本文へ移動
cccskills
無料GitHub で公開

audit-docs

Audits internal (dev/) and external (docs/) documentation completeness for a feature, subject, or set of existing docs, maps changes indicated by the user, across Infrahub's documentation layers, reports gaps, and optionally applies the fixes. TRIGGER when: the user wants to audit or check documentation coverage, find doc gaps after a feature branch, or verify docs are still current for a subject or specific files. DO NOT TRIGGER when: authoring new documentation from scratch → use the add-docs flow; only linting/formatting Markdown → run `uv run invoke docs.lint`.

インストール方法を見る

含まれるファイル(1)

  • SKILL.md5.9 KB

SKILL.md(原文)

インストールする前に、エージェントに与えられる指示の中身を確認できます。

Documentation Audit

User Input

$ARGUMENTS

What this does

Audit documentation completeness — after a feature branch, for a subject, or across a set of existing docs. Scan the relevant changes, cross-reference them against every Infrahub documentation layer, and report the gaps. Then, only with the user's approval, apply the fixes. The goal is coverage without over-documentation: a gap is only a gap if a reader is actually left without something they need.

Workflow

1. Gather scope

Treat $ARGUMENTS as what to audit. It will be one of:

  • A commit range or branch name — audit all changes in that range. Run git log <range> --oneline --stat to understand every change. The input in order of importance is:
    • PR description (if the branch corresponds to a PR or if a PR is provided)
    • Pull Request review comments, resolved and unresolved. Usually this is GOLD information to be included in the internal docs (if the branch corresponds to a PR or if a PR is provided)
    • Any changelog fragment and spec link. If missing, flag it as a gap.
    • The source code diffs themselves.
  • A subject (e.g. "webhooks", "computed attributes", "IPAM") — audit documentation for that topic across the codebase, regardless of branch. Use Grep/Glob to find all related code and docs, then assess coverage against dev/knowledge/, docs/docs/, dev/specs/, the component AGENTS.md files, and the code itself.
  • A set of doc paths (e.g. docs/docs/guides/installation.mdx dev/knowledge/backend/templates.md) — audit only those files. Read each one, identify the feature/topic it covers, then search the codebase for the corresponding implementation to confirm the doc is current and complete, and check whether the code has drifted from what the doc describes.

If $ARGUMENTS is empty, ask what to audit — a commit range, branch name, subject, or list of doc paths (default: current branch vs stable).

2. Map changes to documentation layers

For each changed area, check all five documentation layers:

LayerLocationQuestion
Technical referencedev/knowledge/backend/ or dev/knowledge/frontend/Does the knowledge doc explain how this works?
User-facing docsdocs/docs/topics/ or docs/docs/guides/Can users understand and use this feature?
Feature specdev/specs/Is there a spec, and does it match what was built?
Changelogchangelog/Is there a changelog fragment for user-visible changes?
Cross-referencesAll docsDo related docs link back in both directions?

Also check these secondary locations when relevant:

  • dev/knowledge/backend/architecture.md — component map, if new directories were created.
  • dev/knowledge/backend/testing.md — if new test patterns were introduced.
  • dev/knowledge/backend/schema-definitions.md — if new schema types were added.
  • The AGENTS.md of the component that changed — if new top-level modules were created.

3. Generate the audit report

Present findings in this format:

## Documentation Audit Report

### Scope
<!-- What was scanned: branch/range, subject, or list of doc files -->

### Changes Summary
<!-- Grouped by area: schema, logic, tests, etc. -->

### Documentation Status

For each document checked:

- **File**: path
- **Status**: Current / Outdated / Missing / N/A
- **Details**: What is good, what is missing

### Gaps Found

For each gap:

- **What is missing**: Description
- **Where**: Which file to update or create
- **Severity**: High / Medium / Low
- **Suggested fix**: Concrete content or edit

### Not Gaps (By Design)

Things that look like gaps but are intentionally absent. This section prevents over-documentation.

Common reasons something is not a gap:

- Feature uses standard mechanisms (GraphQL mutations, UI forms) that do not need a dedicated guide.
- Entity is fully documented within a parent feature's knowledge doc.
- An ADR is unnecessary when the spec already captures design decisions.

