AGENT SKILL

checklist-audit

Use when auditing or applying a vault checklist.

License 许可
Unlicense
最近更新
2026-11-03
Stars
★ 1
Forks
0

01 前言 Front Matter

name
checklist-audit
tags
[ "checklist", "audit", "review", "vault", "knowledge-base", "gap-analysis" ]
created
1785974400000
difficulty
senior
description
Use when auditing or applying a vault checklist.

03 SKILL.md 正文

Checklist Audit

Systematic review of checklists, reference documents, and guides against authoritative vault knowledge bases to identify gaps, strengthen items, and ensure cross-reference consistency.

When to Use

  • User asks to "audit", "review", or "check" a checklist or reference document
  • User wants to verify a document against their vault (SWEBOK, CyBOK, career paths, etc.)
  • User wants to identify gaps, missing items, or outdated guidance in a structured document
  • The document has sections, checklists, tier matrices, or sanity checks that need verification
  • User asks to apply a checklist TO a project ("check this checklist and apply it") — the checklist is the standard, the repo/app is the target: use Applying a Checklist to a Project below

Audit Workflow

1. Read the Target Document

read_file(path="<target-checklist-path>")

Understand the document's structure, purpose, audience, and existing cross-references.

2. Gather Vault References in Parallel

Read all relevant knowledge base files that could inform the audit. Use terminal with find if directory structure is unclear:

find "F:/obsidian_note/swe-knowledge/career-path/<domain>" -name "*.md" -type f | sort

Batch reads in parallel for efficiency. Target:

  • Career path overviews and capability notes for the domain
  • Relevant Body of Knowledge chapters (SWEBOK, CyBOK, DMBOK, SEBoK)
  • Any existing specialist notes that cover the same territory

3. Compile Structured Findings

Create a findings table:

# Finding Source Severity Location

Severity levels:

  • 🔴 Gap — Critical missing item that violates authoritative source
  • 🟡 Improve — Existing item needs strengthening or clarification
  • 🟢 Nice-to-have — Optional enhancement
  • ✅ Already Strong — Leave unchanged

4. Apply Patches Systematically

For each finding, use patch tool with mode=replace:

  1. Add new items at appropriate section with clear rationale
  2. Strengthen existing items — expand wording, add context
  3. Update section titles if they no longer reflect content
  4. Maintain numbering — keep sections sequential, no gaps

5. Update Tier Matrix

If the document has a tier applicability matrix:

  • Add rows for new sections
  • Update column entries to reflect new items
  • Ensure row numbers match actual section numbers

6. Update Sanity Checks

Add items for each new critical section item. Keep concise (10-15 items max). Each sanity item should map to a section item.

7. Verify Cross-References

Check:

  • Section numbering is sequential (no gaps)
  • Internal references (§X) point to correct sections
  • Tier matrix rows match actual sections
  • External references ([[Release]], [[API Launch]]) still valid

8. Final Verification Read

read_file(path="<target-checklist-path>")

Confirm all patches applied correctly and document is internally consistent.

9. Compile Summary Report

Provide before/after metrics, changes summary (✅ added, 🟡 improved, 🟢 untouched), cross-reference verification results, and final verdict with risk reduction statement.

Pitfalls

  • Don't add redundancy — if an item is already covered, strengthen the existing item, don't duplicate
  • Don't over-audit — if a section is already strong, leave it alone
  • Don't break cross-references — when adding items or renumbering, verify §X references still work
  • Don't inflate sanity checks — focused subset (10-15 items), not a parallel list
  • Don't ignore tier matrices — if the document has one, you MUST update it
  • Don't invent authoritative sources — only cite vault files you actually read
  • Patch text must match exactly — re-read sections before patching to avoid whitespace/mismatch errors

Examples

QA Checklist Audit (2026-08-06)

Target: checklist/qa-checklist/qa.md Vault refs: 18 files from career-path/10_Quality_and_Test_Engineering/ + SWEBOK Testing chapter Key gaps closed: Systematic test design, exploratory testing, shift-left, test oracles, regression strategy, RCA Result: 152 → 172 lines, added §7 Exploratory Testing, strengthened §1 with shift-left

Security Checklist Audit (2026-08-06)

Target: checklist/security-checklist/security.md Vault refs: Security Engineer overview, CyBOK Software Security, CyBOK Secure SDL, QA Security Testing note Key gaps closed: CSRF, TOCTOU/race conditions, error leakage, IaC scanning, remediation SLAs, crypto API misuse, security champions, abuse cases, hallucination mitigation Result: 203 → 222 lines, 11 → 17 sanity items, full OWASP Top 10 coverage verified

Folder Expansion Workflow

After auditing a checklist, check if its folder follows the vault pattern: general checklist + technology-specific companions. If the folder has only the general checklist, recommend expansion.

