workflow-retrospective
Improve the durable workflow when the same agent, review, test, scope, or validation failure pattern recurs; use after incidents or repeated verifier returns.
- License 许可
- MIT
- 最近更新
- 2026-07-29
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- workflow-retrospective
- metadata
- { "version": "1.0.0" }
- description
- Improve the durable workflow when the same agent, review, test, scope, or validation failure pattern recurs; use after incidents or repeated verifier returns.
- compatibility
- Codex and OpenCode
03 SKILL.md 正文
Workflow Retrospective
Trigger
Use when the same error class occurs twice, an agent weakens a test, reviewers miss a CI/production defect, scope drift recurs, a dangerous command is attempted, or verifier repeatedly returns the task.
Procedure
- Preserve examples and identify the generating stage.
- Distinguish requirement ambiguity, context failure, prompt failure, missing deterministic oracle, permission gap, and flaky infrastructure.
- Choose the narrowest durable control.
- Add a regression scenario for the workflow itself.
- Trial the change on a safe task before scaling.
Allowed durable outcomes
- Specification/template update
AGENTS.mdrule- Shared skill procedure
- Test, linter, static check, or schema
- Permission or Codex rule
- Quality-gate or runner change
Do not finish with only “tell the agent to be more careful.” Record owner, change, evidence, expected metric, and review date.
04 同仓库更多 Skills
workflow-retrospective
Improve the durable workflow when the same agent, review, test, scope, or validation failure pattern recurs; use after incidents or repeated verifier returns.
verification
Perform final independent read-only verification of acceptance criteria, review remediation, test integrity, scope, and quality-gate evidence.
ticket-authoring
Turn a reported or observed defect into one reviewable ticket under tickets/ with reproduction, evidence, impact, and closure criteria, bound to a fix version — without fixing it or rewriting it into a specification.
technical-planning
Create a read-only implementation plan from an approved specification before code changes; use for non-trivial features, bugs, migrations, and cross-module work.
skill-governance
Audit newly installed or updated agent skills as untrusted content and propose required, optional, or denied assignments to existing agent roles. Use after staging a skills.sh installation and before promoting the skill or changing agent permissions.
security-review
Adversarially review authentication, authorization, tenant isolation, sensitive data, abuse resistance, and production reliability for a completed implementation.