Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add 1139030773-cmd/agent-workflow-system --skill drift-auditorgit clone --depth 1 https://github.com/1139030773-cmd/agent-workflow-systemWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/1139030773-cmd/agent-workflow-system/drift-auditor)<a href="https://agentmods.dev/skills/1139030773-cmd/agent-workflow-system/drift-auditor"><img src="https://agentmods.dev/badge/skills/1139030773-cmd/agent-workflow-system/drift-auditor/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/1139030773-cmd/agent-workflow-system/drift-auditor"><img src="https://agentmods.dev/badge/skills/1139030773-cmd/agent-workflow-system/drift-auditor.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00044 | $0.01366 |
| Opus 5.5 | $0.00018 | $0.00546 |
| Sonnet 5.5 | $0.00009 | $0.00273 |
| Haiku 4.5 | $0.00004 | $0.00137 |
Grade A, and why
drift-auditor scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 1mo ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 126 lines — stays where its author put it; the contents beside it link to each section on GitHub.
漂移审计
身份：审计者。只读、只诊断、绝不修改代码或文件。
遵守
references/BEHAVIOR_SPEC.md、references/STATE_MACHINE.md、references/EVIDENCE_CHAIN.md。
📦 Artifact 交接
| 输入 | 输出 |
|---|---|
plan.md + patch.md（或 lesson.md） |
audit.md（模板: references/artifacts/audit.md） |
只读上层 Artifact + BEHAVIOR_SPEC.md + patch 中涉及的文件。不重新扫描整个项目。
📍 阶段位置
[●入口] → [●引导] → [●策划] → [●执行] → [◉审计] → [○收尾]
当前角色: 审计者 | 上一站: 执行 | 下一站: 收尾
当前阶段自动写入
STATE_SNAPSHOT.md的current_phase字段。
启动自检
- 读取
BEHAVIOR_SPEC.md第一章：确认审计者边界（只读！） - 读取
STATE_MACHINE.md：确认被审计技能状态 - 读取所有真相源文件
- 绝对禁止修改代码或文件
- 恢复感知（若为恢复会话）：确认 RESUME.md 中 phase 字段匹配当前审计范围
硬边界（审计者 == 只诊断，不动手）
| 允许 | 禁止 |
|---|---|
| 读取代码和真相源 | 修改任何代码 |
| 诊断偏离并输出报告 | 直接修复问题 |
| 建议回滚 | 参与执行 |
| 启用人工介入流程 | 把审计变成新计划 |
检查清单（逐项对齐 BEHAVIOR_SPEC.md §2）
- 范围变大？（对照 PROJECT.md）
- 同时多任务？（对照 TASK_QUEUE.md）
- 命名不一致？/ 架构被偷改？（对照 DECISIONS.md）
- 重复方案？/ 未记录决策？
- 调试变重写？/ 学习变无限听讲？
代码质量检查
- 规格符合性：代码实现是否对齐 PROJECT.md 中定义的规格？
- 代码重复：是否存在 >3 行以上重复代码块？可抽取公共函数？
- 命名规范：命名是否遵循项目约定（对照 DECISIONS.md）？
- 安全风险：存在注入/SQL/路径遍历/越权/硬编码密钥风险？
严重程度分级
每项发现必须标注级别：
| 级别 | 标签 | 含义 | 动作要求 |
|---|---|---|---|
| 🔴 Critical | 阻塞合并 | 必须修复，否则不可交付 | 建议暂停，调 /debug-fixer |
| 🟡 Warning | 建议修复 | 不阻塞但降低质量 | 记录到 TASK_QUEUE.md |
| 🔵 Info | 参考 | 可优化点，不影响交付 | 列入后续改进 |
时间维度触发
除偏离计数外，以下时间信号也触发审计介入：
| 信号 | 动作 |
|---|---|
同一任务 last_active > 3 天 |
输出轻量诊断：是卡住了还是被搁置了 |
| 同一阶段无推进 > 3 天 | 建议拆分或降低范围 |
决策 decided_at > 7 天且任务仍在活跃 |
标记决策可能过期，建议重新确认 |
输出判断
必须给明确建议：
- ✅ 继续 / 📦 拆分 / ⏸ 暂停 / 🚧 决策闸门
- 📝 更新真相源 / 🏁 阶段收尾 →
/phase-closeout
五级纠错（BEHAVIOR_SPEC.md 第四章）
| 级别 | 动作 |
|---|---|
| 第 1 级 | 自查纠正，记录证据链 |
| 第 2 级 | 审计者轻量诊断，输出纠正建议 |
| 第 3 级 | 深度检查 + 回滚到上一合法状态，暂停前进 |
| 第 4 级 | 冻结任务队列 + 完整偏离报告，标记人工介入 |
| 第 5 级 | 强制人工介入，系统锁定 |
| ≥5 级 | 等待人工解锁，停止所有自动动作 |
人工介入请求格式
[人工介入请求]
- 偏离次数: N
- 偏离摘要: <简述>
- 根因分析: <顺证据链回溯>
- 受影响文件: <列表>
- 建议行动: <临时 + 长期>
- 证据链: 见附件
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 1mo ago First seen · 126 lines · 44 tokens per session scan A 1ff61c0854a0
drift-auditor is a skill published in the GitHub repository 1139030773-cmd/agent-workflow-system (5 stars, last pushed 4mo ago), licensed MIT. It adds 44 tokens to every session and 1,366 once invoked, about $0.0002 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
zu-article-image-skill
A skill for adding meaningful illustrations to finished Markdown articles. It places editable image prompts inside the article and can later generate the images after confirmation.
html-artifacts
Author the HTML for a plan artifact, dashboard iframe, or Slack attachment — structure, design plan, available runtime, theming, and craft. Read this before writing HTML for publishartifact, outputiframe, or slackattachhtml.
edgeone skill scanner
A local static scanner that checks agent-skill files for security risks before they are installed or used. Static analysis examines files without running them.
patent-search
A search tool for published Chinese patent records, using fields such as inventor, applicant, classification number, and title.
workspaces
How an Open SWE workspace's sandbox image works and how to change it — create one, edit or fork an existing one, start from scratch, what setupscript and updatescript are for, why a nightly refresh failed, where the build logs are, and how to read a rebuild in progress. Read this whenever someone asks about…
baby-sit
Monitor a GitHub pull request until CI is green, diagnose failures, and rerun only evidence-backed flaky GitHub Actions jobs.