Identifying Expansion Candidates

  1. List all checklist folders: find "F:/obsidian_note/swe-knowledge/checklist" -type d
  2. Count files per folder — folders with 3+ files have the pattern (e.g., api-checklist/ has api.md + fastapi.md, spring-boot-api.md, etc.)
  3. Folders with only 1 file are expansion candidates

Creating Technology-Specific Checklists

Use delegate_task with tasks array (max 3 per batch due to max_concurrent_children=3 limit). Batch remaining tasks in subsequent dispatches.

Each technology checklist should include:

  • Setup & installation (language-specific package managers, dependencies)
  • Core concepts and lifecycle
  • Common patterns with real code examples
  • CI integration
  • Quick sanity check (10-15 items)
  • Tier applicability matrix (7 tiers matching vault convention)
  • Cross-references back to the general checklist

Delegation Pattern

Batch 1: delegate_task(tasks=[task1, task2, task3])  # max 3
Batch 2: delegate_task(tasks=[task4, task5, task6])  # next 3
...
Final items: write_file directly if subagents can't write

Pitfall: Subagents may report "cannot write files" in sandboxed environments. If so, use their detailed output as content and write_file directly.

Recommended Checklist Set by Domain

Domain General Technology-Specific Companions
QA qa.md pytest, jest-vitest, junit, go-test, playwright, cypress, k6-performance, testcontainers, mutation-testing, accessibility-testing
API api.md Framework-specific (already populated)
Database database.md Engine-specific (already populated)
Web web.md Framework-specific (already populated)

Applying a Checklist to a Project (compliance assessment)

Trigger: "check this checklist and apply it" — the checklist is the standard, the repo/app is the target. (Auditing the checklist FILE itself is the workflow above.)

Workflow

  1. Read the checklist completely first — it is the scope authority; its own tier matrix decides how much rigor fits the target.
  2. Tier call before checkboxes. Using the checklist's scoping matrix, state: the product's risk tier, the repo's real posture (environment model, CI reality), the adopted level (often "Tier N, adapted") and the deltas — which sections are N/A and WHY (e.g. no local webServer, no mocking when the suite's purpose is integration evidence).
  3. Assess section by section with a fixed legend: ✅ done / 🟡 partial / ⬜ gap→planned / ➖ not applicable + reason / 🔒 blocked on a named open question. Back every status with evidence (command output, grep count, file:line) — no vibes.
  4. Apply only safe quick wins now: config defaults that remove observed flakiness, type-check setup (tsconfig.json strict + tsc --noEmit), runbook/debug docs, guard clauses on tests that hang when data is absent. Do NOT mass-refactor working code mid-flight (full POM rewrites etc.) — backlog those with rationale and risk.
  5. Re-verify each applied item with a real command (tsc --noEmit, --list, smoke run) and quote the result in the assessment.
  6. Checklist-discovered gaps become NEW questions in the project's question doc (next number; update localized copies + stale counts too) — applying a checklist reliably surfaces scope gaps nobody asked about yet.
  7. Output shape: NN_<topic>_checklist_assessment.md in the project's docs folder (next number in the set), an index row, applied items logged in the decisions log, and sections for "applied now", "backlog (unblocked)", "blocked on team answers".

Application pitfalls

  • Never silently contradict a recorded team decision (e.g. an evidence-capture policy). If the checklist says X and the team decided Y, record a documented deviation with the decision reference — do not quietly "comply" with the checklist.
  • A N/A status without a stated reason is noise — attach the reason every time.
  • ✅ requires verification in the repo (grep/count/run); "obviously fine" is not evidence.
  • After edits, re-check count-bearing text the doc set states elsewhere (N questions, N docs).

Related Skills

  • playwright-suite-health-review — deep review/triage of E2E suites (a natural application target)
  • dogfood — Exploratory QA of live web apps (different: audits documents, not running systems)
  • spec-driven-code-review — Code review against specs (different: audits checklists, not code)
  • requesting-code-review — How to request reviews (different: performs audits, not requests them)

04 同仓库更多 Skills