_if = "Option::is_none")]
-    reasoning_effort: Option<String>,
-    stream: bool,
-}
-
-#[derive(Debug, Serialize, Deserialize, Clone)]
-struct Message {
-    role: String,
-    content: String,
-}
-
-#[derive(Debug, Deserialize)]
-struct ChatCompletionResponse {
-    choices: Vec<Choice>,
-}
-
-#[derive(Debug, Deserialize)]
-struct Choice {
-    message: Message,
-}
-
-// --- Streaming response structures ---
-
-#[derive(Debug, Deserialize)]
-struct StreamChunk {
-    choices: Vec<StreamChoice>,
-}
-
-#[derive(Debug, Deserialize)]
-struct StreamChoice {
-    delta: StreamDelta,
-    #[serde(default)]
-    finish_reason: Option<String>,
-}
-
-#[derive(Debug, Deserialize, Default)]
-struct StreamDelta {
-    #[serde(default)]
-    content: Option<String>,
-    #[serde(default)]
-    reasoning_content: Option<String>,
-}
-
-#[derive(Debug, Deserialize)]
-struct ErrorResponse {
-    error: ApiError,
-}
-
-#[derive(Debug, Deserialize)]
-struct ApiError {
-    message: String,
-    #[serde(rename = "type")]
-    error_type: String,
-}
-
-impl OpenAiClient {
-    /// Create new OpenAI client
-    pub fn new(base_url: &str, api_key: &str, model: &str) -> Result<Self> {
-        let client = create_http_client(Duration::from_secs(60))?;
-
-        Ok(Self {
-            base_url: base_url.trim_end_matches('/').to_string(),
-            api_key: api_key.to_string(),
-            model: model.to_string(),
-            client,
-            thinking_enabled: false,
-            reasoning_effort: None,
-            max_tokens: 500,
-            temperature: 0.7,
-            top_p: None,
-            thinking_state: None,
-        })
-    }
-
-    pub fn with_timeout(mut self, timeout: Duration) -> Result<Self> {
-        self.client = create_http_client(timeout)?;
-        Ok(self)
-    }
-
-    pub fn with_thinking(mut self, enabled: bool) -> Self {
-        self.thinking_enabled = enabled;
-        self
-    }
-
-    pub fn with_reasoning_effort(mut self, effort: Option<String>) -> Self {
-        self.reasoning_effort = effort;
-        self
-    }
-
-    pub fn with_max_tokens(mut self, max_tokens: u32) -> Self {
-        self.max_tokens = max_tokens;
-        self
-    }
-
-    pub fn with_temperature(mut self, temperature: f32) -> Self {
-        self.temperature = temperature;
-        self
-    }
-
-    pub fn with_top_p(mut self, top_p: f32) -> Self {
-        self.top_p = Some(top_p);
-        self
-    }
-
-    pub fn with_thinking_state(mut self, state: Arc<ThinkingStateManager>) -> Self {
-        self.thinking_state = Some(state);
-        self
-    }
-
-    pub async fn list_models(&self) -> Result<Vec<String>> {
-        let url = format!("{}/models", self.base_url);
-
-        let response = self
-            .client
-            .get(&url)
-            .header("Authorization", format!("Bearer {}", self.api_key))
-            .send()
-            .await
-            .context("Failed to list OpenAI models")?;
-
-        if !response.status().is_success() {
-            let status = response.status();
-            let text = response.text().await.unwrap_or_default();
-            bail!("OpenAI API error: {} - {}", status, text);
-        }
-
-        #[derive(Deserialize)]
-        struct ModelsResponse {
-            data: Vec<Model>,
-        }
-
-        #[derive(Deserialize)]
-        struct Model {
-            id: String,
-        }
-
-        let result: ModelsResponse = response
-            .json()
-            .await
-            .context("Failed to parse OpenAI response")?;
-
-        Ok(result.data.into_iter().map(|m| m.id).collect())
-    }
-
-    pub async fn validate_key(&self) -> Result<bool> {
-        match self.list_models().await {
-            Ok(_) => Ok(true),
-            Err(e) => {
-                let err_str = e.to_string();
-                if err_str.contains("401") || err_str.contains("Unauthorized") {
-                    Ok(false)
-                } else {
-                    Err(e)
-                }
-            }
-        }
-    }
-}
-
-#[async_trait]
-impl LlmProvider for OpenAiClient {
-    async fn generate(&self, prompt: &str) -> Result<String> {
-        let messages = vec![Message {
-            role: "user".to_string(),
-            content: prompt.to_string(),
-        }];
-
-        self.chat_completion_with_retry(messages).await
-    }
-
-    async fn generate_with_system(&self, system: &str, user: &str) -> Result<String> {
-        let mut messages = vec![];
-
-        if !system.is_empty() {
-            messages.push(Message {
-                role: "system".to_string(),
-                content: system.to_string(),
-            });
-        }
-
-        messages.push(Message {
-            role: "user".to_string(),
-            content: user.to_string(),
-        });
-
-        self.chat_completion_with_retry(messages).await
-    }
-
-    async fn is_available(&self) -> bool {
-        self.validate_key().await.unwrap_or(false)
-    }
-
-    fn name(&self) -> &str {
-        "openai"
-    }
-}
-
-impl OpenAiClient {
-    async fn chat_completion_with_retry(&self, messages: Vec<Message>) -> Result<String> {
-        let mut last_error = None;
-
-        for attempt in 1..=3 {
-            match self.chat_completion(messages.clone()).await {
-                Ok(result) => return Ok(result),
-                Err(e) => {
-                    let err_msg = e.to_string();
-                    let is_retryable = err_msg.contains("timeout")
-                        || err_msg.contains("connection")
-                        || err_msg.contains("temporary")
-                        || err_msg.contains("5")
-                            && (err_msg.contains("500")
-                                || err_msg.contains("502")
-                                || err_msg.contains("503")
-                                || err_msg.contains("504"));
-
-                    if !is_retryable || attempt == 3 {
-                        last_error = Some(e);
-                        break;
-                    }
-
-                    tokio::time::sleep(Duration::from_millis(500 * 2u64.pow(attempt - 1))).await;
-                }
-            }
-        }
-
-        Err(last_error.unwrap_or_else(|| anyhow::anyhow!("Request failed after retries")))
-    }
-
-    async fn chat_completion(&self, messages: Vec<Message>) -> Result<String> {
-        if self.thinking_enabled {
-            self.streaming_chat_completion(messages).await
-        } else {
-            self.non_streaming_chat_completion(messages).await
-        }
-    }
-
-    async fn non_streaming_chat_completion(&self, messages: Vec<Message>) -> Result<String> {
-        let url = format!("{}/chat/completions", self.base_url);
-
-        let request = ChatCompletionRequest {
-            model: self.model.clone(),
-            messages,
-            max_tokens: Some(self.max_tokens),
-            temperature: Some(self.temperature),
-            top_p: self.top_p,
-            reasoning_effort: if is_reasoning_model(&self.model) {
-                Some("none".to_string())
-            } else {
-                None
-            },
-            stream: false,
-        };
-
-        let response = self
-            .client
-            .post(&url)
-            .header("Authorization", format!("Bearer {}", self.api_key))
-            .header("Content-Type", "application/json")
-            .json(&request)
-            .send()
-            .await
-            .context("Failed to send request to OpenAI")?;
-
-        let status = response.status();
-
-        if !status.is_success() {
-            let text = response.text().await.unwrap_or_default();
-
-            if let Ok(error) = serde_json::from_str::<ErrorResponse>(&text) {
-                bail!(
-                    "OpenAI API error: {} ({})",
-                    error.error.message,
-                    error.error.error_type
-                );
-            }
-
-            bail!("OpenAI API error: {} - {}", status, text);
-        }
-
-        let result: ChatCompletionResponse = response
-            .json()
-            .await
-            .context("Failed to parse OpenAI response")?;
-
-        result
-            .choices
-            .into_iter()
-            .next()
-            .map(|c| c.message.content.trim().to_string())
-            .filter(|s| !s.is_empty())
-            .ok_or_else(|| anyhow::anyhow!("No response from OpenAI"))
-    }
-
-    /// Streaming request for reasoning mode, filters reasoning_content from output
-    async fn streaming_chat_completion(&self, messages: Vec<Message>) -> Result<String> {
-        let url = format!("{}/chat/completions", self.base_url);
-
-        // For reasoning/thinking mode, omit temperature and top_p
-        let request = ChatCompletionRequest {
-            model: self.model.clone(),
-            messages,
-            max_tokens: Some(self.max_tokens),
-            temperature: None,
-            top_p: None,
-            reasoning_effort: self.reasoning_effort.clone(),
-            stream: true,
-        };
-
-        let response = self
-            .client
-            .post(&url)
-            .header("Authorization", format!("Bearer {}", self.api_key))
-            .header("Content-Type", "application/json")
-            .header("Accept", "text/event-stream")
-            .json(&request)
-            .send()
-            .await
-            .context("Failed to send streaming request to OpenAI")?;
-
-        let status = response.status();
-
-        if !status.is_success() {
-            let text = response.text().await.unwrap_or_default();
-
-            if let Ok(error) = serde_json::from_str::<ErrorResponse>(&text) {
-                bail!(
