/meta-report — Meta-Layer Audit & Report

Generate a comprehensive report on the Claude Code meta-layer for this project and write it to .claude/METAPROCESS.md, overwriting the previous report.

Execution Flow

Phase 0: Usage Data Collection

1. •

   Gather usage metrics from two sources before analysis begins:

   a) Tool usage log (.claude/usage.log):

   • •Read .claude/usage.log if it exists (tab-separated: timestamp, tool_name, path/command)
   • •Compute per-tool invocation counts (e.g., Read: 342, Edit: 87, Grep: 56)
   • •Compute per-file access counts (most-read files, most-edited files)
   • •Compute per-MCP usage counts (group mcp__context7__*, mcp__claude-in-chrome__*, etc.)
   • •Identify skill invocations by looking for Read calls to SKILL.md files followed by the skill's allowed tools
   • •Note the log's date range (earliest → latest entry) and total entry count
   • •If the log doesn't exist or is empty, note "Usage logging enabled but no data yet"

   b) Git history for meta-layer files:

   • •Run git log --oneline --follow -- <file> for each meta-layer file (CLAUDE.md, rules, skills, settings, memory files)
   • •Count commits touching each file and note last-modified date
   • •This captures update frequency even before the usage log existed

   Store these metrics for use in Sections 1–4 and 7 of the report.

Phase 1: Inventory & Analysis

1. •

   Scan all meta-layer files. Read every file in:

   • •.claude/ (settings, rules, skills, plans, METAPROCESS.md)
   • •CLAUDE.md at project root and any parent CLAUDE.md files
   • •~/.claude/projects/.../memory/ (auto-memory files)
   • •.cursor/ if present (rules, plans)
   • •Note file sizes (line counts), purposes, and relationships
2. •

   Analyze each component. For every file/config found:

   • •Summarize its purpose and content
   • •Assess quality (grade A through D) against best practices
   • •Note strengths and remaining issues
   • •Check for staleness, duplication, drift between tools
3. •

   Audit MCP servers & plugins. Review the active MCP configuration:

   • •List all configured MCP servers (from settings or context)
   • •For each: what it does, how often it's used, whether it's well-suited
   • •Identify gaps — things we do manually that an MCP could automate
   • •Note any redundancy between MCPs
4. •

   Analyze meta-skills. Review all skills in .claude/skills/:

   • •For each skill: purpose, usage frequency (if trackable), quality
   • •Specifically track /retro skill existence and usage evolution
   • •Specifically track /meta-report (this skill) — self-referential analysis
   • •Note skill gaps — common workflows that lack a skill

Phase 2: External Research (skip if argument is "quick")

1. •

   Web search: cutting-edge Claude Code techniques. Search for:

   • •"Claude Code advanced techniques 2026"
   • •"Claude Code skills best practices"
   • •"Claude Code MCP servers popular"
   • •Synthesize findings into actionable recommendations
   • •Focus on proven techniques in real-world use, not speculation
2. •

   Web search: MCP ecosystem. Search for:

   • •Popular/useful MCP servers for development workflows
   • •Compare against our current MCP setup
   • •Identify new opportunities that fit this project's needs (Qwik, TypeScript, offline-first, maintenance management)

Phase 3: Report Generation

1. •Write METAPROCESS.md with these sections (in order):

# Meta-Layer Audit Report
*Generated: [date]*

## 1. Inventory
Table of all meta-layer files with purpose, line count, last-modified date, git commits, and access count from usage log.

## 2. Component Assessments
### 2.1 CLAUDE.md — Grade: [X]
### 2.2 Auto-Memory — Grade: [X]
### 2.3 Rules — Grade: [X]
### 2.4 Skills — Grade: [X]
### 2.5 Settings & Hooks — Grade: [X]
### 2.6 Cursor Configuration — Grade: [X]
### 2.7 Cross-Tool Coherence — Grade: [X]

## 3. MCP & Plugin Audit
Current MCPs, usage counts from log (mcp__*__ tool calls), gaps, redundancies, recommendations.

## 4. Meta-Skills Analysis
### 4.1 /meta-report (this skill)
Self-referential: how this report process is working, what changed since last run.
### 4.2 /retro
Usage count, evolution, value delivered. (Zero baseline if not yet created.)
### 4.3 /offline-test
### 4.4 [any other skills]

## 5. Cutting-Edge Techniques
Researched techniques from the Claude Code ecosystem.
What's proven and adopted vs. experimental.
Applicability to this project.

## 6. Anti-Patterns & Technical Debt
Known issues, staleness, drift, duplication.
Resolved items (strikethrough).

## 7. Usage Statistics
Summary tables from Phase 0 data:
- Top 10 tools by invocation count
- Top 10 most-accessed files
- MCP usage breakdown
- Log date range and total entries
- Git commit counts for meta-layer files
If usage log has no data yet, show "Logging active since [date], no data accumulated yet."

## 8. Summary Scorecard
Table: Component | Grade | Status | Change from Last Report

## 9. Meta-on-Meta
Commentary on the report itself:
- How has the report format evolved?
- What new sections were added this run and why?
- Recommendations for the next report cycle.

Phase 4: Verification

1. •Confirm the report by reading back .claude/METAPROCESS.md and verifying all sections are present.
2. •Notify the user with a brief summary of key findings and grade changes.

Quality Standards

• •Be honest in assessments — don't inflate grades
• •Grade scale: A (excellent), B (good, minor issues), C (functional, notable gaps), D (needs rework)
• •Use "+/-" modifiers for nuance (A-, B+, etc.)
• •When comparing to last report, note improvements and regressions
• •Keep the report under 300 lines — dense and useful, not padded
• •Every recommendation should be actionable