Skip to content
Back to skills

Review Docs

ASecurity

Review the docs/ hub for broken links, staleness, duplicate coverage, index currency, and size bloat. Files every finding to the board (tasks for high/critical, a ledger comment for medium/low).

  • 3 stars
  • 0 votes
  • 0 copies
  • 2 views
  • Added September 19, 2026
ai-agentsgobashgitapidatabasesecuritydocumentation

Works with

  • claude code
  • api
  • mcp

Security analysis

A100/100

Scanned September 29, 2026

npx -y skills add HuntsDesk/ve-kit --skill review-docs --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Review Docs?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Review Docs
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/huntsdesk-review-docs/badge)](https://www.skillsdirectory.com/skills/huntsdesk-review-docs)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: review-docs
description: Review the docs/ hub for broken links, staleness, duplicate coverage, index currency, and size bloat. Files every finding to the board (tasks for high/critical, a ledger comment for medium/low).
disable-model-invocation: true
user-invocable: true
---

# Docs Review Protocol

Reviews the project documentation tree at `docs/**/*.md` plus the master index at `docs/README.md`.

Part of the `review-*` skill family. Cites [../_shared/review-checklist.md](../_shared/review-checklist.md).

**Owner**: `docs-manager` agent.

## When to invoke

- After multiple features ship without doc updates
- Quarterly audit
- Before a new team member onboards
- After significant infrastructure changes (service migrations, architecture shifts)

## Output contract

Board tasks only. Lean summary to main agent.

---

## Step 1 — Inventory

- Glob `docs/**/*.md` (excluding `archive/` and `code-backup/`)
- Record: file path, line count, last git commit date (via `git log -1 --format=%cs -- <file>`)
- Parse `docs/README.md` — extract every link and match to actual file

## Step 2 — Index currency

For `docs/README.md`:
- Every top-level doc mentioned? (Glob `docs/*/README.md` + `docs/*.md`)
- Every link resolves (no broken paths)?
- Any docs in `docs/` not referenced from the README?
- Order/tier still accurate?

Flag:
- Orphaned docs (exist but unlisted)
- Dead links (listed but file missing)
- README section headers out of date

## Step 3 — Link integrity

For each doc file, check every markdown link `[text](path)`:
- Relative paths resolve
- External URLs still reachable (spot-check; full check is expensive — skip unless asked)
- Cross-repo links (pointing to CLAUDE.md, .claude/rules/, services/) resolve
- Line-anchor links (`#L42`) valid

## Step 4 — Staleness

For each doc, flag:
- Last commit > 9 months ago + file still referenced as "current" / "primary" (likely stale)
- References to Supabase Edge Functions (retired 2026-02-24)
- References to retired services, tables, or domains
- "Last updated" markers that are older than 6 months if the file implies ongoing currency
- Model ID references (should be Opus 5.5 / Fable 5.1 / Sonnet 5 / Haiku 4.5 — `claude-opus-5-5`, `claude-fable-5-1`; Opus is the agent default)
- Deprecated Claude Code commands
- Stripe API version mismatch (currently `2026-04-22.dahlia`)

## Step 5 — Duplicate content

For the tree as a whole:
- Files covering the same topic in different directories (e.g., `docs/auth/` + `docs/authentication/`)
- Sections within files that duplicate content from other files
- Versioned docs (`-v1.md`, `-new.md`, `-updated.md`) — consolidate into single authoritative source

## Step 6 — Size and shape

| Doc type | Target | Hard limit |
|----------|--------|------------|
| Individual doc | < 500 lines | 800 lines |
| README.md in a sub-dir | < 200 lines | 300 lines |
| Master `docs/README.md` | < 150 lines (it's an index) | 250 lines |

Flag oversized docs for splitting.

## Step 7 — Orphan directories

- Directories with only archived content (should be moved to `docs/archive/`)
- Empty directories
- Directories without a `README.md` (convention violation)
- Nested directories beyond 3 levels deep (restructure)

## Step 8 — Code references

For each doc, spot-check (don't brute-force all):
- Referenced file paths exist (`src/…`, `services/…`)
- Referenced function names still exist (Grep)
- Referenced database tables still exist
- Referenced API endpoints still active

Flag broken references. Priority: docs that are canonical references for a feature (e.g., `CURRENT-TWO-TIER-ARCHITECTURE.md`) get stricter checks.

## Step 9 — Canonical source validation

For key docs flagged as "source of truth" in CLAUDE.md:
- `docs/<subsystem>/ARCHITECTURE.md`
- `docs/chat/v4/README.md`
- `docs/stripe/README.md`
- `docs/deployment/cloud-build-migration.md`
- `docs/mig/README.md`
- `docs/n8n/architecture.md`
- `docs/security/CSP-Policy-Guide.md`

Each must be up-to-date, self-contained, and free of broken refs.

## Step 10 — File findings to the board

Create parent tasks per severity + subtasks on project `<YOUR_AUDIT_PROJECT_ID>`. Assigned agent: `docs-manager`.

## Step 11 — Summary

Return ≤ 100 words:

```
DOCS REVIEW COMPLETE
- [N] docs reviewed ([X] in archive — skipped)
- [C/H/M/L] findings per severity → board parent tasks
- [O] orphan docs
- [D] duplicate-content pairs
- [B] broken internal links
```

## Anti-patterns to flag

- Docs that reference removed features as current
- `-v2.md`, `-new.md`, `-updated.md` file naming
- README-less subdirectories
- Docs > 800 lines without a table of contents
- Index entries that don't match the linked doc's title
- "Coming soon" / "TODO" markers older than 6 months

## Tools used

- Read, Glob, Grep
- Bash (restricted): `git log -1 --format=%cs -- <file>` for last-modified dates
- Board MCP (create_task, update_task, log_activity)

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.