-                    "OpenAI API error: {} ({})",
-                    error.error.message,
-                    error.error.error_type
-                );
-            }
-
-            bail!("OpenAI API error: {} - {}", status, text);
-        }
-
-        let mut content_buffer = String::new();
-        let mut has_reasoning = false;
-        let mut has_content = false;
-
-        let thinking_state = self.thinking_state.as_ref();
-
-        let mut byte_stream = response.bytes_stream();
-        let mut line_buffer = String::new();
-
-        use futures_util::StreamExt;
-
-        while let Some(chunk) = byte_stream.next().await {
-            let chunk = chunk.context("Failed to read streaming response chunk")?;
-            let chunk_str =
-                String::from_utf8(chunk.to_vec()).context("Invalid UTF-8 in stream chunk")?;
-
-            line_buffer.push_str(&chunk_str);
-
-            while let Some(line_end) = line_buffer.find('\n') {
-                let line = line_buffer[..line_end].trim().to_string();
-                line_buffer = line_buffer[line_end + 1..].to_string();
-
-                if line.is_empty() {
-                    continue;
-                }
-
-                if line == "data: [DONE]" {
-                    break;
-                }
-
-                if let Some(json_str) = line.strip_prefix("data: ") {
-                    if let Ok(chunk) = serde_json::from_str::<StreamChunk>(json_str) {
-                        for choice in &chunk.choices {
-                            // Handle reasoning_content (o-series)
-                            if let Some(ref reasoning) = choice.delta.reasoning_content
-                                && !reasoning.is_empty()
-                            {
-                                if !has_reasoning {
-                                    has_reasoning = true;
-                                    if let Some(state) = thinking_state {
-                                        state.start_thinking();
-                                    }
-                                }
-                                continue;
-                            }
-
-                            // Handle content
-                            if let Some(ref content) = choice.delta.content
-                                && !content.is_empty()
-                            {
-                                if has_reasoning
-                                    && !has_content
-                                    && let Some(state) = thinking_state
-                                {
-                                    state.end_thinking();
-                                }
-                                has_content = true;
-                                content_buffer.push_str(content);
-                            }
-                        }
-                    }
-                }
-            }
-        }
-
-        if let Some(state) = thinking_state {
-            state.end_thinking();
-        }
-
-        let result = content_buffer.trim().to_string();
-
-        if result.is_empty() {
-            if has_reasoning && !has_content {
-                bail!(
-                    "OpenAI returned reasoning content but no final answer. \
-                     The model may have entered an incomplete reasoning state. \
-                     Please try again or disable thinking mode."
-                );
-            }
-            bail!(
-                "No response from OpenAI. \
-                 If thinking mode is enabled, try disabling it or ensure the model supports reasoning."
-            );
-        }
-
-        Ok(result)
-    }
-}
-
-/// Azure OpenAI client (extends OpenAI with Azure-specific config)
-pub struct AzureOpenAiClient {
-    endpoint: String,
-    api_key: String,
-    deployment: String,
-    api_version: String,
-    client: reqwest::Client,
-    thinking_enabled: bool,
-    reasoning_effort: Option<String>,
-    max_tokens: u32,
-    temperature: f32,
-    top_p: Option<f32>,
-    thinking_state: Option<Arc<ThinkingStateManager>>,
-}
-
-impl AzureOpenAiClient {
-    pub fn new(endpoint: &str, api_key: &str, deployment: &str, api_version: &str) -> Result<Self> {
-        let client = create_http_client(Duration::from_secs(60))?;
-
-        Ok(Self {
-            endpoint: endpoint.trim_end_matches('/').to_string(),
-            api_key: api_key.to_string(),
-            deployment: deployment.to_string(),
-            api_version: api_version.to_string(),
-            client,
-            thinking_enabled: false,
-            reasoning_effort: None,
-            max_tokens: 500,
-            temperature: 0.7,
-            top_p: None,
-            thinking_state: None,
-        })
-    }
-
-    async fn chat_completion(&self, messages: Vec<Message>) -> Result<String> {
-        let url = format!(
-            "{}/openai/deployments/{}/chat/completions?api-version={}",
-            self.endpoint, self.deployment, self.api_version
-        );
-
-        let request = ChatCompletionRequest {
-            model: self.deployment.clone(),
-            messages,
-            max_tokens: Some(self.max_tokens),
-            temperature: Some(self.temperature),
-            top_p: self.top_p,
-            reasoning_effort: self.reasoning_effort.clone(),
-            stream: false,
-        };
-
-        let response = self
-            .client
-            .post(&url)
-            .header("api-key", &self.api_key)
-            .header("Content-Type", "application/json")
-            .json(&request)
-            .send()
-            .await
-            .context("Failed to send request to Azure OpenAI")?;
-
-        if !response.status().is_success() {
-            let status = response.status();
-            let text = response.text().await.unwrap_or_default();
-            bail!("Azure OpenAI API error: {} - {}", status, text);
-        }
-
-        let result: ChatCompletionResponse = response
-            .json()
-            .await
-            .context("Failed to parse Azure OpenAI response")?;
-
-        result
-            .choices
-            .into_iter()
-            .next()
-            .map(|c| c.message.content.trim().to_string())
-            .filter(|s| !s.is_empty())
-            .ok_or_else(|| anyhow::anyhow!("No response from Azure OpenAI"))
-    }
-}
-
-#[async_trait]
-impl LlmProvider for AzureOpenAiClient {
-    async fn generate(&self, prompt: &str) -> Result<String> {
-        let messages = vec![Message {
-            role: "user".to_string(),
-            content: prompt.to_string(),
-        }];
-
-        self.chat_completion(messages).await
-    }
-
-    async fn generate_with_system(&self, system: &str, user: &str) -> Result<String> {
-        let mut messages = vec![];
-
-        if !system.is_empty() {
-            messages.push(Message {
-                role: "system".to_string(),
-                content: system.to_string(),
-            });
-        }
-
-        messages.push(Message {
-            role: "user".to_string(),
-            content: user.to_string(),
-        });
-
-        self.chat_completion(messages).await
-    }
-
-    async fn is_available(&self) -> bool {
-        let url = format!(
-            "{}/openai/deployments/{}/chat/completions?api-version={}",
-            self.endpoint, self.deployment, self.api_version
-        );
-
-        let request = ChatCompletionRequest {
-            model: self.deployment.clone(),
-            messages: vec![Message {
-                role: "user".to_string(),
-                content: "Hi".to_string(),
-            }],
-            max_tokens: Some(5),
-            temperature: Some(0.0),
-            top_p: None,
-            reasoning_effort: None,
-            stream: false,
-        };
-
-        match self
-            .client
-            .post(&url)
-            .header("api-key", &self.api_key)
-            .json(&request)
-            .send()
-            .await
-        {
-            Ok(response) => response.status().is_success(),
-            Err(_) => false,
-        }
-    }
-
-    fn name(&self) -> &str {
-        "azure-openai"
-    }
-}
-
-/// Available OpenAI models (including o-series with reasoning)
-pub const OPENAI_MODELS: &[&str] = &[
-    "o4-mini",
-    "o3",
-    "o3-mini",
-    "o1",
-    "o1-mini",
-    "o1-pro",
-    "gpt-4.1",
-    "gpt-4.1-mini",
-    "gpt-4.1-nano",
-    "gpt-4o",
-    "gpt-4o-mini",
-    "gpt-4-turbo",
-    "gpt-4",
-    "gpt-3.5-turbo",
-];
-
-pub fn is_valid_model(model: &str) -> bool {
-    OPENAI_MODELS.contains(&model)
-}
-
-fn is_reasoning_model(model: &str) -> bool {
-    model.starts_with("o")
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-
-    #[test]
-    fn test_model_validation_o_series() {
-        assert!(is_valid_model("o4-mini"));
-        assert!(is_valid_model("o3"));
-        assert!(is_valid_model("o1"));
-        assert!(is_valid_model("gpt-4o"));
-        assert!(is_valid_model("gpt-3.5-turbo"));
-        assert!(!is_valid_model("invalid-model"));
-    }
-
-    #[test]
-    fn test_stream_delta_reasoning_parsing() {
-        let json = r#"{"content":null,"reasoning_content":"Let me think..."}"#;
-        let delta: StreamDelta = serde_json::from_str(json).unwrap();
-        assert!(delta.content.is_none());
-        assert_eq!(delta.reasoning_content, Some("Let me think...".to_string()));
-    }
-
-    #[test]
-    fn test_stream_delta_content_parsing() {
-        let json = r#"{"content":"Hello","reasoning_content":null}"#;
-        let delta: StreamDelta = serde_json::from_str(json).unwrap();
-        assert_eq!(delta.content, Some("Hello".to_string()));
-        assert!(delta.reasoning_content.is_none());
-    }
-}