4. Apply fixes

Ask the user which fixes to apply before editing anything:

  1. Apply all — edit or create all proposed documentation.
  2. Cherry-pick — let the user select which changes to apply.
  3. None — keep the report as reference only.

For approved fixes:

  • Prefer editing existing files over creating new ones.
  • Check cross-references in both directions (A links to B and B links to A).
  • Run uv run invoke docs.lint to perform global Markdown/MDX linting across the repo.

Guardrails

  • Never edit docs before the user has approved specific fixes — the report comes first; writing is a separate, opt-in step (Step 4).
  • Report the "Not Gaps" honestly — an audit that flags everything as missing is as useless as one that misses real gaps. Justify why intentional absences are fine.
  • Ground every gap in a concrete reader need — say who is left without what, and point to the exact file to change. No vague "could use more docs".
  • Prefer editing over creating — a new file is only justified when no existing doc is the right home for the content.

レビュー

まだレビューはありません。使ってみた感想をお寄せください。

同じリポジトリのスキル

概要と使いどころ

Analyzes recent CI failures on pull requests to identify flaky tests, using retry outcomes (failed attempt → green re-run) and cross-PR recurrence as evidence, and maintains a local longitudinal ledger so flakiness can be tracked over time. TRIGGER when: the user wants to find flaky tests, correlate recent CI failures, check which tests fail across PRs or recover on retry, or refresh the flakiness trend report. DO NOT TRIGGER when: babysitting a single PR's CI until green → monitoring-pull-requests; diagnosing or fixing one specific failing test → the bug-analysis skills.

日本語の概要は準備中です。原文の説明を表示しています。

opsmill/infrahub5342026年10月10日 更新

commit

無料

Stages and commits the current changes onto a safe working branch, enforcing branch discipline and optionally pushing upstream. TRIGGER when: the user wants to commit, save, or check in the current changes. DO NOT TRIGGER when: opening a pull request → pr.

日本語の概要は準備中です。原文の説明を表示しています。

opsmill/infrahub5342026年10月10日 更新

Use when you've fixed a bug, added a feature, or made any user-facing change in a project that uses Towncrier and need to record it for the changelog — before committing or opening a PR, or whenever asked to add a changelog entry, towncrier fragment, or news fragment.

日本語の概要は準備中です。原文の説明を表示しています。

opsmill/infrahub5342026年10月10日 更新

Turns a single feature idea, improvement, or bug into ONE well-structured GitHub issue. TRIGGER when: the user wants to file/open/create an issue, turn a feature idea or improvement into a ticket, or capture something missing or broken as a ticket. DO NOT TRIGGER when: breaking work into multiple issues or planning a body of work → a planning skill; writing a full Product Requirements Document → creating-prd; the idea is still fuzzy and unhardened → grilling-ideas first.

日本語の概要は準備中です。原文の説明を表示しています。

opsmill/infrahub5342026年10月10日 更新

Synthesises the current conversation context into a Product Requirements Document and publishes it to GitHub (as a comment on a referenced issue, or a new issue). Synthesises from context; does not interview. TRIGGER when: the conversation has produced enough understanding of a feature and the user wants it captured as a PRD. DO NOT TRIGGER when: a single small issue is enough → creating-issues; the idea has not been stress-tested yet → grilling-ideas first; bug reports.

日本語の概要は準備中です。原文の説明を表示しています。

opsmill/infrahub5342026年10月10日 更新

Stress-tests a fuzzy or vague feature idea before any PRD, spec, or ticket is written. TRIGGER when: the user has a fuzzy feature idea — one or two paragraphs, vague on users / scope / success — and wants to harden it, or says "grill / stress-test / pressure-test this idea." DO NOT TRIGGER when: the idea is already turned into a spec or PRD; bug fixes or refactors; the idea is hardened and you are ready to write the PRD → creating-prd.

日本語の概要は準備中です。原文の説明を表示しています。

opsmill/infrahub5342026年10月10日 更新

opsmill のスキルをすべて見る

このスキルの問題を報告する