By noni2she
Automated bug-fix workflow: extract → analyze → implement → test. Run /fix-one-issue <ISSUE-ID> to fix a single issue end-to-end.
批次修復多個 issue。每個 issue 在獨立 worktree 中平行執行,完成後交叉比對修改檔案,標記潛在 merge conflict。
End-to-end bug fix for a single issue. Runs extract → analyze → implement → test with judge gates between phases. One retry on phase failure.
初始化目前專案的 agent-fix 配置。自動偵測框架結構並生成 config.yaml,引導設定 Jira 憑證與測試帳號。不需要傳入路徑:直接在目標專案目錄開啟 Claude Code session,執行此指令即可。
Reproduces a bug in the browser (Step 0) then performs root cause analysis (Steps 1-5) on the codebase. Outputs analyze.md artifact with Root Cause File/Line, Fix Strategy (DIRECT or TACTICAL), and Suggested Fix.
Fetches one issue from the configured source and returns it as standard IssueData JSON (issue_id, summary, description, reproduction_steps, expected, actual, module, attachments).
End-to-end bug fix for a single issue. Runs extract → analyze → implement → test with judge gates. Returns structured RESULT line for batch coordination.
Reads analyze.md, creates a bugfix branch, implements the fix following the specified strategy (DIRECT or TACTICAL), then runs TypeScript and ESLint checks. Outputs implement.md artifact.
Issue Fix Team Orchestrator. Manages the Analyze → Implement → Test cycle for a single bug issue. Makes semantic gate judgments at each phase transition.
定位問題根源 (RCA) 與分析影響範圍。當收到 bug report 或 issue ID,需要分析問題原因、影響範圍、修復策略時使用此 skill。
根據分析報告實作修復。當已完成 bugfix-analyze 分析、有明確的 root cause 和修復策略後使用。
驗證修復結果的品質關卡。完成 bugfix-implement 後,進行靜態分析、策略合規性檢查與行為驗證。
識別外部 issue source 格式(Jira / GitHub Issues),萃取欄位並轉換為 pipeline 統一的標準欄位格式(issue_id / summary / module / description / reproduction_steps / expected / actual / attachments / comments)。
自動偵測目標專案結構,登記 context 文件,並生成 agent-fix config.yaml
Uses power tools
Uses Bash, Write, or Edit tools
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
Claude Code 插件 — 一個指令端對端修復 Bug:issue 擷取 → 重現 + 根源分析 → 實作 → 驗證。
harness_rules.yaml 集中管理 phase tool 限制與截斷規則agent-fix/
├── .claude-plugin/
│ └── plugin.json # Plugin 宣告(name / description / author)
├── commands/
│ └── fix-one-issue.md # 主指令:序列化 sub-agents + judge gate
├── agents/
│ ├── extract.md # Sub-agent: 從來源取 issue,回傳 IssueData JSON
│ ├── analyze.md # Sub-agent: 瀏覽器重現 + 根源分析 + 寫 analyze.md
│ ├── implement.md # Sub-agent: git branch + 實作修復 + 品質檢查
│ └── test.md # Sub-agent: 靜態分析 + 策略合規 + 行為驗證
├── mcp_servers/
│ └── agent_fix_tools/
│ └── server.py # 5-tool MCP server(stdio,FastMCP)
├── batch_runner/
│ └── driver.py # SDK driver:批次發 issue,worktree 隔離
├── engine/
│ ├── config.py # ProjectConfig(Pydantic YAML 驗證)
│ ├── harness_rules.yaml # Phase tool 限制 + 截斷規則(declarative)
│ ├── tools.py # Domain 工具(品質檢查 / 行為驗證 / 技術債)
│ ├── behavior_validation/ # Playwright runner(BehaviorValidator)
│ ├── issue_source/ # Issue 來源 adapter(local_json / Jira / Sheets)
│ └── skill_loader.py # SKILL.md 讀取(park,備 Phase 2/3 使用)
├── skills/ # 現有 SKILL.md(舊式 skill,可被 sub-agents 參考)
├── projects/ # 各目標專案配置
│ └── <slug>/
│ └── config.yaml # 專案配置
└── .mcp.json # MCP server 配置(chrome-devtools + agent-fix-tools)
/fix-one-issue <ISSUE-ID>
│
├─ [extract sub-agent]