src/llm/openrouter.rs

@@ -1,286 +0,0 @@
-use super::{LlmProvider, create_http_client};
-use anyhow::{Context, Result, bail};
-use async_trait::async_trait;
-use serde::{Deserialize, Serialize};
-use std::time::Duration;
-
-/// OpenRouter API client
-pub struct OpenRouterClient {
-    base_url: String,
-    api_key: String,
-    model: String,
-    client: reqwest::Client,
-    max_tokens: u32,
-    temperature: f32,
-    top_p: Option<f32>,
-}
-
-#[derive(Debug, Serialize)]
-struct ChatCompletionRequest {
-    model: String,
-    messages: Vec<Message>,
-    #[serde(skip_serializing_if = "Option::is_none")]
-    max_tokens: Option<u32>,
-    #[serde(skip_serializing_if = "Option::is_none")]
-    temperature: Option<f32>,
-    stream: bool,
-}
-
-#[derive(Debug, Serialize, Deserialize)]
-struct Message {
-    role: String,
-    content: String,
-}
-
-#[derive(Debug, Deserialize)]
-struct ChatCompletionResponse {
-    choices: Vec<Choice>,
-}
-
-#[derive(Debug, Deserialize)]
-struct Choice {
-    message: Message,
-}
-
-#[derive(Debug, Deserialize)]
-struct ErrorResponse {
-    error: ApiError,
-}
-
-#[derive(Debug, Deserialize)]
-struct ApiError {
-    message: String,
-    #[serde(rename = "type")]
-    error_type: String,
-}
-
-impl OpenRouterClient {
-    /// Create new OpenRouter client
-    pub fn new(api_key: &str, model: &str) -> Result<Self> {
-        let client = create_http_client(Duration::from_secs(60))?;
-
-        Ok(Self {
-            base_url: "https://openrouter.ai/api/v1".to_string(),
-            api_key: api_key.to_string(),
-            model: model.to_string(),
-            client,
-            max_tokens: 500,
-            temperature: 0.7,
-            top_p: None,
-        })
-    }
-
-    /// Create with custom base URL
-    pub fn with_base_url(api_key: &str, model: &str, base_url: &str) -> Result<Self> {
-        let client = create_http_client(Duration::from_secs(60))?;
-
-        Ok(Self {
-            base_url: base_url.trim_end_matches('/').to_string(),
-            api_key: api_key.to_string(),
-            model: model.to_string(),
-            client,
-            max_tokens: 500,
-            temperature: 0.7,
-            top_p: None,
-        })
-    }
-
-    /// Set timeout
-    pub fn with_timeout(mut self, timeout: Duration) -> Result<Self> {
-        self.client = create_http_client(timeout)?;
-        Ok(self)
-    }
-
-    pub fn with_max_tokens(mut self, max_tokens: u32) -> Self {
-        self.max_tokens = max_tokens;
-        self
-    }
-
-    pub fn with_temperature(mut self, temperature: f32) -> Self {
-        self.temperature = temperature;
-        self
-    }
-
-    pub fn with_top_p(mut self, top_p: f32) -> Self {
-        self.top_p = Some(top_p);
-        self
-    }
-
-    /// List available models
-    pub async fn list_models(&self) -> Result<Vec<String>> {
-        let url = format!("{}/models", self.base_url);
-
-        let response = self
-            .client
-            .get(&url)
-            .header("Authorization", format!("Bearer {}", self.api_key))
-            .header("HTTP-Referer", "https://quicommit.dev")
-            .header("X-Title", "QuiCommit")
-            .send()
-            .await
-            .context("Failed to list OpenRouter models")?;
-
-        if !response.status().is_success() {
-            let status = response.status();
-            let text = response.text().await.unwrap_or_default();
-            bail!("OpenRouter API error: {} - {}", status, text);
-        }
-
-        #[derive(Deserialize)]
-        struct ModelsResponse {
-            data: Vec<Model>,
-        }
-
-        #[derive(Deserialize)]
-        struct Model {
-            id: String,
-        }
-
-        let result: ModelsResponse = response
-            .json()
-            .await
-            .context("Failed to parse OpenRouter response")?;
-
-        Ok(result.data.into_iter().map(|m| m.id).collect())
-    }
-
-    /// Validate API key
-    pub async fn validate_key(&self) -> Result<bool> {
-        match self.list_models().await {
-            Ok(_) => Ok(true),
-            Err(e) => {
-                let err_str = e.to_string();
-                if err_str.contains("401") || err_str.contains("Unauthorized") {
-                    Ok(false)
-                } else {
-                    Err(e)
-                }
-            }
-        }
-    }
-}
-
-#[async_trait]
-impl LlmProvider for OpenRouterClient {
-    async fn generate(&self, prompt: &str) -> Result<String> {
-        let messages = vec![Message {
-            role: "user".to_string(),
-            content: prompt.to_string(),
-        }];
-
-        self.chat_completion(messages).await
-    }
-
-    async fn generate_with_system(&self, system: &str, user: &str) -> Result<String> {
-        let mut messages = vec![];
-
-        if !system.is_empty() {
-            messages.push(Message {
-                role: "system".to_string(),
-                content: system.to_string(),
-            });
-        }
-
-        messages.push(Message {
-            role: "user".to_string(),
-            content: user.to_string(),
-        });
-
-        self.chat_completion(messages).await
-    }
-
-    async fn is_available(&self) -> bool {
-        self.validate_key().await.unwrap_or(false)
-    }
-
-    fn name(&self) -> &str {
-        "openrouter"
-    }
-}
-
-impl OpenRouterClient {
-    async fn chat_completion(&self, messages: Vec<Message>) -> Result<String> {
-        let url = format!("{}/chat/completions", self.base_url);
-
-        let request = ChatCompletionRequest {
-            model: self.model.clone(),
-            messages,
-            max_tokens: Some(self.max_tokens),
-            temperature: Some(self.temperature),
-            stream: false,
-        };
-
-        let response = self
-            .client
-            .post(&url)
-            .header("Authorization", format!("Bearer {}", self.api_key))
-            .header("Content-Type", "application/json")
-            .header("HTTP-Referer", "https://quicommit.dev")
-            .header("X-Title", "QuiCommit")
-            .json(&request)
-            .send()
-            .await
-            .context("Failed to send request to OpenRouter")?;
-
-        let status = response.status();
-
-        if !status.is_success() {
-            let text = response.text().await.unwrap_or_default();
-
-            // Try to parse error
-            if let Ok(error) = serde_json::from_str::<ErrorResponse>(&text) {
-                bail!(
-                    "OpenRouter API error: {} ({})",
-                    error.error.message,
-                    error.error.error_type
-                );
-            }
-
-            bail!("OpenRouter API error: {} - {}", status, text);
-        }
-
-        let result: ChatCompletionResponse = response
-            .json()
-            .await
-            .context("Failed to parse OpenRouter response")?;
-
-        result
-            .choices
-            .into_iter()
-            .next()
-            .map(|c| c.message.content.trim().to_string())
-            .ok_or_else(|| anyhow::anyhow!("No response from OpenRouter"))
-    }
-}
-
-/// Popular OpenRouter models
-pub const OPENROUTER_MODELS: &[&str] = &[
-    "openai/gpt-3.5-turbo",
-    "openai/gpt-4",
-    "openai/gpt-4-turbo",
-    "anthropic/claude-3-opus",
-    "anthropic/claude-3-sonnet",
-    "anthropic/claude-3-haiku",
-    "google/gemini-pro",
-    "meta-llama/llama-2-70b-chat",
-    "mistralai/mixtral-8x7b-instruct",
-    "01-ai/yi-34b-chat",
-];
-
-/// Check if a model name is valid
-pub fn is_valid_model(_model: &str) -> bool {
-    // Since OpenRouter supports many models, we'll allow any model name
-    // but provide some popular ones as suggestions
-    true
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-
-    #[test]
-    fn test_model_validation() {
-        assert!(is_valid_model("openai/gpt-4"));
-        assert!(is_valid_model("custom/model"));
-    }
-}

src/llm/parsing.rs

@@ -0,0 +1,281 @@
+use anyhow::{Result, bail};
+
+/// Parse commit response from LLM
+pub(crate) fn parse_commit_response(
+    response: &str,
+    format: crate::config::CommitFormat,
+) -> Result<GeneratedCommit> {
+    // Clean markdown code fences from the response
+    let cleaned = strip_code_fences(response);
+
+    let lines: Vec<&str> = cleaned
+        .lines()
+        .map(|l| l.trim())
+        .filter(|l| !l.is_empty())
+        .collect();
+
+    if lines.is_empty() {
+        let preview: String = response.chars().take(200).collect();
+        bail!(
+            "LLM returned empty or whitespace-only response. \
+             Raw response preview: '{}'. \
+             Hint: If using DeepSeek/Kimi with thinking enabled, \
+             the model may have returned reasoning_content only. \
+             Try disabling thinking mode or switching models.",
+            preview
+        );
+    }
+
+    // Find the line most likely to be the commit subject
+    let first_line = find_commit_subject_line(&lines, format);
+
+    // Parse based on format
+    match format {
+        crate::config::CommitFormat::Conventional => {
+            parse_conventional_commit(first_line, &lines, response)
+        }
+        crate::config::CommitFormat::Commitlint => {
+            parse_commitlint_commit(first_line, &lines, response)
+        }
+    }
+}
+
+/// Remove surrounding markdown code fences (```) from LLM output
+fn strip_code_fences(response: &str) -> String {
+    let mut lines: Vec<&str> = response.lines().collect();
+
+    // Strip leading fence lines (``` or ```lang)
+    while lines.first().map_or(false, |l| l.trim().starts_with("```")) {
+        lines.remove(0);
+    }
+
+    // Strip trailing fence lines
+    while lines.last().map_or(false, |l| l.trim() == "```") {
+        lines.pop();
+    }
+
+    lines.join("\n")
+}
+
+/// Find the line that is most likely the commit subject among extracted lines
+fn find_commit_subject_line<'a>(
+    lines: &[&'a str],
+    format: crate::config::CommitFormat,
+) -> &'a str {
+    let valid_types = crate::utils::validators::get_commit_types(matches!(
+        format,
+        crate::config::CommitFormat::Commitlint
+    ));
+
+    // First pass: line starting with a known type that also has proper syntax
+    // (e.g. "type:", "type(scope):", "type!:")
+    for &line in lines {
+        let trimmed = line.trim();
+        for &t in valid_types {
+            if let Some(rest) = trimmed.strip_prefix(t) {
+                if rest.starts_with(':') || rest.starts_with('(') || rest.starts_with("!:") {
+                    return trimmed;
+                }
+            }
+        }
+    }
+
+    // Second pass: any line containing a colon (generic "prefix: description")
+    for &line in lines {
+        if line.contains(':') {
+            return line.trim();
+        }
+    }
+
+    // Fallback: return the first line as-is
+    lines[0].trim()
+}
+
+fn parse_conventional_commit(
+    first_line: &str,
+    lines: &[&str],
+    raw_response: &str,
+) -> Result<GeneratedCommit> {
+    // Parse: type(scope)!: description
+    let parts: Vec<&str> = first_line.splitn(2, ':').collect();
+    if parts.len() != 2 {
+        let preview: String = raw_response.chars().take(300).collect();
+        bail!(
+            "Invalid conventional commit format: missing colon.\n\
+             Parsed subject line: '{}'\n\
+             Raw response preview: '{}'\n\
+             Expected: <type>[optional scope]: <description>",
+            first_line,
+            preview
+        );
+    }
+
+    let type_part = parts[0];
+    let description = parts[1].trim();
+
+    // Extract type, scope, and breaking indicator
+    let breaking = type_part.ends_with('!');
+    let type_part = type_part.trim_end_matches('!');
+
+    let (commit_type, scope) = if let Some(start) = type_part.find('(') {
+        if let Some(end) = type_part.find(')') {
+            let t = &type_part[..start];
+            let s = &type_part[start + 1..end];
+            (t.to_string(), Some(s.to_string()))
+        } else {
+            bail!("Invalid scope format: missing closing parenthesis");
+        }
+    } else {
+        (type_part.to_string(), None)
+    };
+
+    // Extract body and footer
+    let (body, footer) = extract_body_footer(lines);
+
+    Ok(GeneratedCommit {
+        commit_type,
+        scope,
+        description: description.to_string(),
+        body,
+        footer,
+        breaking,
+    })
+}
+
+fn parse_commitlint_commit(
+    first_line: &str,
+    lines: &[&str],
+    raw_response: &str,
+) -> Result<GeneratedCommit> {
+    // Similar parsing but with commitlint rules
+    let parts: Vec<&str> = first_line.splitn(2, ':').collect();
+    if parts.len() != 2 {
+        let preview: String = raw_response.chars().take(300).collect();
+        bail!(
+            "Invalid commit format: missing colon.\n\
+             Parsed subject line: '{}'\n\
+             Raw response preview: '{}'\n\
+             Expected: <type>[optional scope]: <subject>",
+            first_line,
+            preview
+        );
+    }
+
+    let type_part = parts[0];
+    let subject = parts[1].trim();
+
+    let (commit_type, scope) = if let Some(start) = type_part.find('(') {
+        if let Some(end) = type_part.find(')') {
+            let t = &type_part[..start];
+            let s = &type_part[start + 1..end];
+            (t.to_string(), Some(s.to_string()))
+        } else {
+            (type_part.to_string(), None)
+        }
+    } else {
+        (type_part.to_string(), None)
+    };
+
+    let (body, footer) = extract_body_footer(&lines);
+
+    Ok(GeneratedCommit {
+        commit_type,
+        scope,
+        description: subject.to_string(),
+        body,
+        footer,
+        breaking: false,
+    })
+}
+
+fn extract_body_footer(lines: &[&str]) -> (Option<String>, Option<String>) {
+    if lines.len() <= 1 {
+        return (None, None);
+    }
+
+    let rest: Vec<&str> = lines[1..]
+        .iter()
+        .skip_while(|l| l.trim().is_empty())
+        .copied()
+        .collect();
+
+    if rest.is_empty() {
+        return (None, None);
+    }
+
+    // Look for footer markers
+    let footer_markers = [
+        "BREAKING CHANGE:",
+        "Closes",
+        "Fixes",
+        "Refs",
+        "Co-authored-by:",
+    ];
+
+    let mut body_lines = vec![];
+    let mut footer_lines = vec![];
+    let mut in_footer = false;
+
+    for line in &rest {
+        if footer_markers.iter().any(|m| line.starts_with(m)) {
+            in_footer = true;
+        }
+
+        if in_footer {
+            footer_lines.push(*line);
+        } else {
+            body_lines.push(*line);
+        }
+    }
+
+    let body = if body_lines.is_empty() {
+        None
+    } else {
+        Some(body_lines.join("\n"))
+    };
+
+    let footer = if footer_lines.is_empty() {
+        None
+    } else {
+        Some(footer_lines.join("\n"))
+    };
+
+    (body, footer)
+}
+
+/// Generated commit structure
+#[derive(Debug, Clone)]
+pub struct GeneratedCommit {
+    pub commit_type: String,
+    pub scope: Option<String>,
+    pub description: String,
+    pub body: Option<String>,
+    pub footer: Option<String>,
+    pub breaking: bool,
+}
+
+impl GeneratedCommit {
+    /// Format as conventional commit
+    pub fn to_conventional(&self) -> String {
+        crate::utils::formatter::format_conventional_commit(
+            &self.commit_type,
+            self.scope.as_deref(),
+            &self.description,
+            self.body.as_deref(),
+            self.footer.as_deref(),
+            self.breaking,
+        )
+    }
+
+    /// Format as commitlint commit
+    pub fn to_commitlint(&self) -> String {
+        crate::utils::formatter::format_commitlint_commit(
+            &self.commit_type,
+            self.scope.as_deref(),
+            &self.description,
+            self.body.as_deref(),
+            self.footer.as_deref(),
+            None,
+        )
+    }
+}

src/llm/prompts.rs

@@ -0,0 +1,618 @@
+use crate::config::Language;
+
+/// Get commit system prompt based on format and language
+pub(crate) fn get_commit_system_prompt(
+    format: crate::config::CommitFormat,
+    language: Language,
+) -> &'static str {
+    match (format, language) {
+        (crate::config::CommitFormat::Conventional, Language::Chinese) => {
+            CONVENTIONAL_COMMIT_SYSTEM_PROMPT_ZH
+        }
+        (crate::config::CommitFormat::Conventional, Language::Japanese) => {
+            CONVENTIONAL_COMMIT_SYSTEM_PROMPT_JA
+        }
+        (crate::config::CommitFormat::Conventional, Language::Korean) => {
+            CONVENTIONAL_COMMIT_SYSTEM_PROMPT_KO
+        }
+        (crate::config::CommitFormat::Conventional, Language::Spanish) => {
+            CONVENTIONAL_COMMIT_SYSTEM_PROMPT_ES
+        }
+        (crate::config::CommitFormat::Conventional, Language::French) => {
+            CONVENTIONAL_COMMIT_SYSTEM_PROMPT_FR
+        }
+        (crate::config::CommitFormat::Conventional, Language::German) => {
+            CONVENTIONAL_COMMIT_SYSTEM_PROMPT_DE
+        }
+        (crate::config::CommitFormat::Conventional, _) => CONVENTIONAL_COMMIT_SYSTEM_PROMPT,
+        (crate::config::CommitFormat::Commitlint, Language::Chinese) => COMMITLINT_SYSTEM_PROMPT_ZH,
+        (crate::config::CommitFormat::Commitlint, Language::Japanese) => {
+            COMMITLINT_SYSTEM_PROMPT_JA
+        }
+        (crate::config::CommitFormat::Commitlint, Language::Korean) => COMMITLINT_SYSTEM_PROMPT_KO,
+        (crate::config::CommitFormat::Commitlint, Language::Spanish) => COMMITLINT_SYSTEM_PROMPT_ES,
+        (crate::config::CommitFormat::Commitlint, Language::French) => COMMITLINT_SYSTEM_PROMPT_FR,
+        (crate::config::CommitFormat::Commitlint, Language::German) => COMMITLINT_SYSTEM_PROMPT_DE,
+        (crate::config::CommitFormat::Commitlint, _) => COMMITLINT_SYSTEM_PROMPT,
+    }
+}
+
+pub(crate) fn get_tag_system_prompt(language: Language) -> &'static str {
+    match language {
+        Language::Chinese => TAG_MESSAGE_SYSTEM_PROMPT_ZH,
+        Language::Japanese => TAG_MESSAGE_SYSTEM_PROMPT_JA,
+        Language::Korean => TAG_MESSAGE_SYSTEM_PROMPT_KO,
+        Language::Spanish => TAG_MESSAGE_SYSTEM_PROMPT_ES,
+        Language::French => TAG_MESSAGE_SYSTEM_PROMPT_FR,
+        Language::German => TAG_MESSAGE_SYSTEM_PROMPT_DE,
+        _ => TAG_MESSAGE_SYSTEM_PROMPT,
+    }
+}
+
+pub(crate) fn get_changelog_system_prompt(language: Language) -> &'static str {
+    match language {
+        Language::Chinese => CHANGELOG_SYSTEM_PROMPT_ZH,
+        Language::Japanese => CHANGELOG_SYSTEM_PROMPT_JA,
+        Language::Korean => CHANGELOG_SYSTEM_PROMPT_KO,
+        Language::Spanish => CHANGELOG_SYSTEM_PROMPT_ES,
+        Language::French => CHANGELOG_SYSTEM_PROMPT_FR,
+        Language::German => CHANGELOG_SYSTEM_PROMPT_DE,
+        _ => CHANGELOG_SYSTEM_PROMPT,
+    }
+}
+
+// System prompts for LLM
+
+const CONVENTIONAL_COMMIT_SYSTEM_PROMPT: &str = r#"You are a helpful assistant that generates conventional commit messages.
+
+Analyze the git diff provided and generate a commit message following the Conventional Commits specification.
+
+Format: <type>[optional scope]: <description>
+
+Types:
+- feat: A new feature
+- fix: A bug fix
+- docs: Documentation only changes
+- style: Changes that don't affect code meaning (formatting, semicolons, etc.)
+- refactor: Code change that neither fixes a bug nor adds a feature
+- perf: Code change that improves performance
+- test: Adding or correcting tests
+- build: Changes to build system or dependencies
+- ci: Changes to CI configuration
+- chore: Other changes that don't modify src or test files
+- revert: Reverts a previous commit
+
+Rules:
+1. Use lowercase for type and scope
+2. Keep description under 100 characters
+3. Use imperative mood ("add" not "added")
+4. Don't capitalize first letter
+5. No period at the end
+6. Include scope if the change is specific to a module/component
+
+Output ONLY the commit message, nothing else.
+"#;
+
+const CONVENTIONAL_COMMIT_SYSTEM_PROMPT_ZH: &str = r#"你是一个生成符合 Conventional Commits 规范的提交消息的助手。
+
+分析提供的 git diff，并生成符合 Conventional Commits 规范的提交消息。
+
+格式: <type>[可选作用域]: <描述>
+
+类型:
+- feat: 新功能
+- fix: 修复错误
+- docs: 仅文档更改
+- style: 不影响代码含义的更改（格式化、分号等）
+- refactor: 既不修复错误也不添加功能的代码更改
+- perf: 提高性能的代码更改
+- test: 添加或更正测试
+- build: 更改构建系统或依赖项
+- ci: 更改 CI 配置
+- chore: 其他不修改 src 或测试文件的更改
+- revert: 撤销之前的提交
+
+规则:
+1. 类型和小写使用小写
+2. 描述保持在 100 个字符以内
+3. 使用祈使语气（"添加"而不是"已添加"）
+4. 不要大写首字母
+5. 结尾不要句号
+6. 如果更改特定于模块/组件，请包含作用域
+7. 仅输出提交消息，不要输出其他内容。
+"#;
+
+const CONVENTIONAL_COMMIT_SYSTEM_PROMPT_JA: &str = r#"あなたはConventional Commits仕様に従ったコミットメッセージを生成するアシスタントです。
+
+提供されたgit diffを分析し、Conventional Commits仕様に従ったコミットメッセージを生成してください。
+
+形式: <type>[オプションのスコープ]: <説明>
+
+タイプ:
+- feat: 新機能
+- fix: バグ修正
+- docs: ドキュメントのみの変更
+- style: コードの意味に影響しない変更（フォーマット、セミコロンなど）
+- refactor: バグ修正や機能追加を伴わないコード変更
+- perf: パフォーマンスを向上させるコード変更
+- test: テストの追加または修正
+- build: ビルドシステムまたは依存関係の変更
+- ci: CI設定の変更
+- chore: srcやテストファイルを変更しないその他の変更
+- revert: 以前のコミットを取り消す
+
+ルール:
+1. タイプとスコープは小文字を使用
+2. 説明は100文字以内にする
+3. 命令形を使用する（"追加"ではなく"追加する"）
+4. 先頭を大文字にしない
+5. 最後にピリオドを付けない
+6. 変更がモジュール/コンポーネントに固有の場合はスコープを含める
+7. コミットメッセージのみを出力し、それ以外は出力しないでください。
+"#;
+
+const CONVENTIONAL_COMMIT_SYSTEM_PROMPT_KO: &str = r#"당신은 Conventional Commits 사양에 따른 커밋 메시지를 생성하는 도우미입니다.
+
+제공된 git diff를 분석하고 Conventional Commits 사양에 따른 커밋 메시지를 생성하세요.
+
+형식: <type>[선택적 범위]: <설명>
+
+유형:
+- feat: 새 기능
+- fix: 버그 수정
+- docs: 문서 변경만
+- style: 코드 의미에 영향을 주지 않는 변경(서식, 세미콜론 등)
+- refactor: 버그를 수정하거나 기능을 추가하지 않는 코드 변경
+- perf: 성능을 향상시키는 코드 변경
+- test: 테스트 추가 또는 수정
+- build: 빌드 시스템 또는 종속성 변경
+- ci: CI 구성 변경
+- chore: src 또는 테스트 파일을 수정하지 않는 기타 변경
+- revert: 이전 커밋 되돌리기
+
+규칙:
+1. 유형과 범위는 소문자 사용
+2. 설명은 100자 이내로 유지
+3. 명령형 사용("추가"가 아닌 "추가하다")
+4. 첫 글자 대문자화하지 않음
+5. 끝에 마침표 사용하지 않음
+6. 변경 사항이 모듈/구성 요소에 특정한 경우 범위 포함
+7. 커밋 메시지만 출력하고 다른 내용은 출력하지 마세요.
+"#;
+
+const CONVENTIONAL_COMMIT_SYSTEM_PROMPT_ES: &str = r#"Eres un asistente que genera mensajes de commit siguiendo la especificación Conventional Commits.
+
+Analiza el diff de git proporcionado y genera un mensaje de commit siguiendo la especificación Conventional Commits.
+
+Formato: <tipo>[alcance opcional]: <descripción>
+
+Tipos:
+- feat: Una nueva característica
+- fix: Una corrección de error
+- docs: Solo cambios en documentación
+- style: Cambios que no afectan el significado del código (formato, punto y coma, etc.)
+- refactor: Cambio de código que no corrige un error ni agrega una característica
+- perf: Cambio de código que mejora el rendimiento
+- test: Agregar o corregir pruebas
+- build: Cambios en el sistema de construcción o dependencias
+- ci: Cambios en la configuración de CI
+- chore: Otros cambios que no modifican archivos src o de prueba
+- revert: Revierte un commit anterior
+
+Reglas:
+1. Usa minúsculas para tipo y alcance
+2. Mantén la descripción bajo 100 caracteres
+3. Usa modo imperativo ("agregar" no "agregado")
+4. No capitalices la primera letra
+5. Sin punto al final
+6. Incluye alcance si el cambio es específico de un módulo/componente
+7. Genera SOLO el mensaje de commit, nada más.
+"#;
+
+const CONVENTIONAL_COMMIT_SYSTEM_PROMPT_FR: &str = r#"Vous êtes un assistant qui génère des messages de commit suivant la spécification Conventional Commits.
+
+Analysez le diff git fourni et générez un message de commit suivant la spécification Conventional Commits.
+
+Format: <type>[portée optionnelle]: <description>
+
+Types:
+- feat: Une nouvelle fonctionnalité
+- fix: Une correction de bug
+- docs: Changements de documentation uniquement
+- style: Changements qui n'affectent pas la signification du code (formatage, points-virgules, etc.)
+- refactor: Changement de code qui ne corrige pas un bug ni n'ajoute une fonctionnalité
+- perf: Changement de code qui améliore les performances
+- test: Ajout ou correction de tests
+- build: Changements du système de build ou des dépendances
+- ci: Changements de la configuration CI
+- chore: Autres changements qui ne modifient pas les fichiers src ou de test
+- revert: Révertit un commit précédent
+
+Règles:
+1. Utilisez des minuscules pour le type et la portée
+2. Gardez la description sous 100 caractères
+3. Utilisez le mode impératif ("ajouter" non "ajouté")
+4. Ne capitalisez pas la première lettre
+5. Pas de point à la fin
+6. Incluez la portée si le changement est spécifique à un module/composant
+7. Générez SEULEMENT le message de commit, rien d'autre.
+"#;
+
+const CONVENTIONAL_COMMIT_SYSTEM_PROMPT_DE: &str = r#"Sie sind ein Assistent, der Commit-Nachrichten gemäß der Conventional Commits-Spezifikation generiert.
+
+Analysieren Sie den bereitgestellten git diff und generieren Sie eine Commit-Nachricht gemäß der Conventional Commits-Spezifikation.
+
+Format: <typ>[optionaler Bereich]: <beschreibung>
+
+Typen:
+- feat: Eine neue Funktion
+- fix: Ein Bugfix
+- docs: Nur Dokumentationsänderungen
+- style: Änderungen, die die Code-Bedeutung nicht beeinflussen (Formatierung, Semikolons usw.)
+- refactor: Code-Änderung, die weder einen Bug behebt noch eine Funktion hinzufügt
+- perf: Code-Änderung, die die Leistung verbessert
+- test: Hinzufügen oder Korrigieren von Tests
+- build: Änderungen am Build-System oder Abhängigkeiten
+- ci: Änderungen an der CI-Konfiguration
