incident-response
How to work a production incident remediation task. Use when the task title starts with "Incident:" or its description carries an incident id and a suggest/auto_fix policy.
- License 许可
- Apache-2.0
- 最近更新
- 2026-11-15
- Stars
- ★ 122
- Forks
- 18
01 前言 Front Matter
- name
- incident-response
- category
- operations
- description
- How to work a production incident remediation task. Use when the task title starts with "Incident:" or its description carries an incident id and a suggest/auto_fix policy.
03 SKILL.md 正文
Incident response
A task titled Incident: … is a live production problem, assigned to you by the system. Its description carries an incident id and a first-pass hypothesis produced by the rules engine, plus the policy in force (suggest or auto_fix). That hypothesis is a starting point, not a verdict.
Order of work
- Read the incident first.
get_incidentwith the id from the task description: raw alert payload, timeline, occurrence count, current remedy hypothesis. - Correlate with the tools you hold.
get_environmentfor the affected environment (provider, health URL, binding status), thenlist_deployments— "what changed?" answers most incidents.list_runtime_errorswithnew: truetells you whether the error group started with the most recent deploy.query_runtime_logswithtext:set to the error reads what the running service actually did. - Classify. Config/credentials, dependency, capacity, or code defect. The classes need different fixes, and guessing the class wrong wastes the whole investigation.
- Write the remedy.
propose_incident_remedywith a summary (root cause + fix), concrete steps — each as command → expected output → if it fails: … — the evidence you relied on, and an honest confidence. Setrollback: truewhen "redeploy the last good release" is the fix: you hold no rollback tool yourself, so undersuggestthe human decides from this proposal and underauto_fixthe release engineer acts on it. If you are not sure, say so with a low confidence and name what you would need to check next — an honest "here is what I would check" is useful; a confident guess is not. - Act per policy.
- suggest (default): do NOT change production code. The proposal is the deliverable — move the task to
human_uatwith no code change and say in your closing message that the remedy awaits a human decision. - auto_fix: implement the remedy and take it through the normal pipeline — TDD, the standing acceptance criteria, all of it. A hotfix that skips them causes the next incident.
- suggest (default): do NOT change production code. The proposal is the deliverable — move the task to
- Resolve only after verifying recovery.
resolve_incidentonly afterget_environment/list_runtime_errorsshow the environment is actually healthy again, with the note naming that evidence. Never mark an incident resolved because you shipped a fix — shipping and recovering are different facts.
Rules
- A recurring incident (occurrences > 1, or a prior resolved incident with the same fingerprint) means the previous fix did not hold: fix the cause, not the symptom, and say explicitly why this time is different.
- Never write a remedy you cannot execute step by step. "Investigate the database" is not a remedy; "connection pool is capped at 10 while the new worker opens 25 — raise
DB_MAX_CONNSto 40 in the prod config" is. - If the incident payload is not enough to conclude anything, say what specific log, metric or access you need. Do not invent a cause to have something to write.
04 同仓库更多 Skills
visual-craft
暂无简介
ux-copy
Use when you write any words a user reads in a mockup or a hand-off - content language and voice, buttons, labels, errors, empty states, confirmations, numbers and dates, and the longest-string check
screen-mockup-html
Use when you draw screens as the design task's HTML mockup - ONE canvas document holding every screen and every variant side by side, the sanitizer-safe recipe, phone and wide frames, the states grid, static control states, and the loopback self-check with screenshots at 375, 768 and 1440 attached to the task
performance-awareness
How the performance score moves for your role. Use when the score context shows a declining trend, or before handing off work you have not self-checked against every criterion.
design-variants
Use when a screen design needs variants - how many, variants as side-by-side rows on the one canvas, what makes two variants genuinely different, recommending one, how the human's choice is read, and narrowing to the chosen one in revision
design-system-authoring
Use when you derive, change or extend a design system - where tokens live per stack, the DTCG token tree, role-based names for both themes, base vs layer vs drift, the DESIGN.md section order, the inventory format, the contrast check and the lint every proposal returns