You've learned about rules vs skills, the alwaysApply tax, and discovery mechanics. Now what? This post provides a step-by-step audit process for cleaning up an existing AI configuration. The goal: fewer rules, faster responses, better AI behavior.

Related: This is a practical follow-up to the Configuring Your AI Assistant series.

The Audit Framework

Phase Time Goal
Inventory 30 min Count and categorize
Evaluate 1-2 hours Identify what should change
Migrate 2-4 hours Execute the changes
Verify 30 min Confirm improvements

Block an afternoon. You'll thank yourself later.

Phase 1: Inventory

Step 1.1: Count Everything

# Rules
find .cursor/rules -name "*.md" | wc -l

# Skills
find .cursor/skills -type d -mindepth 1 | wc -l

# Commands (if using)
find .cursor/commands -name "*.md" 2>/dev/null | wc -l

# Agents (if using)
find .cursor/agents -name "*.md" 2>/dev/null | wc -l

Step 1.2: Measure alwaysApply Load

# Count alwaysApply rules
grep -rl "alwaysApply: true" .cursor/rules | wc -l

# Total lines in alwaysApply rules
grep -rl "alwaysApply: true" .cursor/rules | xargs wc -l | tail -1

Benchmarks:

  • Healthy: < 1,000 lines always loaded
  • Concerning: 1,000-2,000 lines
  • Critical: > 2,000 lines

Step 1.3: Find Oversized Rules

# List rules by size
find .cursor/rules -name "*.md" -exec wc -l {} + | sort -rn | head -20

Benchmarks:

  • Good: < 100 lines
  • Large: 100-200 lines
  • Oversized: > 200 lines (candidate for splitting)

Step 1.4: Document Current State

Create an inventory file:

# AI Config Audit - [Date]

## Inventory

| Category                | Count |
| ----------------------- | ----- |
| Rules                   | X     |
| Skills                  | X     |
| Commands                | X     |
| Agents                  | X     |
| alwaysApply rules       | X     |
| Total alwaysApply lines | X     |
| Oversized rules (>200)  | X     |

## Largest Rules

1. rule-name (X lines)
2. ...

Phase 2: Evaluate

Step 2.1: Audit Each alwaysApply Rule

For each rule with alwaysApply: true, ask:

Question If Yes If No
Needed in EVERY conversation? Keep alwaysApply Convert to globs
Is it a workflow (step-by-step)? Extract to skill Keep as rule
Could it be glob-triggered? Add globs, remove alwaysApply -
Is it > 100 lines? Split or extract -

Step 2.2: Create Migration Plan

Document what needs to change:

Rule Current Action Target
core-logging alwaysApply, 240 lines Add globs Glob-triggered
core-diagrams alwaysApply, 80 lines Add globs Glob-triggered
core-workflow alwaysApply, 89 lines Keep alwaysApply (core)
content-gen alwaysApply, 150 lines Extract workflow Rule + Skill

Step 2.3: Evaluate Skills

For each skill, check:

  1. Description quality: Does it have rich triggers?
  2. Type: Is it actually a skill (workflow) or a template/reference?
  3. Discovery: Can you trigger it with natural language?
Skill Description Quality Is Workflow? Action
/ship Good Yes Keep
/comic-hero Poor No (template) Consider consolidating
/init-project Good Yes Keep

Phase 3: Migrate

Step 3.1: Convert alwaysApply to Globs

Before:

---
description: "Logging standards"
alwaysApply: true
---

After:

---
description: "Logging standards"
globs:
  - "**/log*"
  - "**/logger*"
  - "**/*logging*"
alwaysApply: false
---

Step 3.2: Extract Workflows to Skills

If a rule has step-by-step instructions:

  1. Create new skill: skills/new-skill/SKILL.md
  2. Move workflow content to skill
  3. Keep only reference content in rule
  4. Add rule reference to skill: "See @rule-name for standards"

Before (in rule):

## How to Generate Content

1. First, check the brief
2. Then, draft the outline
3. Review against standards
4. Finalize and submit

After (split):

Rule keeps standards:

## Content Standards

- Title ≤42 characters
- Must have counterargument
- Enterprise examples required

Skill gets workflow:

# Content Generation Skill

## Instructions

1. Check brief for requirements
2. Draft outline (see @content-standards)
3. Review against standards
4. Finalize and submit

Step 3.3: Split Oversized Rules

If a rule is > 200 lines:

