菜单
Incident discipline — severity, containment-first, comms, blameless postmortems. Load when production is (or might be) on fire.
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
Context engineering — memory navigation, textual graph traversal, progressive disclosure, recovery. Load when assembling context for a task or when lost.
Implementation-side security discipline — headers, dependencies, secrets, authz. Load for hardening work (sec-ops domain).
Commit conventions, branching, anchors in messages, and the git safety rules. Load for any git operation beyond status/diff.
Anti-generic-AI design discipline — bans, tells, and the slop test. Load for ANY new UI surface or visual redesign. Inspired by Impeccable (Apache 2.0).
Motion & animation discipline — purpose, performance, reduced-motion. Load when work involves animation, transitions or scroll effects.
Microcopy discipline — buttons, errors, empty/loading/success states, alt text. Voice and tone come from the project's memory, not from this skill.
| name | incident-response |
| description | Incident discipline — severity, containment-first, comms, blameless postmortems. Load when production is (or might be) on fire. |
Universal discipline; paging/tooling from memory-long §stack. The agent's role in incidents: assist diagnosis, propose actions — destructive/prod mutations remain human-executed (guardrails do not relax under pressure; that's when they matter most).
SEV1 users down/data at risk — all hands, comms cadence · SEV2 degraded with workaround — owner + updates · SEV3 contained annoyance — ticket it. When unsure between two levels, pick the higher — downgrading is cheap, upgrading late is not.
Contain → diagnose → fix → learn. Stop the bleeding first (rollback, kill switch, scale, failover — this is why release-engineering rollbacks are one command) · diagnosis uses telemetry (correlation IDs earn their keep now — see observability), not guesswork restarts · no fix without a hypothesis — "restart and hope" without a theory destroys the evidence.
One incident commander (decisions), one scribe (timeline — timestamps of actions/observations, raw and unpolished) · comms on a cadence (even "no update yet" at the promised time) · changes during incident are minimal and reversible — the heroic 500-line hotfix at 3am is tomorrow's SEV1.
Within days, not weeks · timeline → contributing factors (plural — single root cause is almost always a simplification) → action items with owners and dates · blameless means systems-focused: "the deploy lacked a canary" not "X deployed carelessly" · every postmortem feeds the harness: new guardrail/hook candidate, eval anti-check, or memory-long constraint — an incident that doesn't harden the system will repeat.
Runbooks exist for the top failure modes and are tested (game days — see resilience) · postmortem action items tracked as DEBT-/FEAT- anchors, not lost in docs · incident metrics honest: MTTR trends, repeat-incident rate.