+- chore: Andere Änderungen, die src- oder Testdateien nicht ändern
+- revert: Setzt einen vorherigen Commit zurück
+
+Regeln:
+1. Verwenden Sie Kleinbuchstaben für Typ und Bereich
+2. Halten Sie die Beschreibung unter 100 Zeichen
+3. Verwenden Sie den Imperativ ("hinzufügen" nicht "hinzugefügt")
+4. Großschreiben Sie den ersten Buchstaben nicht
+5. Kein Punkt am Ende
+6. Fügen Sie einen Bereich ein, wenn die Änderung spezifisch für ein Modul/Komponente ist
+7. Geben Sie NUR die Commit-Nachricht aus, nichts anderes.
+"#;
+
+const COMMITLINT_SYSTEM_PROMPT: &str = r#"You are a helpful assistant that generates commit messages following @commitlint/config-conventional.
+
+Analyze the git diff and generate a commit message.
+
+Format: <type>[optional scope]: <subject>
+
+Types: feat, fix, docs, style, refactor, perf, test, build, ci, chore, revert
+
+Rules:
+1. Subject should not start with uppercase
+2. Subject should not end with period
+3. Subject should be 4-100 characters
+4. Use imperative mood
+5. Be concise but descriptive
+6. Output ONLY the commit message, nothing else.
+"#;
+
+const COMMITLINT_SYSTEM_PROMPT_ZH: &str = r#"你是一个生成符合 @commitlint/config-conventional 规范的提交消息的助手。
+
+分析 git diff 并生成提交消息。
+
+格式: <type>[可选作用域]: <主题>
+
+类型: feat, fix, docs, style, refactor, perf, test, build, ci, chore, revert
+
+规则:
+1. 主题不应以大写字母开头
+2. 主题不应以句号结尾
+3. 主题应为 4-100 个字符
+4. 使用祈使语气
+5. 简洁但描述性强
+6. 仅输出提交消息，不要输出其他额外内容。
+"#;
+
+const COMMITLINT_SYSTEM_PROMPT_JA: &str = r#"あなたは@commitlint/config-conventionalに従ったコミットメッセージを生成するアシスタントです。
+
+git diffを分析し、コミットメッセージを生成してください。
+
+形式: <type>[オプションのスコープ]: <件名>
+
+タイプ: feat, fix, docs, style, refactor, perf, test, build, ci, chore, revert
+
+ルール:
+1. 件名は大文字で始めないでください
+2. 件名はピリオドで終わらないでください
+3. 件名は4-100文字である必要があります
+4. 命令形を使用してください
+5. 簡潔ですが説明的であること
+6. コミットメッセージのみを出力し、それ以外は出力しないでください。
+"#;
+
+const COMMITLINT_SYSTEM_PROMPT_KO: &str = r#"당신은 @commitlint/config-conventional에 따른 커밋 메시지를 생성하는 도우미입니다.
+
+git diff를 분석하고 커밋 메시지를 생성하세요.
+
+형식: <type>[선택적 범위]: <제목>
+
+유형: feat, fix, docs, style, refactor, perf, test, build, ci, chore, revert
+
+규칙:
+1. 제목은 대문자로 시작하지 않아야 합니다
+2. 제목은 마침표로 끝나지 않아야 합니다
+3. 제목은 4-100자여야 합니다
+4. 명령형을 사용하세요
+5. 간결하지만 설명적이어야 합니다
+6. 커밋 메시지만 출력하고 다른 내용은 출력하지 마세요.
+"#;
+
+const COMMITLINT_SYSTEM_PROMPT_ES: &str = r#"Eres un asistente que genera mensajes de commit siguiendo @commitlint/config-conventional.
+
+Analiza el diff de git y genera un mensaje de commit.
+
+Formato: <tipo>[alcance opcional]: <asunto>
+
+Tipos: feat, fix, docs, style, refactor, perf, test, build, ci, chore, revert
+
+Reglas:
+1. El asunto no debe comenzar con mayúscula
+2. El asunto no debe terminar con punto
+3. El asunto debe tener 4-100 caracteres
+4. Usa modo imperativo
+5. Sé conciso pero descriptivo
+6. Genera SOLO el mensaje de commit, nada más.
+"#;
+
+const COMMITLINT_SYSTEM_PROMPT_FR: &str = r#"Vous êtes un assistant qui génère des messages de commit suivant @commitlint/config-conventional.
+
+Analysez le diff git et générez un message de commit.
+
+Format: <type>[portée optionnelle]: <sujet>
+
+Types: feat, fix, docs, style, refactor, perf, test, build, ci, chore, revert
+
+Règles:
+1. Le sujet ne doit pas commencer par une majuscule
+2. Le sujet ne doit pas se terminer par un point
+3. Le sujet doit avoir 4-100 caractères
+4. Utilisez le mode impératif
+5. Soyez concis mais descriptif
+6. Générez SEULEMENT le message de commit, rien d'autre.
+"#;
+
+const COMMITLINT_SYSTEM_PROMPT_DE: &str = r#"Sie sind ein Assistent, der Commit-Nachrichten gemäß @commitlint/config-conventional generiert.
+
+Analysieren Sie den git diff und generieren Sie eine Commit-Nachricht.
+
+Format: <typ>[optionaler Bereich]: <betreff>
+
+Typen: feat, fix, docs, style, refactor, perf, test, build, ci, chore, revert
+
+Regeln:
+1. Der Betreff sollte nicht mit einem Großbuchstaben beginnen
+2. Der Betreff sollte nicht mit einem Punkt enden
+3. Der Betreff sollte 4-100 Zeichen haben
+4. Verwenden Sie den Imperativ
+5. Seien Sie prägnant aber beschreibend
+6. Geben Sie NUR die Commit-Nachricht aus, nichts anderes.
+"#;
+
+const TAG_MESSAGE_SYSTEM_PROMPT: &str = r#"You are a helpful assistant that generates git tag annotation messages.
+
+Given a version number and a list of commits, generate a concise but informative tag message.
+
+The message should:
+1. Start with a brief summary of the release
+2. Group changes by type (features, fixes, etc.)
+3. Be suitable for a git annotated tag
+
+Format:
+<version> Release
+
+Summary of changes...
+
+Changes:
+- Feature: description
+- Fix: description
+...
+"#;
+
+const TAG_MESSAGE_SYSTEM_PROMPT_ZH: &str = r#"你是一个生成 git 标签注释消息的助手。
+
+给定版本号和提交列表，生成简洁但信息丰富的标签消息。
+
+消息应该：
+1. 以发布的简要摘要开始
+2. 按类型分组更改（功能、修复等）
+3. 适合 git 标注标签
+
+格式：
+<version> 发布
+
+更改摘要...
+
+更改：
+- 功能：描述
+- 修复：描述
+...
+"#;
+
+const TAG_MESSAGE_SYSTEM_PROMPT_JA: &str = r#"あなたはgitタグ注釈メッセージを生成するアシスタントです。
+
+バージョン番号とコミットのリストを考慮して、簡潔ですが情報豊富なタグメッセージを生成してください。
+
+メッセージは以下のようであるべきです：
+1. リリースの簡単な要約から始める
+2. タイプ別に変更をグループ化する（機能、修正など）
+3. git注釈タグに適している
+
+形式：
+<version> リリース
+
+変更の概要...
+
+変更：
+- 機能：説明
+- 修正：説明
+...
+"#;
+
+const TAG_MESSAGE_SYSTEM_PROMPT_KO: &str = r#"당신은 git 태그 주석 메시지를 생성하는 도우미입니다.
+
+버전 번호와 커밋 목록을 고려하여 간결하지만 정보가 풍부한 태그 메시지를 생성하세요.
+
+메시지는 다음과 같아야 합니다:
+1. 릴리스의 간단한 요약으로 시작
+2. 유형별로 변경 사항 그룹화(기능, 수정 등)
+3. git 주석 태그에 적합
+
+형식:
+<version> 릴리스
+
+변경 사항 요약...
+
+변경 사항:
+- 기능: 설명
+- 수정: 설명
+...
+"#;
+
+const TAG_MESSAGE_SYSTEM_PROMPT_ES: &str = r#"Eres un asistente que genera mensajes de anotación de etiquetas git.
+
+Dado un número de versión y una lista de commits, genera un mensaje de etiqueta conciso pero informativo.
+
+El mensaje debe:
+1. Comenzar con un resumen breve de la versión
+2. Agrupar cambios por tipo (características, correcciones, etc.)
+3. Ser adecuado para una etiqueta git anotada
+
+Formato:
+<version> Versión
+
+Resumen de cambios...
+
+Cambios:
+- Característica: descripción
+- Corrección: descripción
+...
+"#;
+
+const TAG_MESSAGE_SYSTEM_PROMPT_FR: &str = r#"Vous êtes un assistant qui génère des messages d'annotation de balises git.
+
+Étant donné un numéro de version et une liste de commits, générez un message de balise concis mais informatif.
+
+Le message doit :
+1. Commencer par un bref résumé de la version
+2. Grouper les changements par type (fonctionnalités, corrections, etc.)
+3. Être adapté à une balise git annotée
+
+Format :
+<version> Version
+
+Résumé des changements...
+
+Changements :
+- Fonctionnalité : description
+- Correction : description
+...
+"#;
+
+const TAG_MESSAGE_SYSTEM_PROMPT_DE: &str = r#"Sie sind ein Assistent, der git-Tag-Anmerkungsnachrichten generiert.
+
+Gegeben eine Versionsnummer und eine Liste von Commits, generieren Sie eine prägnante aber informative Tag-Nachricht.
+
+Die Nachricht sollte:
+1. Mit einer kurzen Zusammenfassung der Version beginnen
+2. Änderungen nach Typ gruppieren (Funktionen, Fixes, etc.)
+3. Für ein git-annotiertes Tag geeignet sein
+
+Format:
+<version> Version
+
+Zusammenfassung der Änderungen...
+
+Änderungen:
+- Funktion: Beschreibung
+- Fix: Beschreibung
+...
+"#;
+
+const CHANGELOG_SYSTEM_PROMPT: &str = r#"You are a helpful assistant that generates changelog entries.
+
+Given a version and a list of commits, generate a well-formatted changelog section.
+
+Group commits by:
+- Features (feat)
+- Bug Fixes (fix)
+- Documentation (docs)
+- Other Changes
+
+Format in markdown with proper headings and bullet points.
+"#;
+
+const CHANGELOG_SYSTEM_PROMPT_ZH: &str = r#"你是一个生成变更日志条目的助手。
+
+给定版本和提交列表，生成格式良好的变更日志部分。
+
+按以下方式分组提交：
+- 功能 (feat)
+- 错误修复 (fix)
+- 文档 (docs)
+- 其他更改
+
+使用适当的标题和项目符号以 markdown 格式输出。
+"#;
+
+const CHANGELOG_SYSTEM_PROMPT_JA: &str = r#"あなたは変更ログエントリを生成するアシスタントです。
+
+バージョンとコミットのリストを考慮して、適切にフォーマットされた変更ログセクションを生成してください。
+
+コミットを以下でグループ化してください：
+- 機能 (feat)
+- バグ修正 (fix)
+- ドキュメント (docs)
+- その他の変更
+
+適切な見出しと箇条書きを使用してmarkdown形式でフォーマットしてください。
+"#;
+
+const CHANGELOG_SYSTEM_PROMPT_KO: &str = r#"당신은 변경 로그 항목을 생성하는 도우미입니다.
+
+버전과 커밋 목록을 고려하여 잘 포맷된 변경 로그 섹션을 생성하세요.
+
+다음으로 커밋을 그룹화하세요:
+- 기능 (feat)
+- 버그 수정 (fix)
+- 문서 (docs)
+- 기타 변경 사항
+
+적절한 제목과 글머리 기호를 사용하여 markdown 형식으로 포맷하세요.
+"#;
+
+const CHANGELOG_SYSTEM_PROMPT_ES: &str = r#"Eres un asistente que genera entradas de registro de cambios.
+
+Dada una versión y una lista de commits, genera una sección de registro de cambios bien formateada.
+
+Agrupa los commits por:
+- Características (feat)
+- Correcciones de errores (fix)
+- Documentación (docs)
+- Otros cambios
+
+Formatea en markdown con encabezados y viñetas apropiados.
+"#;
+
+const CHANGELOG_SYSTEM_PROMPT_FR: &str = r#"Vous êtes un assistant qui génère des entrées de journal des modifications.
+
+Étant donné une version et une liste de commits, générez une section de journal des modifications bien formatée.
+
+Groupez les commits par :
+- Fonctionnalités (feat)
+- Corrections de bugs (fix)
+- Documentation (docs)
+- Autres modifications
+
+Formatez en markdown avec des en-têtes et des puces appropriés.
+"#;
+
+const CHANGELOG_SYSTEM_PROMPT_DE: &str = r#"Sie sind ein Assistent, der Changelog-Einträge generiert.
+
+Gegeben eine Version und eine Liste von Commits, generieren Sie einen gut formatierten Changelog-Abschnitt.
+
+Gruppieren Sie Commits nach:
+- Funktionen (feat)
+- Bugfixes (fix)
+- Dokumentation (docs)
+- Andere Änderungen
+
+Formatieren Sie in Markdown mit geeigneten Überschriften und Aufzählungspunkten.
+"#;

