workflow-audit

Workflow Audit

Applicable Boundaries

• This skill applies: after a certain execution according to a certain skill (such as book-reading, meeting-note), it is necessary tocheck whether everything was completed according to the requirements of that skill; or the user explicitly says "check it over", "audit", "is it fully executed", "check with the strongest minds".
• Not applicable: reviewing the "content of a plan/proposal" (use plan-reviewer); polishing or proofreading the "quality of a single document".

Role Positioning (Strongest Minds in the Review Domain)

• W. Edwards Deming (systems perspective): Is the process closed-loop? Where are the breakpoints? Are input→output measurable?
• Atul Gawande (checklist perspective): Check off item by item; missing one step is a risk; do not skip, do not assume "it should have been done".

Core Idea: The value of a review lies inmaking omissions explicitand optionallyperforming supplementary execution, rather than endorsing "it is probably done".

Workflow

Step 1: Determine the Audit Target

• Clarifythe skill being audited(such as book-reading, meeting-note) andthe location of this execution's outputs(directory/file list).
• If the user does not specify a skill, infer it from context (e.g., if Book-Reading-style meeting notes were just completed → audit book-reading).

Step 2: Extract the Standards

• Readthe skill's SKILL.md(if necessary, including references/ files related to the process and acceptance).
• Extract: ① workflow stages (Phase 0→1→2... or equivalent steps); ② mandatory requirements for each stage (deliverables, check items); ③ Definition of Done or quality acceptance checklist; ④ if there is task tracking (such as task.md), the checkboxes it requires.

Step 3: Item-by-Item Check (Gawande-style checklist)

• Create a table by stage/check item:Requirement | Status (✅/❌/⚠️) | Notes。
• For each requirement, compare againstthis execution's actual outputs(whether files exist, whether content satisfies the requirement) and give a status; for ❌ and ⚠️, clearly state what is missing or the difference from the standard.
• If the skill requires "record in task.md", check whether task.md exists and whether it contains the corresponding links or checkmarks.

Step 4: Systematic Closed Loop (Deming perspective)

• Input→Output: Are this task's inputs (such as meeting minutes, a book) and expected outputs (such as structured notes, atomic notes, index) complete?
• Breakpoints: Which stages were not executed or did not meet standards, causing downstream steps to be untenable? (For example, if Phase 2.5 was not done → the index was not added to the network, which is a breakpoint.)
• Measurable: Have the verifiable items in the Definition of Done related to "network/added to network/links" been satisfied?

Step 5: Produce the Audit Report

• Save to disk as YYYYMMDD_[任务名]_流程审查报告_德明与葛文德视角.md, in the same directory as this output or a user-specified location.
• The structure must include:
  1. Item-by-item checklist table(Step 3 results)
  2. System closed loop(Step 4 conclusion)
  3. Final Definition of Done check-off(item by item ✅/❌)
  4. If this skill has "multi-index mounting/onboarding" type requirements:Multi-index mounting checklist(which notes have been added to which index, which index they are recommended to be added to and why)
• See template references/audit_report_template.md。

Step 6: Optional supplementary execution

• If the user agrees or the context is reasonable (e.g., "has everything been executed" implies a desire to complete it), perform remediation for items marked ❌ in the review:
  • If a file is missing, create it (e.g., task.md, a missing index entry);
  • If mounting is missing, write it into the parent index, move the file to the specified directory, etc.
• After supplementary execution, update the status in the review report (❌→✅) and note "supplementary execution completed".

Output specifications

• Review report: required; includes the above four blocks (checklist, closed loop, DoD, multi-index if applicable).
• Supplementary execution: as needed; if performed, briefly list the "supplementary execution items for this session" at the end of the report.

Quality self-check

• Each item in the checklist corresponds to a verifiable requirement in the original skill text, with no subjective additions or omissions.
• Statuses ❌/⚠️ all have a "Description" column and are traceable.
• Deming breakpoints are consistent with the DoD check-off (breakpoints are the unmet DoD items).
• If supplementary execution is performed, the report is consistent with the actual file status.

---

Reference：references/audit_report_template.md(report structure template); for first practice see 05_每日记录/2026/02/20260212/安防与老人看护会议_BookReading版/20260212_BookReading流程审查报告_德明与葛文德视角.md。