Skip to main content
skill CX Codex

scorecard — Claude Code skill: CLAUDE.md compliance audit

scorecard renders a worst-first card of your CLAUDE.md rules: which ones Claude actually follows while editing, which ones it keeps breaking.

MIT ★ 586 updated 2026-10-26

Install into Claude Code · Plugin marketplace All methods and platforms ↓
/plugin marketplace add karanb192/claude-code-hooks
/plugin install dead-rules-audit@claude-code-hooks
Open on GitHub

What scorecard does

The skill prints a compliance card for CLAUDE.md. Every rule gets a line: how often it mattered for an edit, how often it was broken, and a heuristic compliance figure. Worst offenders sit at the top, and rules Claude keeps ignoring carry a ⚠ promote→hook flag.

The card is rendered by dead-rules-audit.js with --render. The numbers behind it come from the dead-rules-audit hook, which records during your sessions; the skill groups them across recent work. After that it tells Claude to point out 1-3 of the worst rules and, for flagged ones, to suggest a deterministic check instead of memory: a PreToolUse or PostToolUse hook, or a linter rule.

Two things have to be in place. The render step calls node, so Node.js must be available. The hook also has to be installed, because it owns the history and the skill only displays it. Invocation is manual — the skill file sets disable-model-invocation: true.

What it helps with

  • see which CLAUDE.md rules Claude has stopped following
  • find rules that belong in a hook rather than in a prompt
  • compare rule compliance across recent sessions
  • drop CLAUDE.md lines that no longer affect edits
  • show where the instruction file fails in practice

Example prompts for the agent

  • /scorecard — show me the CLAUDE.md compliance card
  • /scorecard, then name the three worst rules
  • /scorecard and tell me which rules should become hooks

How to install scorecard

1Plugin marketplace

/plugin marketplace add karanb192/claude-code-hooks
/plugin install dead-rules-audit@claude-code-hooks

This installs the whole dead-rules-audit plugin with all its skills

2Skills CLI (Vercel)

npx skills add karanb192/claude-code-hooks --skill scorecard -a claude-code

3Manually: put the files in place

~/.claude/skills/scorecard/SKILL.md

Copy the whole skill folder; for one project use .claude/skills/scorecard/

Step by step, with commands for each platform: How to install skills in Claude Code: five ways

Questions about scorecard

How do I install scorecard in Claude Code?

From the marketplace: run /plugin marketplace add karanb192/claude-code-hooks, then /plugin install dead-rules-audit@claude-code-hooks. The skill ships inside the dead-rules-audit plugin.

What does scorecard do?

It shows a card for your CLAUDE.md rules with how often each was relevant, how often violated, a heuristic compliance percentage, and a ⚠ promote→hook flag on rules Claude chronically ignores.

Do I need the dead-rules-audit hook for it?

Yes, otherwise the card has nothing to show. The hook records edits and violations as you work; the skill only renders what it collected.

Source text

SKILL.md · MIT
---
name: scorecard
description: Show the CLAUDE.md compliance scorecard: which rules Claude actually follows vs ignores
disable-model-invocation: true
---

# CLAUDE.md compliance scorecard

!`node "${CLAUDE_SKILL_DIR}/../../dead-rules-audit.js" --render 2>/dev/null || node "${CLAUDE_PLUGIN_ROOT}/dead-rules-audit.js" --render`

The card above is a worst-first audit of your CLAUDE.md rules, aggregated across
your recent sessions: how often each rule was relevant to an edit, how often it was
violated, its heuristic compliance %, and a `⚠ promote→hook` flag for rules Claude
chronically ignores. It is produced by the `dead-rules-audit` hook, which records
as you work.

Briefly call out the 1-3 worst offenders and, for any rule flagged
`promote→hook`, suggest making it deterministic (a PreToolUse/PostToolUse hook or
a lint rule) instead of relying on Claude to remember it. Keep it short; do not
re-run any commands.

© Karan Bansal · MIT · karanb192/claude-code-hooks

Similar skills

All skills for Agent setup: memory, context, orchestration →

From the same set28

The whole set →

Project sponsors

Companies that keep this analytics open