src/llm/thinking.rs

@@ -64,18 +64,10 @@ impl Default for ThinkingStateManager {
 /// 线程安全的思考状态管理器引用
 pub type SharedThinkingState = Arc<ThinkingStateManager>;
 
-/// 创建带有默认控制台输出的思考状态管理器
-/// 在思考开始时打印 "thinking..."，在思考结束时清除该标识
+/// 创建 LLM 流式使用的共享思考状态。
+/// 进度显示由 AI 生成 spinner 负责（issue 21），此处不再附加控制台输出。
 pub fn create_console_thinking_state() -> SharedThinkingState {
-    Arc::new(
-        ThinkingStateManager::new()
-            .on_thinking_start(|| {
-                eprint!("\rthinking...");
-            })
-            .on_thinking_end(|| {
-                eprint!("\r          \r");
-            }),
-    )
+    Arc::new(ThinkingStateManager::new())
 }
 
 #[cfg(test)]

src/main.rs

@@ -1,5 +1,3 @@
-#![allow(dead_code)]
-
 use anyhow::Result;
 use clap::{Parser, Subcommand};
 use std::path::PathBuf;
@@ -22,7 +20,7 @@ use quicommit::commands::{
 #[command(propagate_version = true)]
 #[command(arg_required_else_help = true)]
 struct Cli {
-    /// Enable verbose output
+    /// Increase verbosity (-v: info, -vv: debug, -vvv: trace)
     #[arg(short, long, global = true, action = clap::ArgAction::Count)]
     verbose: u8,
 
@@ -31,9 +29,17 @@ struct Cli {
     config: Option<String>,
 
     /// Disable colored output
-    #[arg(long, global = true, env = "NO_COLOR")]
+    #[arg(long, global = true)]
     no_color: bool,
 
