AGENT SKILL

resilience-review

Run a risk-ranked Murphy review when credible failure could cause data loss, security or privacy harm, irreversible action, broken contracts, or a likely user dead end.

License 许可
MIT
最近更新
2026-11-11
Stars
★ 4
Forks
1

01 前言 Front Matter

name
resilience-review
description
Run a risk-ranked Murphy review when credible failure could cause data loss, security or privacy harm, irreversible action, broken contracts, or a likely user dead end.

02 资源文件 1

03 SKILL.md 正文

On interface/type/shared-state changes, read Poteto.

Murphy pass, not edge-case harvest.

Evidence first

Credibility needs a trust boundary, irreversible effect, contract, incident, demonstrated scale, or likely path. Could happen is insufficient; skip low-risk work.

Map action, state changes, effects, dependencies, scale. Codex runs inline unless explicit agents or /swarm were requested. Probe:

  • Input: malformed/stale trust-boundary data.
  • Timing: duplicate/out-of-order corruption or deception.
  • System: dependency failure breaks contract.
  • State: reachable impossible state.
  • Recovery: user stuck/fake success; retryable writes: repeat and resume after a partial effect per Poteto.

Findings need evidence, trigger, expected behavior, smallest guard and public-contract test. No evidence means no finding.

Security/privacy/data loss/destructive actions fail closed. Otherwise prefer clear failure over speculative retries/fallbacks/caches/flags/observability.

Use /read-the-damn-docs for external contracts, /diagnosing-bugs to confirm real defects, then one RED regression test. /visual-review only for customer recovery UI.

Output

## Resilience review
Risk surface:
Evidence:
Credible findings: | Scenario | Evidence | Smallest guard | Contract test |
Verdict: PASS | NEEDS_GUARDS | BLOCKED

Cite files/routes/forms/API; rank, do not count. High-impact gaps block; hypothetical edges do not. Reference.

04 同仓库更多 Skills