菜单
Collaborative multi-agent fix workflow using codex, gemini-cli, and an independent subagent. Use when the user asks to "collab fix", "multi-agent fix", or wants several AI tools to jointly analyze and fix a bug or task with cross-review.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Revise an academic paper sentence by sentence with the author, stripping AI writing tells (stop-slop) while preserving the author's own voice. Use this whenever the user wants to polish, tighten, edit, proofread, or "de-slop" a paper, manuscript, draft, abstract, or thesis chapter — especially when they want careful interactive sentence-level revision rather than a one-shot rewrite. Trigger on requests like "help me revise my paper", "clean up my draft", "make my writing sound less AI", "go through my abstract line by line", or when they hand over a .tex/.md/.txt paper to improve.
Wield Google's Gemini CLI as a powerful auxiliary tool for code generation, review, analysis, and web research. Use when tasks benefit from a second AI perspective, current web information via Google Search, codebase architecture analysis, or parallel code generation. Also use when user explicitly requests Gemini operations.
Scan this machine for the Miasma / Phantom Gyp ("TeamPCP") npm supply-chain compromise that plants backdoors in Claude Code (~/.claude/settings.json) and VS Code (.vscode/tasks.json). Use when the user asks to check if their PC/machine is infected, compromised, or affected by the npm/Claude Code backdoor attack, credential-stealing worm, or Shai-Hulud campaign.
Remove AI writing patterns from prose. Use when drafting, editing, or reviewing text to eliminate predictable AI tells.
Use when the user asks to run Codex CLI (codex exec, codex resume) or references OpenAI Codex for code analysis, code review, or automated editing
Test-Driven Development debug workflow combined with multi-agent collab-fix. Use when the user asks to "TDD debug", "reproduce and fix with a test", or wants a bug fixed with a failing test written first and verified after the fix.
| name | collab-fix |
| description | Collaborative multi-agent fix workflow using codex, gemini-cli, and an independent subagent. Use when the user asks to "collab fix", "multi-agent fix", or wants several AI tools to jointly analyze and fix a bug or task with cross-review. |
| allowed-tools | ["Bash","Read","Edit","Write","Grep","Glob","Agent","AskUserQuestion"] |
Fix the user's task using codex, gemini-cli, and an independent subagent in coordinated rounds.
codex and gemini-cli skills must be available. If missing, report the error and stop.code-reviewer subagent if available; otherwise use the general-purpose agent.--sandbox read-only. For gemini-cli: do NOT pass --yolo or -s.timeout: 600000 (10 min) when calling Bash for codex/gemini commands.Propose plans in parallel. Ask codex, gemini-cli, and a subagent to analyze the problem and propose fix plans:
echo "Analyze <description>. Propose a fix plan with steps and tradeoffs." | codex exec --skip-git-repo-check --sandbox read-only - 2>/dev/nullgemini "Analyze <description>. Propose a fix plan with steps and tradeoffs." -o json 2>/dev/null | jq -r '.response'code-reviewer) to analyze independentlyCompare and choose. Summarize tradeoffs across the 3 plans. Ask the user only the necessary clarifying questions via AskUserQuestion to pick the best fix.
Implement. Ultrathink and implement the fix yourself. Do not git commit.
Cross-review the diff in parallel.
(echo "Review the following uncommitted diff."; git diff) | codex exec --skip-git-repo-check --sandbox read-only - 2>/dev/null(echo "Review the following uncommitted diff."; git diff) | gemini -o json 2>/dev/null | jq -r '.response'Resolve feedback. If any item depends on human preference, ask the user via AskUserQuestion.
Iterate. Repeat steps 3–5 until all three reviewers are satisfied, or 5 rounds are reached. If no consensus after 5 rounds, report the root cause and what remains disputed.