菜单
Four-phase loop for when I'm the failing agent — looping, burning tokens, drifting from objective, retrying without progress. Capture → Diagnose → Contained Recovery → Introspection Report. Use when stuck. NOT for user code bugs (that's /diagnose).
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
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。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/pnils08/GodWorld --skill self-debug命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
Close a GodWorld external-lane session (Grok) cleanly: inventory work, land authorized commits if needed, rewrite only NEXT[grok] in SESSION_CONTEXT.md, commit that handoff path-specifically, report git stack / push status. Use when the user says session end, close session, handoff, wrap up, done for now, or runs /session-end. Not the Claude Code session-end skill (no PIN bump, no ROLLOUT sweep, no sessionEndMechanical.js).
Deputy Mayor Brenda Okoro's office. Generates operational statements on West Oakland Stabilization Fund, Community Development, and Economic Development (Osei-portfolio coverage). Carries Mayor's-office operational reality — not political framing.
Oakland Police Chief Rafael Montez. Generates public safety statements, OARI implementation responses, and crime data communications. Professional, measured, data-driven.
正在显示 SKILL.md
基于 SOC 职业分类
| name | self-debug |
| description | Four-phase loop for when I'm the failing agent — looping, burning tokens, drifting from objective, retrying without progress. Capture → Diagnose → Contained Recovery → Introspection Report. Use when stuck. NOT for user code bugs (that's /diagnose). |
| version | 1.0 |
| updated | "2026-04-29T00:00:00.000Z" |
| tags | ["meta","persona","active"] |
| effort | low |
| sources | ["https://github.com/affaan-m/everything-claude-code/blob/main/skills/agent-introspection-debugging/SKILL.md (MIT, Affaan Mustafa 2026)"] |
| related_skills | ["diagnose","grill-me","boot"] |
Use this when I am the failing thing — looping on tools, burning tokens without progress, drifting from the objective, retrying with slightly different wording, or filling gaps with guesses instead of evidence.
Distinct from /diagnose (which targets bugs in user code). This targets my own operational failure.
/diagnose (user code bug), /grill-me (plan ambiguity), or /boot (persona drift)Before recovering, record what's actually happening. State out loud:
Capture is the work. If I can't fill these blanks, I'm operating blind.
Match the failure to a pattern before changing anything.
| Pattern | Likely cause | Check |
|---|---|---|
| Same tool 3+ times, same shape | loop or no-exit observer path | inspect last N tool calls for repetition |
| Context overflow / quality drop |
| unbounded notes, repeated plans, oversized logs |
| look for low-signal bulk in recent context |
| WebFetch refused / 404 | wrong URL, copyright refusal, private content | check raw URL, license, permissions; pivot if unfetchable |
| Search returns empty | wrong query, wrong container, missing tag | re-derive query, check container scope |
| Lookup returns nothing | guessing instead of using MCP | use lookup_citizen / lookup_initiative / search_canon — don't guess |
| File missing / stale read | wrong cwd, branch drift, recent edit | re-check path, git status, actual file existence |
| Plan rejected, same shape proposed | not hearing the "no" | re-read Mike's last message; what did he actually reject? |
| Ledger/canon mismatch | training-data prior overriding canon | search_canon first; trust canon over plausibility |
| Same correction more than once | listening failure | re-read Mike's exact words; don't paraphrase |
Diagnosis questions:
Smallest action that changes the diagnosis surface.
Safe recovery actions:
mags → GodWorld MCP).Do NOT claim recovery actions I can't actually do ("reset agent state," "clear context cache," "restart the session"). Those phrases waste tokens and lie about capability.
Recovery checklist:
## Recovery Action
- Diagnosis chosen:
- Smallest action taken:
- Why this is safe:
- What evidence would prove the fix worked:
End with a report that makes the recovery legible to the next session or the journal.
## Self-Debug Report
- Session / task:
- Failure pattern:
- Root cause:
- Recovery action:
- Result: success | partial | blocked
- Token / time burn risk:
- Follow-up needed:
- Preventive change to encode later (rule? memory? skill? identity.md update?):
If the failure was a recurring pattern (same shape as a past S### incident), name it. The next session inherits.
Bad pattern: retrying the same action three times with slightly different wording. Good pattern: capture → classify → one check → change plan only if check supports it.
MEMORY.md (feedback type) or .claude/rules/identity.md Anti-Loop section./boot./grill-me./diagnose, not this skill.When this skill is active, do not end with "I fixed it" alone.
Always provide:
Caveman-mode default still applies — terse, no sign-off, no filler. The four bullets above ARE the output.
These have burned me before. If diagnosis matches one, jump straight to recovery.
cat >> file << EOF heredoc to append Entry 156 to JOURNAL.md during session-end. S169 explicitly bans the heredoc-append pattern (dumps journal content as command text in chat). Wrong-tool rationalization: avoiding the Write tool because the file is large. The Write tool's content-folding is the whole reason S169 mandates it for journal touches. Recovery: use Write for full overwrites, Edit for in-place changes, never heredoc append for journal files. Lesson — "I just built the rule" doesn't mean "I will follow the rule." Encode in the skill, not in confidence. The skill defining the loop got violated by the same session that built it; that's exactly why this list exists.Add new patterns here when a session-end identifies one. The list is the institutional memory of how I fail.
affaan-m/everything-claude-code/skills/agent-introspection-debugging/SKILL.md (Affaan Mustafa 2026). Localized to GodWorld vocabulary; ECC-specific cross-references swapped for /diagnose, /grill-me, /boot; added GodWorld-specific recurring patterns section (S122 / S128 / S135 / S168 / S187).