Option A: Extract examples to docs

rules/my-rule/RULE.md (core guidance)
rules/my-rule/examples.md (examples, referenced)

Option B: Split by concern

rules/my-rule-core/RULE.md (always needed)
rules/my-rule-advanced/RULE.md (glob-triggered)

Option C: Move detail to skill references

rules/my-rule/RULE.md (core guidance)
skills/my-skill/references/detailed-guide.md (on demand)

Step 3.4: Improve Skill Descriptions

Before:

description: "Helps with project setup"

After:

description: |
  Initialize project structure. Create directories and config files.
  Use when user asks to: set up project, initialize repo, create structure.
  Triggers: "init project", "set up", "bootstrap", "create structure"

Step 3.5: Consolidate Redundant Skills

If you have multiple similar skills with different entry points, consider consolidating:

Before:

skills/comic-hero-prompt/
skills/social-card-prompt/
skills/icon-set-prompt/

After:

skills/image-handler/
├── SKILL.md (unified entry point)
└── templates/
    ├── comic-hero.md
    ├── social-card.md
    └── icon-set.md

Phase 4: Verify

Step 4.1: Re-measure

# Count alwaysApply rules again
grep -rl "alwaysApply: true" .cursor/rules | wc -l

# Total lines again
grep -rl "alwaysApply: true" .cursor/rules | xargs wc -l | tail -1

Step 4.2: Test Key Workflows

  1. Start a new conversation
  2. Ask simple question ("what's 2+2?")
  3. Time the response
  4. Ask domain-specific question
  5. Verify appropriate rules/skills activate

Step 4.3: Document Results

Update your audit file:

## Results

| Metric                 | Before | After | Change |
| ---------------------- | ------ | ----- | ------ |
| alwaysApply rules      | 22     | 8     | -64%   |
| alwaysApply lines      | 2,700  | 900   | -67%   |
| Oversized rules        | 5      | 1     | -80%   |
| Response time (simple) | ~3s    | ~1.5s | -50%   |

Maintenance Schedule

Weekly

  • Quick check: Any new alwaysApply rules added?

Monthly

  • Audit new rules for appropriate activation
  • Check for skill description quality

Quarterly

  • Full re-audit using this process
  • Review and remove unused rules/skills

Decision Trees

Should This Be alwaysApply?

Is it needed in EVERY conversation?
├── No → Use globs or skill
└── Yes
    ├── Is it about safety/security?
    │   └── Yes → alwaysApply ✓
    ├── Is it about AI persona/behavior?
    │   └── Yes → alwaysApply ✓
    └── Is it domain-specific?
        └── Yes → Use globs

Should This Be a Skill or Rule?

Does it have step-by-step instructions?
├── Yes → Skill
└── No
    ├── Is it reference material for editing?
    │   └── Yes → Rule with globs
    └── Is it universal standards?
        └── Yes → Rule (maybe alwaysApply)

How to Improve Skill Discovery?

Is description > 3 lines?
├── No → Add more trigger phrases
└── Yes
    ├── Does it include "Use when..."?
    │   └── No → Add use cases
    └── Does it include "Triggers: ..."?
        └── No → Add explicit triggers

Pre-Flight Checklist

Before Starting

  • [ ] Block 2-3 hours
  • [ ] Have terminal access
  • [ ] Back up current config (git commit)

Inventory Complete

  • [ ] Counted rules, skills, agents
  • [ ] Counted alwaysApply rules and lines
  • [ ] Listed oversized rules

Evaluation Complete

  • [ ] Reviewed each alwaysApply rule
  • [ ] Checked skill descriptions
  • [ ] Created migration plan

Migration Complete

  • [ ] Converted alwaysApply → globs (where appropriate)
  • [ ] Extracted workflows → skills
  • [ ] Split oversized rules
  • [ ] Improved skill descriptions

Verification Complete

  • [ ] Re-measured metrics
  • [ ] Tested key workflows
  • [ ] Documented results
  • [ ] Committed changes

Key Takeaways

  1. You can't optimize what you don't measure. Start with inventory.

  2. Every rule is a tax on every conversation. Be ruthless about alwaysApply.

  3. The best audit finds things to remove. Smaller is better.

  4. Schedule regular maintenance. Rules accumulate; performance degrades.

  5. Document your changes. Future you will thank present you.

Related: Rules vs Skills | The alwaysApply Tax | How Cursor Finds Skills