| name | bug-triage |
| description | Triage and fix GitHub bug reports for Aries AI. Use when a GitHub issue is classified as a bug, regression, defect, outage, broken flow, or runtime failure that should be reproduced, fixed on a branch, validated before/after, deployed to staging, and tracked in data/feedback-processing-log.json. |
Bug Triage
Use this skill for GitHub issues that represent defects in DeliciousHouse/aries-app.
Read this when needed
- Severity rubric:
references/severity-rubric.md
Required inputs
- GitHub issue number
- Repo root:
/app/aries-app
- Feedback log:
data/feedback-processing-log.json
Required procedure
- Confirm the issue record is pending in the feedback log:
- Inspect
data/feedback-processing-log.json for the issue entry.
- Inspect the issue in GitHub:
gh issue view <number> --repo DeliciousHouse/aries-app --json title,body,labels,comments,url
- Classify severity using the rubric.
- Reproduce the bug before changing code.
- Prefer an existing failing test.
- If none exists, add the smallest targeted regression test that fails first.
- If automation is impossible, record the exact manual repro steps and expected/actual behavior.
- Locate the likely root cause in the checked-in code.
- Create a branch named
bugfix/issue-<number>-<slug>.
- Implement the smallest safe fix.
- Run validation in this order:
- targeted regression test(s)
- nearest related test file(s)
- broader repo validation only as needed for confidence
- Deploy the branch to staging:
- Use the project's staging deploy process; if none is configured, record
staging: missing-config and continue.
- Verify the bug is reproducible before the fix and not reproducible after the fix.
- Update
data/feedback-processing-log.json so the record includes:
severity
rootCause
branch
validationBefore
validationAfter
stagingUrl
status
summaryPending
- If severity is
critical, make the result explicit in your final output so the caller can alert immediately.
Guardrails
- Do not claim a fix without before/after evidence.
- Do not skip staging verification unless the staging deploy hook is missing; if it is missing, fail loudly with the exact missing env/config.
- Prefer minimal, reversible fixes over broad refactors.
- Keep unsupported scope creep out of the bug fix.
Output format
Return only:
status: success or failed
issue: #<number>
severity: critical/high/medium/low
root_cause: short phrase
branch: branch name or none
validation_before: one line
validation_after: one line
staging: URL or failed or missing-config
alert: critical or none
next_step: short operator follow-up or none