菜单
Live debug, diagnose & fix on main. The ONE command allowed to edit code directly on main. Invoked as `$jc <bug description or service>`. Targeted hotfixes only.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
| name | jc |
| description | Live debug, diagnose & fix on main. The ONE command allowed to edit code directly on main. Invoked as `$jc <bug description or service>`. Targeted hotfixes only. |
Read .claude/commands/jc.md in full — it is your complete role manual. Follow it verbatim.
Argument: bug description, service name, or diagnostic request.
.claude/agents/gitter.md Phase 4 for the exact protocol.JC-COMMIT is LOCAL ONLY. Phase 4 = commit + commit-docs. Nothing else. Never run git push during /jc. If the founder wants the commit pushed, they invoke $git push separately. Stop after local commit; report hash; wait.
Cross-project build pipeline. Invoked as $wave-build <feature> or $wave-build with Pipeline/Task manifest/Wave args. Runs the full pipeline end to end including all git work.
Wave task runner. Reads a wave.md file and runs each pipeline sequentially. Invoked as `$wave <path/to/wave.md>`. Argument is the path to the wave manifest file.
Attributes Claude Code runtime token spend per sub-agent and per workflow run, parsed from local JSONL transcripts, and ranks the heaviest burners with estimated USD cost. Answers 'which agent burned the most tokens', 'what did this workflow/wave/pipeline cost', 'token breakdown', 'per-operation tokens', and any retrospective spend analysis. Triggered by 'token ledger', 'token attribution', 'heaviest token burner', 'which agent burned the most tokens', 'what did the run/workflow/wave cost', 'token breakdown', 'per-agent tokens', 'per-operation tokens'. Complements /pcm:context-meter: that audits STATIC context size, this covers RUNTIME spend attribution — route static-budget questions to context-meter, after-the-fact spend questions here.
Code hygiene audit — ghost fields, dead code, deps, arch, types, naming, quality. Domain-hydrated at install time.
Security deep scan — injection, auth, API, LLM/prompt, sensitive data, crypto, transport, supply-chain. Domain-hydrated at install time.
Exhaustive multi-angle analysis. Invoked by `360`, `three-sixty`, `do a 360`, "what could go wrong", or explicit blind-spot sweep requests.