菜单
Diagnosis loop for hard bugs and performance regressions. Use when the user says "diagnose"/"debug this", or reports something broken/throwing/failing/slow.
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
Ask which skill or flow fits your situation. A router over the skills in this repo.
Implement a piece of work based on a spec or set of tickets.
Configure this repo for the engineering skills — set up its issue tracker, triage label vocabulary, and domain doc layout. Run once before first use of the other engineering skills.
Turn the current conversation into a spec and publish it to the project issue tracker — no interview, just synthesis of what you've already discussed.
Break a plan, spec, or the current conversation into a set of tracer-bullet tickets, each declaring its blocking edges, published to the configured tracker — edges as text in one file per ticket locally, or native blocking links on a real tracker.
Plan a huge chunk of work — more than one agent session can hold — as a shared map of investigation tickets on your issue tracker, and resolve them one at a time until the way to the destination is clear.
| name | diagnosing-bugs |
| description | Diagnosis loop for hard bugs and performance regressions. Use when the user says "diagnose"/"debug this", or reports something broken/throwing/failing/slow. |
Uses scratch working memory for hypothesis tracking.
RULE: glossary + ADRs. Skip phase only with explicit reason. DO NOT hypothesise before deterministic loop.
MUST build fast pass/fail loop. Try order:
Bash vs dev server.Bash.git bisect run harness via Bash.scripts/hitl-loop.template.sh.Rules:
Initialize session: .scratch/diagnoses/<YYYY-MM-DD>/zoo-diagnosing-bugs-<slug>/
Write <session-dir>/session.md with symptom, repro steps, error signatures.
Generate 3–5 ranked falsifiable hypotheses. Write each to <session-dir>/hypothesis-<N>.md:
If {cause}, then {probe} will {result}Drop vague hypotheses. Show ranked list. User AFK → proceed top hypothesis.
Before each probe, read all <session-dir>/hypothesis-*.md files to maintain context.
[DEBUG-xxxx].After each probe, update the corresponding hypothesis file with:
Continue until root cause is confirmed or all hypotheses are rejected.
OWNER: Implementer profile (requires source-code edits)
Delegate to implementer profile via Agent tool before this phase. Implementer executes:
Implementer does NOT return completion here — more phases remain. Continue to Phase 6.
OWNER: Implementer profile (requires source-code edits and git commit)
MUST finish:
[DEBUG-...] logs removed from source./zoo-improve-codebase-architecture.Implementer returns results back to architect profile for Phase 7.
Before broad reads, locate relevant files/symbols with Glob, Grep, or targeted Read.
Prefer targeted Read with line ranges or block reads once the relevant area is known.
Read full files only when structure, ordering, or surrounding context is required for correctness.
Do not re-read unchanged files; use prior findings unless the file changed.
For logs or large outputs, search for the failing marker/error first using Grep. Read only surrounding ranges needed to prove or disprove a hypothesis.
OWNER: Architect profile (writes to .scratch/)
Read all <session-dir>/hypothesis-*.md files.
Write <session-dir>/root-cause.md:
Write <session-dir>/diagnosis.md with full log (hypotheses, instrumentation results, timeline).
COMPLETION PROTOCOL: Your final text message is your return value. Output this summary, then STOP — do not call any more tools.
<session-dir>/root-cause.md path/zoo-verify then /zoo-review then /zoo-commit-and-document)