+    /// Force emoji decorations on (overrides config)
+    #[arg(long, global = true, conflicts_with = "no_emoji")]
+    emoji: bool,
+
+    /// Disable emoji/symbol decorations
+    #[arg(long, global = true)]
+    no_emoji: bool,
+
     #[command(subcommand)]
     command: Commands,
 }
@@ -73,6 +79,31 @@ enum Commands {
 async fn main() -> Result<()> {
     let cli = Cli::parse();
 
+    // Apply the global color decision before any output is produced.
+    // --no-color disables colors; NO_COLOR follows the no-color.org spec
+    // (any presence, regardless of value, disables color).
+    let no_color = cli.no_color || std::env::var_os("NO_COLOR").is_some();
+    colored::control::set_override(!no_color);
+
+    // Resolve the decoration switch (issue 14): explicit flag > config >
+    // default; --no-color disables decorations too (ADR-0003).
+    let emoji_from_config = cli
+        .config
+        .as_deref()
+        .map(PathBuf::from)
+        .or_else(|| quicommit::config::AppConfig::default_path().ok())
+        .and_then(|path| quicommit::config::AppConfig::load(&path).ok())
+        .map(|config| config.output.emoji)
+        .unwrap_or(true);
+    let emoji_enabled = if no_color || cli.no_emoji {
+        false
+    } else if cli.emoji {
+        true
+    } else {
+        emoji_from_config
+    };
+    quicommit::utils::set_emoji_enabled(emoji_enabled);
+
     let log_level = match cli.verbose {
         0 => "warn",
         1 => "info",
@@ -80,8 +111,11 @@ async fn main() -> Result<()> {
         _ => "trace",
     };
 
+    // RUST_LOG takes precedence when set; otherwise -v decides (issue 23).
+    let filter = tracing_subscriber::EnvFilter::try_from_default_env()
+        .unwrap_or_else(|_| tracing_subscriber::EnvFilter::new(log_level));
     tracing_subscriber::fmt()
-        .with_env_filter(log_level)
+        .with_env_filter(filter)
         .with_target(false)
         .init();
 

src/utils/keyring.rs

@@ -232,9 +232,7 @@ impl KeyringManager {
                     "Keyring is available".to_string()
                 }
             }
-            KeyringStatus::Unavailable => {
-                "Keyring is not available. Set QUICOMMIT_API_KEY environment variable.".to_string()
-            }
+            KeyringStatus::Unavailable => "Keyring is not available on this system.".to_string(),
         }
     }
 }

src/utils/mod.rs

@@ -6,48 +6,144 @@ pub mod validators;
 
 use anyhow::{Context, Result};
 use colored::Colorize;
-use std::io::{self, Write};
+use std::io;
+use std::sync::OnceLock;
+use std::time::Duration;
+
+/// Process-global decoration switch (issue 14).
+static EMOJI_ENABLED: OnceLock<bool> = OnceLock::new();
+
+/// Set whether emoji/symbol decorations are shown. Called once from `main`
+/// after resolving config + flags.
+pub fn set_emoji_enabled(enabled: bool) {
+    let _ = EMOJI_ENABLED.set(enabled);
+}
+
+/// Whether emoji/symbol decorations are shown.
+pub fn emoji_enabled() -> bool {
+    *EMOJI_ENABLED.get_or_init(|| true)
+}
+
+/// Styled decoration prefix for status lines; empty when decorations are off
+/// (ADR-0003: message text never embeds decorations).
+pub fn success_prefix() -> colored::ColoredString {
+    if emoji_enabled() {
+        "✓".green().bold()
+    } else {
+        "".normal()
+    }
+}
+
+pub fn error_prefix() -> colored::ColoredString {
+    if emoji_enabled() {
+        "✗".red().bold()
+    } else {
+        "".normal()
+    }
+}
+
+pub fn warning_prefix() -> colored::ColoredString {
+    if emoji_enabled() {
+        "⚠".yellow().bold()
+    } else {
+        "".normal()
+    }
+}
+
+pub fn info_prefix() -> colored::ColoredString {
+    if emoji_enabled() {
+        "ℹ".blue().bold()
+    } else {
+        "".normal()
+    }
+}
+
+pub fn progress_prefix() -> colored::ColoredString {
+    if emoji_enabled() {
+        "→".cyan()
+    } else {
+        "".normal()
+    }
+}
+
+fn print_decorated(prefix: colored::ColoredString, msg: &str, to_stderr: bool) {
+    if emoji_enabled() {
+        if to_stderr {
+            eprintln!("{} {}", prefix, msg);
+        } else {
+            println!("{} {}", prefix, msg);
+        }
+    } else if to_stderr {
+        eprintln!("{}", msg);
+    } else {
+        println!("{}", msg);
+    }
+}
 
 /// Print success message
 pub fn print_success(msg: &str) {
-    println!("{} {}", "✓".green().bold(), msg);
+    print_decorated(success_prefix(), msg, false);
 }
 
-/// Print error message
+/// Print error message to stderr
 pub fn print_error(msg: &str) {
-    eprintln!("{} {}", "✗".red().bold(), msg);
+    print_decorated(error_prefix(), msg, true);
 }
 
 /// Print warning message
 pub fn print_warning(msg: &str) {
-    println!("{} {}", "⚠".yellow().bold(), msg);
+    print_decorated(warning_prefix(), msg, false);
+}
+
+/// Print warning message to stderr (stdout must stay clean, e.g. exports)
+pub fn eprint_warning(msg: &str) {
+    print_decorated(warning_prefix(), msg, true);
 }
 
 /// Print info message
 pub fn print_info(msg: &str) {
-    println!("{} {}", "ℹ".blue().bold(), msg);
+    print_decorated(info_prefix(), msg, false);
 }
 
-/// Confirm action with user
-pub fn confirm(prompt: &str) -> Result<bool> {
-    print!("{} [y/N] ", prompt);
-    io::stdout().flush()?;
-
-    let mut input = String::new();
-    io::stdin().read_line(&mut input)?;
-
-    Ok(input.trim().to_lowercase().starts_with('y'))
+/// Print progress/status message (→ prefix when decorations are on)
+pub fn print_progress(msg: &str) {
+    print_decorated(progress_prefix(), msg, false);
 }
 
-/// Get user input
-pub fn input(prompt: &str) -> Result<String> {
-    print!("{}: ", prompt);
-    io::stdout().flush()?;
+/// Progress spinner for long-running operations (issue 21).
+/// Degrades to a static line when stdout is not a terminal or when
+/// decorations are disabled.
+pub struct Spinner {
+    bar: Option<indicatif::ProgressBar>,
+}
 
-    let mut input = String::new();
-    io::stdin().read_line(&mut input)?;
+impl Spinner {
+    pub fn start(msg: &str) -> Self {
+        if emoji_enabled() && std::io::IsTerminal::is_terminal(&io::stdout()) {
+            let bar = indicatif::ProgressBar::new_spinner();
+            bar.set_message(msg.to_string());
+            bar.enable_steady_tick(Duration::from_millis(100));
+            Self { bar: Some(bar) }
+        } else {
+            println!("{}", msg);
+            Self { bar: None }
+        }
+    }
 
-    Ok(input.trim().to_string())
+    /// Stop the spinner, erasing it from the terminal (no completion line).
+    pub fn finish_clear(&self) {
+        if let Some(bar) = &self.bar {
+            bar.finish_and_clear();
+        }
+    }
+
+    /// Stop the spinner and print a completion line.
+    pub fn finish_with(&self, msg: &str) {
+        if let Some(bar) = &self.bar {
+            bar.finish_and_clear();
+        }
+        println!("{}", msg);
+    }
 }
 
 /// Get password input (hidden)
@@ -59,19 +155,3 @@ pub fn password_input(prompt: &str) -> Result<String> {
         .interact()
         .context("Failed to read password")
 }
-
-/// Check if running in a terminal
-pub fn is_terminal() -> bool {
-    atty::is(atty::Stream::Stdout)
-}
-
-/// Format duration in human-readable format
-pub fn format_duration(secs: u64) -> String {
-    if secs < 60 {
-        format!("{}s", secs)
-    } else if secs < 3600 {
-        format!("{}m {}s", secs / 60, secs % 60)
-    } else {
-        format!("{}h {}m", secs / 3600, (secs % 3600) / 60)
-    }
-}

src/utils/validators.rs

@@ -1,6 +1,6 @@
 use anyhow::{Result, bail};
-use lazy_static::lazy_static;
 use regex::Regex;
+use std::sync::LazyLock;
 
 /// Conventional commit types
 pub const CONVENTIONAL_TYPES: &[&str] = &[
@@ -37,32 +37,33 @@ pub const COMMITLINT_TYPES: &[&str] = &[
     "security", // Security-related changes
 ];
 
-lazy_static! {
-    /// Regex for conventional commit format
-    static ref CONVENTIONAL_COMMIT_REGEX: Regex = Regex::new(
-        r"^(?P<type>feat|fix|docs|style|refactor|perf|test|build|ci|chore|revert)(?:\((?P<scope>[^)]+)\))?(?P<breaking>!)?: (?P<description>.+)$"
-    ).unwrap();
+/// Regex for conventional commit format
+static CONVENTIONAL_COMMIT_REGEX: LazyLock<Regex> = LazyLock::new(|| {
+    Regex::new(
+        r"^(?P<type>feat|fix|docs|style|refactor|perf|test|build|ci|chore|revert)(?:\((?P<scope>[^)]+)\))?(?P<breaking>!)?: (?P<description>.+)$",
+    )
+    .unwrap()
+});
 
-    /// Regex for scope validation
-    static ref SCOPE_REGEX: Regex = Regex::new(
-        r"^[a-z0-9-]+$"
-    ).unwrap();
+/// Regex for scope validation
+static SCOPE_REGEX: LazyLock<Regex> = LazyLock::new(|| Regex::new(r"^[a-z0-9-]+$").unwrap());
 
