一键导入
validate
Validate workflow markdown documents such as plans, reviews, implementation reports, and audits against the required format contract.
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。
菜单
Validate workflow markdown documents such as plans, reviews, implementation reports, and audits against the required format contract.
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
Audit a feature implementation against its approved plan and implementation report, then produce the architect verdict and follow-up direction.
Create a feature document covering business context and initial architecture for a new feature using the PM and Architect workflow.
Execute a review-approved feature plan and produce or refine the implementation report for that feature.
Orchestrate the full feature workflow loop by detecting artifact state and routing through review, implementation, audit, and follow-up steps.
Create or update a feature implementation plan with concrete steps, risks, and acceptance criteria for the Architect workflow.
Revise an existing feature plan after review findings, implementation issues, or manual edits while preserving plan history.
| name | validate |
| description | Validate workflow markdown documents such as plans, reviews, implementation reports, and audits against the required format contract. |
Purpose: Check workflow Markdown against the format contract (plans, reviews, implementation reports, audits).
Audience: Validator (subagent, CI, human running /validate) uses Commands below. Other skills only delegate using § Validator subagent — no shell in parent prompts.
Normative formats: .claude/skills/workflow/references/plan-format.md, review-format.md, implementation-format.md, audit-format.md.
Operator reference: scripts/README-validate-workflow-doc.md
| Pattern | Type |
|---|---|
*.plan.md | plan |
*.review.md | review |
*.implementation.md | implementation |
*.audit.md | audit |
If the basename does not end with one of these suffixes, pass --type plan|review|implementation|audit.
Do not redirect command output into the repo (no > file.txt / tee).
bash scripts/validate-workflow-doc.sh /absolute/path/to/file.md
bash scripts/validate-workflow-doc.sh file1.md file2.md
bash scripts/validate-workflow-doc.sh --type plan /path/to/unusual.md
bash scripts/validate-workflow-doc.sh --mdlint /path/to/file.plan.md # optional; requires markdownlint on PATH
Run from this repository root. File arguments may be absolute or relative paths to .md files.
0 — passed contract check.1 — contract violation.2 — usage / missing file / unknown type.0 in this session..txt/.log or any capture files in the workspace; no >/tee into the repo. The script only reads the target .md; stream results to the terminal and chat only.Parent skills should run validation in a separate context when possible. Parents must not paste bash lines — only paths and this template.
You are the Validator subagent.
1. Open and follow ONLY the /validate skill at:
{ABS_PATH_TO_VALIDATE_SKILL}
2. Validate exactly this workflow Markdown file:
{ABS_PATH_TO_WORKFLOW_MD}
3. Run the procedure defined inside that skill (cd, env — all details are there). Do not edit the Markdown file. **Do not** write any files (no logs, no redirects into the workspace).
4. Reply: exit code, full stderr if non-zero, PASS or FAIL.
If “spawn Validator subagent” is ambiguous: plan, replan, review, implement, audit, and loop each document (Concrete tooling — …) under Validation (or the orchestrator block for /loop). See .agents/skills/review/SKILL.md for the Reviewer pattern; sibling skills mirror the same Task / subagent_type / handoff-only rules for their role.
0, or inline fallback: you opened {ABS_PATH_TO_VALIDATE_SKILL} and completed one run for {ABS_PATH_TO_WORKFLOW_MD} per that document.See /loop skill.
| Action | File |
|---|---|
| Plan created / replanned | *.plan.md |
| Review round written | *.review.md |
| Implementation report updated | *.implementation.md |
| Audit written | *.audit.md |
Structural / enum checks from format specs only, not semantic quality of review or audit text.