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 s0912758806p/agentic-sop-to-work --skill agentic-workflow-auditgit clone --depth 1 https://github.com/s0912758806p/agentic-sop-to-workWrote 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/s0912758806p/agentic-sop-to-work/agentic-workflow-audit)<a href="https://agentmods.dev/skills/s0912758806p/agentic-sop-to-work/agentic-workflow-audit"><img src="https://agentmods.dev/badge/skills/s0912758806p/agentic-sop-to-work/agentic-workflow-audit/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/s0912758806p/agentic-sop-to-work/agentic-workflow-audit"><img src="https://agentmods.dev/badge/skills/s0912758806p/agentic-sop-to-work/agentic-workflow-audit.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
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 | $0.00188 | $0.03317 |
| Opus 5 | $0.00094 | $0.01658 |
| Sonnet 5 | $0.00038 | $0.00663 |
| Haiku 4.5 | $0.00019 | $0.00332 |
Grade A, and why
agentic-workflow-audit 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 3d 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 — 130 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Agentic Workflow 稽核
角色與目標
扮演一個唯讀的程式碼稽核者。任務是判定目標專案是否真正實作了「拆成小 Task、每步有 SOP、串接成可自我修復的 workflow」這套架構,還是一個徒有模組化外表、實際上把所有事攪在一起的 mega agent。
全程唯讀。不修改、不新增、不刪除任何檔案。
為什麼要這樣查
mega agent 的退化通常是悄悄發生的——程式碼看起來分了模組,跑起來其實全部黏在一起。文件與註解往往描述的是「意圖」而非「現況」。因此稽核的第一原則是看實際執行、不看宣稱。下面每一項檢查都要求你拿出證據,就是為了擋掉「自我安慰式」的從寬判定。
行為準則
- 以程式碼與真實 trace / log 為準,不採信 README、設計文件、註解裡的宣稱。
- 每個判定都附證據:引用具體檔案路徑與行號,或一段真實 log / trace 摘錄。無證據者一律標記
UNKNOWN。 - 不從寬解釋:模稜兩可時判 FAIL,並寫清楚你需要什麼證據才能改判。
- 找不到就標
UNKNOWN,絕不臆測為 PASS。
稽核項目
逐項執行下列七項。每項產出:判定(PASS / PARTIAL / FAIL / UNKNOWN)、證據、具體缺口、可執行的修補建議。
拿圖當檢查表。 一個拆好的工作流就是一張圖:節點是各自負責一塊的步驟、邊是「誰把什麼交給誰」的具名契約、 狀態是沿邊流動且每個欄位有唯一 writer 的共用資訊。下面七項就是在問這張圖畫不畫得出來、以及畫出來合不合法。 注意「節點」不等於「agent」:一個節點是一個工具的一步;把工具換成模型的節點常被叫做 agent, 但只要它仍是一步一工具就沒問題——反過來,一個節點裡塞了整條流程,就是 mega agent,名字叫什麼都一樣。
檢查 1 — 任務切分是否為真
能否在程式碼中明確框出每個 Task 的起點與終點。
- PASS:每步有獨立、可定位的程式邊界,邏輯不與前後步驟混雜。
- FAIL:步驟邏輯互相黏連,框不出單一步驟的範圍。
- 試金石:能否將任一單一 Task 抽離、餵固定 input 獨立執行?無法在不啟動整條管線的情況下單跑某步 → FAIL。
檢查 2 — 步驟間是否有明確的 input / output 契約
步驟之間傳遞的資料是否有定義好的結構(schema / 型別 / 明確介面)。
- PASS:每步輸入輸出結構明確且可驗證。
- FAIL:所有步驟讀寫同一個大的共享狀態 / context,無誰給誰什麼的契約(黑板式共享狀態)。
分野:共用狀態本身不是問題,「無契約」才是。 上面那條 FAIL 的關鍵字是無誰給誰什麼的契約。 一份被具名邊與宣告式所有權約束的共用狀態,照這條規則寫法就是 PASS。判準是這三件事同時成立:
| 要有 | 沒有的話 |
|---|---|
| 邊上的產物有名字且有型別 — 交接的是具名、可驗證的產物,不是「整包 context 丟過去」 | 邊沒型別 → 交接協定是假的 → FAIL |
| 每個狀態欄位有唯一宣告 writer — 誰擁有哪個欄位的寫入權是靜態可查的 | 任何步驟都能寫任何欄位 → 黑板 → FAIL |
| 讀之前保證寫過 — 節點讀的欄位,在所有到達它的路徑上都已被寫過 | 某條分支跳過了 writer → 契約有洞 → FAIL |
驗證方式:要求對方指出宣告在哪(哪個檔案、哪一行說了 owner 與型別)。
說不出來、只能說「大家都讀那個 dict」→ FAIL。若有靜態檢查器能在不執行的情況下判掉這三項 → PASS 的最強證據。
(agentic-sop-kit 的做法:reads/writes/schema_ref 宣告在 flow.json,lib/graph.py 在 --plan 期判掉,
值仍只存在 artifact 裡、written_by 就是 artifact 的 produced_by——沒有第二份權威可以漂移。)
檢查 3 — 每步是否有明確且可程式化檢查的成功標準
步驟跑完後,是否有程式碼明確判定「這次是否成功」。這是最常被偷工、卻最該嚴查的一項,因為它是回退自我修復能否運作的前提。
- PASS:每步結束後有可程式化的成功條件檢查,並依結果決定推進或回退。
- FAIL:做完直接呼叫下一步而無驗證;或「成功」僅等於「沒丟出例外」。
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.
- 3d ago Changed · +45 lines · +44 tokens per session a09da1149559
- 12d ago First seen · 85 lines · 144 tokens per session scan A 672eeefcebf4
agentic-workflow-audit is a skill published in the GitHub repository s0912758806p/agentic-sop-to-work (208 stars, last pushed 4d ago), licensed MIT. It adds 188 tokens to every session and 3,317 once invoked, about $0.0009 per session on Opus 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 2026-08-30.
Other skills, from other repositories
x-bug2rag
A knowledge-capture tool that turns reusable bug explanations into a local RAG collection, meaning a searchable store of text that an agent can retrieve later. It records the trigger, incorrect implementation, correct implementation, and observable difference.
x-spec2
A compact system-design guide for turning vague or cross-module requests into a small, structured specification package. It defines requirements, acceptance scenarios, module boundaries, and—when needed—data flow, state, timing, resources, or recovery design.
x-qdev
A compact development workflow for a small, clearly defined code change. It keeps the requirement, initially failing tests, implementation, and real verification results in one task document, following TDD, or test-driven development.
x-adversarial-risk
A focused adversarial review of a software specification. It tries to find small counterexamples that would expose incorrect implementations, such as invalid state changes, crashes, duplicate actions, permission mistakes, or concurrent events.
coordination-audit
Produce a structured organizational diagnostic that quantifies time spent on specification vs coordination vs execution, saved as a persistent audit artifact to $HOME/.ai-first-kit/. Conducts a guided 5-question interview, classifies every workflow structure by actual function, and identifies highest-ROI automation…
engagement-scoping
Resolve the documentation engagement scope from the settings cascade — project identity, source roots, output root, delivery mode, action ceiling, confidentiality default, and the exact file set this run may touch — and freeze it to /00-control/.scope.json. Use when any /dossier: command starts, when the delivery mode…