-    /// Regex for version tag validation (semver)
-    static ref SEMVER_REGEX: Regex = Regex::new(
-        r"^v?(?P<major>0|[1-9]\d*)\.(?P<minor>0|[1-9]\d*)\.(?P<patch>0|[1-9]\d*)(?:-(?P<prerelease>(?:0|[1-9]\d*|\d*[a-zA-Z-][0-9a-zA-Z-]*)(?:\.(?:0|[1-9]\d*|\d*[a-zA-Z-][0-9a-zA-Z-]*))*))?(?:\+(?P<buildmetadata>[0-9a-zA-Z-]+(?:\.[0-9a-zA-Z-]+)*))?$"
-    ).unwrap();
+/// Regex for version tag validation (semver)
+static SEMVER_REGEX: LazyLock<Regex> = LazyLock::new(|| {
+    Regex::new(
+        r"^v?(?P<major>0|[1-9]\d*)\.(?P<minor>0|[1-9]\d*)\.(?P<patch>0|[1-9]\d*)(?:-(?P<prerelease>(?:0|[1-9]\d*|\d*[a-zA-Z-][0-9a-zA-Z-]*)(?:\.(?:0|[1-9]\d*|\d*[a-zA-Z-][0-9a-zA-Z-]*))*))?(?:\+(?P<buildmetadata>[0-9a-zA-Z-]+(?:\.[0-9a-zA-Z-]+)*))?$",
+    )
+    .unwrap()
+});
 
-    /// Regex for email validation
-    static ref EMAIL_REGEX: Regex = Regex::new(
-        r"^[a-zA-Z0-9._%+-]+@[a-zA-Z0-9.-]+\.[a-zA-Z]{2,}$"
-    ).unwrap();
+/// Regex for email validation
+static EMAIL_REGEX: LazyLock<Regex> = LazyLock::new(|| {
+    Regex::new(r"^[a-zA-Z0-9._%+-]+@[a-zA-Z0-9.-]+\.[a-zA-Z]{2,}$").unwrap()
+});
 
-    /// Regex for GPG key ID validation
-    static ref GPG_KEY_ID_REGEX: Regex = Regex::new(
-        r"^[A-F0-9]{16,40}$"
-    ).unwrap();
-}
+/// Regex for GPG key ID validation
+static GPG_KEY_ID_REGEX: LazyLock<Regex> =
+    LazyLock::new(|| Regex::new(r"^[A-F0-9]{16,40}$").unwrap());
 
 /// Validate conventional commit message
 pub fn validate_conventional_commit(message: &str) -> Result<()> {

tests/gitignore_tests.rs

@@ -115,7 +115,8 @@ fn test_stage_all_removes_ignored_tracked_files() {
     write_file(repo_path, "__pycache__/foo.pyc", "modified bytecode");
 
     let repo = GitRepo::open(repo_path).expect("Failed to open repo");
-    let removed = repo.stage_all().expect("stage_all should succeed");
+    let messages = quicommit::i18n::Messages::new(quicommit::config::Language::English);
+    let removed = repo.stage_all(&messages).expect("stage_all should succeed");
 
     assert!(
         removed.iter().any(|f| f == "__pycache__/foo.pyc"),
@@ -142,7 +143,8 @@ fn test_stage_all_no_ignored_files_returns_empty() {
     write_file(repo_path, "src/main.rs", "fn main() {}");
 
     let repo = GitRepo::open(repo_path).expect("Failed to open repo");
-    let removed = repo.stage_all().expect("stage_all should succeed");
+    let messages = quicommit::i18n::Messages::new(quicommit::config::Language::English);
+    let removed = repo.stage_all(&messages).expect("stage_all should succeed");
 
     assert!(
         removed.is_empty(),

tests/integration_tests.rs

@@ -258,6 +258,190 @@ mod config_command {
             .success()
             .stdout(predicate::str::contains("config.toml"));
     }
+
+    #[test]
+    fn test_no_color_disables_ansi() {
+        let temp_dir = TempDir::new().unwrap();
+        let config_path = temp_dir.path().join("config.toml");
+
+        let mut cmd = cargo_bin_cmd!("quicommit");
+        cmd.args(&["init", "--yes", "--config", config_path.to_str().unwrap()]);
+        cmd.assert().success();
+
+        let mut cmd = cargo_bin_cmd!("quicommit");
+        cmd.args(&[
+            "config",
+            "show",
+            "--no-color",
+            "--config",
+            config_path.to_str().unwrap(),
+        ]);
+
+        let output = cmd.output().unwrap();
+        assert!(output.status.success());
+        let stdout = String::from_utf8_lossy(&output.stdout);
+        assert!(
+            !stdout.contains("\x1b["),
+            "ANSI color codes present despite --no-color"
+        );
+    }
+
+    #[test]
+    fn test_emoji_switch_precedence() {
+        let temp_dir = TempDir::new().unwrap();
+        let config_path = temp_dir.path().join("config.toml");
+
+        let mut cmd = cargo_bin_cmd!("quicommit");
+        cmd.env_remove("NO_COLOR");
+        cmd.args(&["init", "--yes", "--config", config_path.to_str().unwrap()]);
+        cmd.assert().success();
+
+        let set_format = |extra: &[&str], cfg: &std::path::Path| {
+            let mut c = cargo_bin_cmd!("quicommit");
+            c.env_remove("NO_COLOR");
+            c.args(&["config", "set", "commit.format", "conventional"])
+                .args(extra)
+                .args(&["--config", cfg.to_str().unwrap()]);
+            c.output().unwrap()
+        };
+
+        // Default (config true): decorations shown
+        let out = set_format(&[], &config_path);
+        assert!(out.status.success());
+        assert!(String::from_utf8_lossy(&out.stdout).contains("✓"));
+
+        // Config output.emoji=false: no decorations
+        let mut cmd = cargo_bin_cmd!("quicommit");
+        cmd.env_remove("NO_COLOR");
+        cmd.args(&[
+            "config",
+            "set",
+            "output.emoji",
+            "false",
+            "--config",
+            config_path.to_str().unwrap(),
+        ]);
+        cmd.assert().success();
+
+        let out = set_format(&[], &config_path);
+        assert!(out.status.success());
+        assert!(!String::from_utf8_lossy(&out.stdout).contains("✓"));
+
+        // Explicit --emoji overrides config false
+        let out = set_format(&["--emoji"], &config_path);
+        assert!(out.status.success());
+        assert!(String::from_utf8_lossy(&out.stdout).contains("✓"));
+
+        // Config back to true, then --no-emoji overrides
+        let mut cmd = cargo_bin_cmd!("quicommit");
+        cmd.env_remove("NO_COLOR");
+        cmd.args(&[
+            "config",
+            "set",
+            "output.emoji",
+            "true",
+            "--config",
+            config_path.to_str().unwrap(),
+        ]);
+        cmd.assert().success();
+
+        let out = set_format(&["--no-emoji"], &config_path);
+        assert!(out.status.success());
+        assert!(!String::from_utf8_lossy(&out.stdout).contains("✓"));
+    }
+
+    #[test]
+    fn test_chinese_output_smoke() {
+        let temp_dir = TempDir::new().unwrap();
+        let repo_path = temp_dir.path().to_path_buf();
+        setup_test_repo_with_file(&repo_path, "test.txt", "content");
+
+        let config_path = repo_path.join("config.toml");
+        init_quicommit(&repo_path, &config_path);
+
+        let mut cmd = cargo_bin_cmd!("quicommit");
+        cmd.env_remove("NO_COLOR");
+        cmd.args(&[
+            "config",
+            "set",
+            "language.output_language",
+            "zh",
+            "--config",
+            config_path.to_str().unwrap(),
+        ]);
+        cmd.assert().success();
+
+        let mut cmd = cargo_bin_cmd!("quicommit");
+        cmd.env_remove("NO_COLOR");
+        cmd.args(&[
+            "commit",
+            "--manual",
+            "-m",
+            "feat: 中文冒烟",
+            "--dry-run",
+            "--yes",
+            "--config",
+            config_path.to_str().unwrap(),
+        ])
+        .current_dir(&repo_path);
+
+        cmd.assert()
+            .success()
+            .stdout(predicate::str::contains("试运行"));
+    }
+
+    #[test]
+    fn test_rust_log_env() {
+        let temp_dir = TempDir::new().unwrap();
+        let config_path = temp_dir.path().join("config.toml");
+
+        let mut cmd = cargo_bin_cmd!("quicommit");
+        cmd.env_remove("NO_COLOR");
+        cmd.args(&["init", "--yes", "--config", config_path.to_str().unwrap()]);
+        cmd.assert().success();
+
+        let mut cmd = cargo_bin_cmd!("quicommit");
+        cmd.env_remove("NO_COLOR");
+        cmd.env("RUST_LOG", "debug");
+        cmd.args(&["config", "path", "--config", config_path.to_str().unwrap()]);
+
+        cmd.assert()
+            .success()
+            .stdout(predicate::str::contains("Starting quicommit"));
+    }
+
+    #[test]
+    fn test_auto_generate_defaults_true() {
+        // Documents the unified --yes semantics (issue 05): with default
+        // config, --yes still AI-generates; deterministic output needs
+        // --no-generate (changelog) or -m (tag).
+        let temp_dir = TempDir::new().unwrap();
+        let config_path = temp_dir.path().join("config.toml");
+
+        let mut cmd = cargo_bin_cmd!("quicommit");
+        cmd.env_remove("NO_COLOR");
+        cmd.args(&["init", "--yes", "--config", config_path.to_str().unwrap()]);
+        cmd.assert().success();
+
+        for key in [
+            "commit.auto_generate",
+            "tag.auto_generate",
+            "changelog.auto_generate",
+        ] {
+            let mut cmd = cargo_bin_cmd!("quicommit");
+            cmd.env_remove("NO_COLOR");
+            cmd.args(&[
+                "config",
+                "get",
+                key,
+                "--config",
+                config_path.to_str().unwrap(),
+            ]);
+            cmd.assert()
+                .success()
+                .stdout(predicate::str::contains("true"));
+        }
+    }
 }
 
 mod commit_command {
@@ -438,6 +622,8 @@ mod tag_command {
             "tag",
             "--name",
             "v0.1.0",
+            "-m",
+            "Release v0.1.0",
             "--dry-run",
             "--yes",
             "--config",
@@ -469,6 +655,8 @@ mod tag_command {
             "--think",
             "--name",
             "v0.2.0",
+            "-m",
+            "Release v0.2.0",
             "--dry-run",
             "--yes",
             "--config",
@@ -528,6 +716,7 @@ mod changelog_command {
             "changelog",
             "--dry-run",
             "--yes",
+            "--no-generate",
             "--config",
             config_path.to_str().unwrap(),
         ])