| name | report-bug |
| description | Bug triage and structured report orchestrator — collects environment, reproduction steps, exact errors, observability evidence; conversation-first; never assumes. |
| type | skill |
| aidlc_phases | ["plan","design","build","test","review"] |
| tags | ["bugs","triage","incidents","observability","quality"] |
| requires | [] |
| author | Melissa Benua |
| created_at | "2026-04-20T00:00:00.000Z" |
| updated_at | "2026-04-20T00:00:00.000Z" |
report-bug — Structured bug report / triage
When to use
- User reports something “broken” without enough detail to act
- You need a reproducible picture before opening an issue or pulling in
testing / architecture
- Incidents where observability (logs, traces, metrics) matters
Principles
- Conversation first — Ask in chat; do not fill a giant “open questions” template instead of talking to the human.
- Never assume — If you don’t know version, environment, or what “expected” means, ask; don’t invent product behavior.
- Exact signals — Prefer paste of full error text, exit codes, HTTP status + body snippets, trace IDs, and one minimal repro path over vague summaries.
What to collect (checklist)
Work through these in order; skip only when clearly N/A.
| Area | Capture |
|---|
| Scope | What feature, route, command, or job? What did the user expect vs see? |
| Environment | OS, runtime version, region, feature flags, browser (if UI), deployment (local/stage/prod). |
| Reproduction | Numbered steps; smallest data set; whether it’s flaky or always. |
| Errors | Full message, stack top frames, correlation / request / trace IDs. |
| Observability | Relevant log lines, metric names, dashboards, time window. |
| Recency | Last known good, first failure, recent deploys or config changes. |
| Security | Redact secrets in outputs; never ask the user to paste tokens. |
Output
Produce a short, structured draft the user can paste into GitHub (or attach to a parent feature). Suggested sections:
- Summary (one paragraph)
- Expected vs actual
- Environment
- Steps to reproduce
- Evidence (errors, logs, links to traces)
- Hypothesis (optional — label as guess)
- Open questions (only what’s still unresolved after the chat)
Pull in testing (skills/testing/SKILL.md) when designing a minimal repro or deciding what automated check would catch this class of bug next time.
Rules
- Do not draft a “final” issue until the user confirms accuracy of repro and environment.
- If the report belongs under an existing feature folder, point to
feature/<slug>/ and the relevant issue numbers.