Eval Rules

Rules Evaluator

Discover all rule files, validate their structure and glob patterns against the real project, then run an interactive session to confirm (or improve) each rule with the user.

The goal is not just to score; it is to leave the rules directory in better shape than it was.

When to Use

• First time writing .claude/rules/ files (validate before committing)
• A rule seems to never trigger, or fires on every file
• Migrating @ imports from CLAUDE.md to path-scoped rules
• Periodic hygiene: "are these rules still relevant to how we work?"
• After onboarding to a new codebase

Key Concepts

| Mechanism | When it loads | Notes | |---|---|---| | @file in CLAUDE.md | Session start, always | Even inside a conditional sentence | | No paths: in rule | Session start, always | Same cost as @import | | paths: frontmatter | When Claude reads a matching file | Trigger = Read tool on a matched file | | User-level rule (~/.claude/rules/) | Session start, always | Applies to all projects on the machine |

The paths: field is the main lever for keeping rules contextual. An always-on rule with 80 lines loads on every session even if you're fixing a typo in README.md.

Glob patterns in paths: support brace expansion: "src/**/*.{ts,tsx}" matches both .ts and .tsx files with one entry.

---

Scoring Criteria (12 pts per rule)

| # | Criterion | Max | What is checked | |---|-----------|-----|-----------------| | 1 | frontmatter block | 1 | File has YAML frontmatter (--- delimited) | | 2 | paths: field | 2 | Present (1pt) + at least one pattern listed (1pt) | | 3 | pattern validity | 3 | Each pattern matches >= 1 file in project (up to 3 patterns checked) | | 4 | scope | 2 | Not dead (>= 1 match) + not too broad (= 10/12 (>= 83%)

• Needs work: 7-9/12 (58-82%)
• Fix: /dev/null | wc -l

If no `.claude/rules/` directory exists, report it and stop.

### Step 2: Parse each rule

For each `.md` file found:
1. Read the full file
2. Extract YAML frontmatter (content between first `---` and second `---`)
3. Parse `paths:` field: collect all glob patterns as a list
4. Classify: **conditional** (has `paths:`), **always-on** (no frontmatter or no `paths:`), or **user-level** (from `~/.claude/rules/`)
5. Read the body: line count, presence of a clear title/header, whether rules are specific

### Step 3: Resolve glob patterns

For each rule with `paths:`, use Glob to resolve each pattern against the project:

- Collect total matched files per pattern
- Show up to 10 sample paths
- Flag dead patterns (0 matches) and broad patterns (>30% of source files)

### Step 4: Interactive review (core of the skill)

Process rules **one by one**. Do not batch and skip the interaction.

**For each conditional rule (has `paths:`):**

Show:

Rule: payments.md [conditional] paths: ["/payments/", "src/billing/**"] Matches: 12 files

• src/payments/stripe.py
• src/payments/webhook.py
• src/billing/invoice.py

... (9 more)

Ask three questions:
1. "Is this scope right? (y = yes / n = needs adjustment)"
2. "Is this rule still useful day-to-day? (y / n / unsure)"
3. "Anything to add, remove, or update in the rule content? (describe or skip)"

**For each always-on rule (no `paths:`):**

Show:

Rule: database.md [always-on, loads every session] Content: 91 lines, 8 rules

Ask:
1. "This rule loads at every session. Should it stay always-on, or be scoped to specific files? (keep / scope / skip)"
2. "Is the content still accurate and useful? (y / n)"

If the user says **scope**: help them define a `paths:` pattern based on the rule content (e.g., database rules -> `**/models/**`, `**/migrations/**`, `**/*.sql`). Propose the frontmatter block and ask for confirmation before editing.

**If the user provides corrections or updates during the interaction**: apply them directly using Edit, confirm each change, then move to the next rule.

### Step 5: Output report

After all rules are reviewed:

Rules Audit: [project name or path]

Date: [today] | Scanned: N rules (X conditional, Y always-on)

Summary

| Status | Count | |--------|-------| | Good (>= 83%) | N | | Needs work (58-82%) | N | | Fix (= 1 file | | scope | 2/2 | 3.5%, well-scoped | | content quality | 2/3 | 158 lines (over 150) |

User feedback: scope confirmed ("fires exactly when we work on Stripe") Content: no changes needed

---

database.md (3/5) [always-on]

No paths: field; loads at every session.

| Criterion | Score | Notes | |-----------|-------|-------| | frontmatter | 0/1 | No frontmatter block | | content quality | 3/3 | 91 lines, specific rules, clear header | | focus | 1/1 | 8 rules |

User feedback: scope -> added paths: ["/models/", "/migrations/", "**/*.sql"] Edit applied

---

### Step 6: Fix Summary

What Changed This Session

database.md:

• Added frontmatter with paths: ["/models/", "/migrations/", "**/*.sql"]

testing.md:

• User confirmed useful, no changes

git-workflow.md:

• User flagged as stale (marked for deletion, not deleted yet, awaiting confirmation)

---

N rules audited / N edits applied / N rules flagged as stale

For any rule the user marked as stale or outdated: ask for explicit confirmation before deleting. Never delete without a clear "yes, delete it".

---

## Edge Cases

- **Subdirectory rules** (e.g., `.claude/rules/frontend/react.md`): discovered and processed normally
- **Symlinked files or directories** in `.claude/rules/`: supported by design, discovered and processed normally
- **Invalid YAML frontmatter**: report parse error, score frontmatter as 0/1
- **Empty file**: flag as broken, skip interactive step, ask if it should be deleted
- **Dead pattern** (0 matches): flag, suggest fix based on rule content
- **Pattern matching >30% of files**: flag as too broad, suggest narrowing
- **User says "unsure" about usefulness**: note it in the report, do not edit, add a comment "Review in next audit"
- **Rules not loading as expected**: advise the user to add the `InstructionsLoaded` hook, which logs exactly which rule files load and when

## Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [FlorianBruniaux](https://github.com/FlorianBruniaux)
- **Source:** [FlorianBruniaux/claude-code-plugins](https://github.com/FlorianBruniaux/claude-code-plugins)
- **License:** MIT

Install and usage instructions live in the source repository linked above.