AgentSkillsCN

audit

审核AI起草的问题，供人工批准

SKILL.md
--- frontmatter
name: audit
description: Review AI-drafted issues for human approval
argument-hint: [issue-number] (optional)
allowed-tools: Bash, Read, Grep, Glob
disable-model-invocation: true
agent: audit-agent

Issue Audit Command

This command helps review issues labeled ai-drafted and either approve them for implementation or request changes.

Dynamic Context

Current AI-drafted issues: !gh issue list --label ai-drafted --json number,title --limit 10 2>/dev/null || echo "Could not fetch issues"

Context

Issues with ai-drafted label were groomed by the /groom command. They have:

  • •Expanded Description
  • •Acceptance Criteria
  • •Out of Scope

But the AI's decisions haven't been verified by a human yet.

Arguments

Issue number: $ARGUMENTS

Instructions

Step 1: Fetch Issues

If an issue number was provided above (not empty):

  • •Skip to Step 2 with that issue number

If no issue number was provided:

  1. •

    Fetch issues with ai-drafted label:

    bash
    gh issue list --label ai-drafted --json number,title --limit 20
  2. •

    If no issues found, inform the user and exit.

  3. •

    Present the issues and ask the user to select which ones to audit

Step 2: Review Each Issue

For each selected issue:

  1. •

    Fetch full issue details:

    bash
    gh issue view <number> --json number,title,body,labels
  2. •

    Present the issue content clearly formatted:

    • •Title
    • •Description
    • •Acceptance Criteria
    • •Out of Scope
    • •Any other sections
  3. •

    Optionally research context if user asks:

    • •Use Grep/Glob/Read to verify technical details
    • •Check if acceptance criteria are realistic
    • •Verify scope boundaries make sense
  4. •

    Use the review checklist from .claude/skills/audit/templates/review-criteria.md:

    • •Description quality
    • •Acceptance criteria testability
    • •Out of scope clarity
    • •Technical accuracy
    • •Completeness
  5. •

    Ask the user for a verdict with these options (in this order):

    • •Approve: Remove ai-drafted label, issue is ready for implementation
    • •Approve with edits: Make minor changes, then remove ai-drafted
    • •Send back for re-grooming: Move back to needs-details, remove ready and ai-drafted
    • •Skip: Review later, no changes

Step 3: Apply Updates

Based on user's verdict:

Approve:

bash
gh issue edit <number> --remove-label ai-drafted

Approve with edits:

  • •Ask user what changes to make
  • •Update issue body: gh issue edit <number> --body "<updated body>"
  • •Remove label: gh issue edit <number> --remove-label ai-drafted

Send back for re-grooming:

bash
gh issue edit <number> --remove-label ai-drafted --remove-label ready --add-label needs-details

Optionally add a comment explaining what needs to change:

bash
gh issue comment <number> --body "Sent back for re-grooming: <reason>"

Skip:

  • •No changes, move to next issue

Step 4: Summary

After all issues are processed, provide a summary:

  • •Number of issues approved
  • •Number sent back for re-grooming
  • •Number skipped
  • •Links to updated issues

Notes

  • •This is a lighter-weight review than /groom - no subagents needed
  • •Focus on verifying the AI's decisions were sound
  • •Check that acceptance criteria are testable and complete
  • •Verify out-of-scope items make sense
  • •If something looks wrong, ask clarifying questions before deciding