│ fetch_issue(issue_id) → IssueData JSON
│ Gate 0: ❌ → CHECKPOINT
│
├─ [analyze sub-agent] ← 收到 IssueData JSON
│ Step 0: 瀏覽器重現(chrome-devtools MCP)
│ → Evidence Package(reproduce_confidence)
│ Gate 1a: confidence < 0.5 → RETRY once → CHECKPOINT
│ Steps 1–5: 讀原始碼 → 定位根因 → 評估影響 → 選策略
│ → 寫 issues/reports/<id>/analyze.md
│ Gate 1b: 缺 Root Cause File/Line → RETRY once → CHECKPOINT
│
├─ [implement sub-agent] ← 收到 analyze.md 路徑
│ git worktree branch → 實作修復(DIRECT / TACTICAL)
│ → run_typescript_check + run_eslint
│ → 寫 issues/reports/<id>/implement.md
│ Gate 2: TS FAILED → RETRY once → CHECKPOINT
│
└─ [test sub-agent] ← 收到 analyze.md + implement.md 路徑
靜態分析 + 策略合規 + 行為驗證(≤ 3 次 Playwright)
→ 寫 issues/reports/<id>/test.md
Gate 3: FAIL → RETRY once → CHECKPOINT
PASS / SKIPPED → 輸出 Fix Complete 摘要
| Phase | Sub-agent | 主要工具 | 輸出 |
|---|---|---|---|
| extract | extract | fetch_issue MCP | IssueData JSON |
| analyze | analyze | chrome-devtools MCP + Read/Grep | analyze.md |
| implement | implement | Edit/Write/Bash + run_typescript_check / run_eslint MCP | implement.md |
| test | test | run_behavior_validation / record_tech_debt MCP | test.md |
git clone https://github.com/noni2she/agent-fix.git
cd agent-fix
uv sync # 安裝 Python 依賴(含 mcp, playwright)
複製範本並填入目標專案資訊:
cp config-template.yaml projects/<slug>/config.yaml
# 編輯 config.yaml:project_root, issue_source, quality_checks, dev_server
# 必填
export PROJECT_CONFIG=./projects/<slug>/config.yaml
export ANTHROPIC_API_KEY=... # 批次 SDK driver 使用
# 行為驗證(選填,需要登入時)
export PROJ_TEST_USERNAME=...
export PROJ_TEST_PASSWORD=...
claude plugin install .
在 Claude Code 對話中:
/fix-one-issue PROJ-001
或使用批次 SDK driver:
python -m batch_runner.driver --issues PROJ-001,PROJ-002
python -m batch_runner.driver --issues-file issues.txt
python -m batch_runner.driver --source jira --jql "project = PROJ AND status = 'To Do'"
| 指令 | 說明 |
|---|---|
/fix-one-issue <ISSUE-ID> | 在 Claude Code 中端對端修復單一 issue |
python -m batch_runner.driver --issues A,B,C | 批次修復(sequential,worktree 隔離) |
python -m batch_runner.driver --batch-id <id> | 恢復中斷的批次 |
詳見 config-template.yaml 查看所有選項。
.mcp.json 宣告兩個 MCP server,Claude Code 自動管理生命週期:
npx claudepluginhub noni2she/agent-fixHarness-native ECC operator layer - 67 agents, 271 skills, 92 legacy command shims, reusable hooks, rules, selective install profiles, and production-ready workflows for Claude Code, Codex, OpenCode, Cursor, and related agent harnesses
Tools to maintain and improve CLAUDE.md files - audit quality, capture session learnings, and keep project memory current.
Comprehensive feature development workflow with specialized agents for codebase exploration, architecture design, and quality review
v9.44.1 — Patch release for Gemini environment/version detection and qwen auth gating. Run /octo:setup.
Superpowers Plus core skills library for Claude Code: planning, execution routing, TDD, debugging, and collaboration workflows
Browser automation and end-to-end testing MCP server by Microsoft. Enables Claude to interact with web pages, take screenshots, fill forms, click elements, and perform automated browser testing workflows.