docs-auditor

name	docs-auditor
description	セッションtranscriptを分析してドキュメントの効果を評価する。 doc参照がAgentの行動を改善したか、Context浪費か、腐敗して有害かを判定。 last-validated がないドキュメントにもgit履歴ベースで対応。 HTML reportでper-doc健全性、Context budget、freshness、推奨アクションを表示。 Use when user says "audit docs", "docs audit", "run docs-auditor", "ドキュメント監査", "doc ROI", "ドキュメントの効果分析".
disable-model-invocation	true

Docs Auditor

Transcript-based documentation effectiveness analysis. Evaluates whether reading docs actually improves agent behavior or wastes context tokens. Generates an interactive HTML report with per-doc health, freshness dashboard, context budget, and actionable recommendations.

Prerequisites

• pip install tiktoken (optional — falls back to character-based estimation)
• No external API keys required. Analysis uses Claude sub-agents.

Workflow

Run all steps sequentially. The coordinator (you) manages data flow between scripts and sub-agents.

Step 1: Initial Questions

Ask the user for the report language using AskUserQuestion: "レポートの言語は？ (e.g. 日本語, English)" Default to the user's conversation language if not specified. Pass the language choice to all sub-agents as an instruction prefix: "Write all output text in [chosen language]."

Scope is always current project only. Do NOT ask.

Step 2: Workspace Setup & Data Collection

Set up a timestamped workspace:

RUN_ID=$(date +%Y-%m-%dT%H-%M-%S)
WORKSPACE=<project>/.claude/docs-report/${RUN_ID}
mkdir -p ${WORKSPACE}

Collect transcripts: