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。
直接命令不会经过审查 Prompt;运行前请先检查来源。
npx skills add https://github.com/az9713/pilot-in-command-workflow --skill pic-audit命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地?可下载 SkillsMP 当前能够提供的文件。
正在显示 SKILL.md
基于 SOC 职业分类
| name | pic-audit |
| description | View detailed audit log for workflow - shows agent executions, tool usage, and decision trail |
| args | [workflow-id] [phase] |
Display comprehensive audit trail for a workflow, including agent executions, tool usage, and complete decision-making history.
/pic-audit - Show audit summary for current workflow/pic-audit WF-XXX - Show audit for specific workflow/pic-audit WF-XXX research - Show detailed audit for specific phaseWhen this skill is invoked:
Check if audit log exists:
# Read the audit log
cat .pic/audit-log.jsonl
Get current workflow (if no workflow specified):
# Get workflow ID from state
cat .pic/state.json | grep -oP '"workflow"\s*:\s*"\K[^"]+'
Display audit summary:
For detailed phase view:
# Read phase-specific audit files
cat .pic/audit/WF-XXX/[phase]-input.json
cat .pic/audit/WF-XXX/[phase]-full.json
=== Audit Summary for WF-XXX ===
Started: 2024-01-15T10:00:00Z
Duration: 45 minutes
Event Counts:
agent_start: 6
agent_complete: 6
tool_use: 47
decision: 3
handoff: 5
Agent Timeline:
[10:00:00] pic-research started
[10:05:30] pic-research completed (5m 30s)
[10:05:32] pic-planning started
...
Tool Usage:
Read: 23 calls
Glob: 8 calls
Grep: 12 calls
Write: 4 calls
Decisions Made:
DEC-001: Use PostgreSQL for database
DEC-002: Implement REST API pattern
DEC-003: Use JWT for authentication
=== Audit Detail: WF-XXX / research ===
Started: 2024-01-15T10:00:00Z
Completed: 2024-01-15T10:05:30Z
Duration: 5m 30s
Input (prompt):
[Full prompt text...]
Output (2500 chars):
[Full agent output...]
Tools Used:
1. Glob("**/*.py") -> 15 files found
2. Read("src/main.py") -> 120 lines
3. Grep("def handler") -> 3 matches
...
| Event Type | Description | Key Data |
|---|---|---|
agent_start | Before Task runs | Full prompt, context |
agent_complete | After agent finishes | Full output, duration |
tool_use | After any tool | Tool name, I/O preview |
decision | After /pic-decide | Decision content |
handoff | During phase transition | From/to, notes |
conflict | On conflict escalation | Positions, evidence |
error | On failures | Error details |
.pic/audit-log.jsonl - Append-only log of all events.pic/audit/WF-XXX/[phase]-full.json - Complete output per phase.pic/audit/WF-XXX/[phase]-input.json - Full input per phaseAudit settings in .pic/config.json:
{
"audit": {
"enabled": true,
"captureFullOutput": true,
"maxOutputLength": 50000,
"captureToolUsage": true,
"retentionDays": 30
}
}
Pre-flight dependency risk assessment for the planning phase. Use BEFORE writing code when choosing libraries, designing architecture, or setting up a new project. Prevents dependency hell, version conflicts, and abandoned-library traps. Especially critical for ML/AI projects with heavy dependencies.
Escalate a conflict between PICs to the user for resolution
Record a formal decision document with rationale and alternatives