project-inventory

name: project-inventory metadata: compatible_agents: [claude-code] tags: [inventory, discovery, claude-code, ai, projects, reporting, dashboard, metrics] description: > Discovers and inventories all Claude Code and AI-integrated projects across specified directories. Scans for CLAUDE.md files, .claude/ directories, AI SDK imports, and MCP configurations, then generates structured reports in Markdown, JSON, and a self-contained HTML portfolio dashboard with AI collaboration metrics. Use this skill whenever the user asks to list all their Claude Code projects, discover AI-powered repos, audit which projects use Claude, generate a project inventory, build an AI portfolio dashboard, or export project data — including: "list semua project Claude Code", "scan for AI projects", "inventory my repos", "which projects use Claude", "export project list", "cari semua project yang guna AI", "senarai project Claude", "audit Claude Code usage", "buat dashboard AI", "generate AI portfolio", "show my AI stats".

Project Inventory — Claude Code & AI Project Discovery

Discovers, audits, and reports on all Claude Code and AI-integrated projects across your workspace. Generates human-readable Markdown, machine-readable JSON inventories, and a visual HTML portfolio dashboard with AI collaboration metrics.

Command Reference

/inventory scan

Scan one or more directories to discover projects that use Claude Code or AI integrations.

Detection Signals

Scan for these indicators (in priority order):

Steps

1. Ask user for directories to scan — prompt for one or more root directories (e.g., ~/Projects, ~/Packages, ~/Work)

2. Discover projects — for each directory, find all git repositories or directories containing a composer.json, package.json, Cargo.toml, go.mod, pyproject.toml, or similar project markers

3. Detect AI signals — for each discovered project, check for the detection signals listed above

4. Extract metadata — for each matching project:

   • Name: directory name
   • Path: absolute path
   • Description: from composer.json description, package.json description, README first line, or CLAUDE.md first paragraph
   • URL: from git remote origin URL
   • Organization: extracted from git remote URL (GitHub org/user)
   • Has CLAUDE.md: boolean
   • AI signals found: list of detected signals
   • Tech stack: detected from project files (Laravel, React, Go, Python, etc.)

5. Present summary — show count and grouped list to user for review before generating reports

Example Output

Found 48 projects with AI integration:

~/Projects — 30 projects
  2025/ — 11 projects
  2026/ — 6 projects
  nadi-pro/ — 11 projects

~/Packages — 18 projects
  claude-tools/ — 4 projects
  laravel-packages/ — 14 projects

/inventory report

Generate structured inventory reports in Markdown and JSON formats.

Steps

1. Run scan if not already done — execute /inventory scan first

2. Ask for output directory — where to save the reports

3. Generate Markdown report (claude-code-projects.md):

   Follow this structure:

   # Claude Code Project Inventory
   
   > Generated: YYYY-MM-DD
   > Total: **N projects** across [directories listed]
   
   ---
   
   ## [Location 1] (count)
   
   ### [Category/Subfolder]
   
   | # | Project | Description | URL |
   |---|---------|-------------|-----|
   | 1 | **project-name** | Short description | https://github.com/org/repo |
   
   ---
   
   ## Summary
   
   | Metric | Count |
   |--------|-------|
   | **Total projects** | **N** |
   | Projects with CLAUDE.md | N |
   | Projects with .claude/ only | N |
   
   ### By Organization
   | Organization | Count |
   |-------------|-------|
   | org-name | N |
   
   ### By Tech Stack
   | Stack | Projects |
   |-------|----------|
   | Laravel | ~N |
   

4. Generate JSON report (claude-code-projects.json):

   Array of objects with this schema:

   [
     {
       "name": "project-name",
       "path": "/absolute/path/to/project",
       "description": "Short project description",
       "url": "https://github.com/org/repo",
       "location": "Projects",
       "category": "Category Name",
       "hasClaudeMd": true,
       "org": "github-org-name",
       "techStack": "Laravel",
       "aiSignals": ["CLAUDE.md", ".claude/"]
     }
   ]
   

   Rules:

   • url: set to null if no git remote configured
   • org: set to null if local-only project
   • hasClaudeMd: boolean based on file existence
   • aiSignals: array of all detected signals

5. Validate outputs:

   • Verify JSON is valid: python3 -m json.tool <file> > /dev/null
   • Verify project count matches scan results
   • Report file paths and sizes to user

/inventory audit

Audit Claude Code adoption quality across discovered projects.

Steps

1. Run scan if not already done

2. Check each project for:

   • Has CLAUDE.md (not just .claude/ directory)
   • CLAUDE.md has meaningful content (not empty/boilerplate)
   • Has .claude/settings.json with configured permissions
   • Has MCP servers configured
   • Has project-specific instructions (not generic template)

3. Generate audit report:

   ## Claude Code Adoption Audit
   
   ### Fully Configured (CLAUDE.md + settings)
   - project-a, project-b, ...
   
   ### Partial (CLAUDE.md only)
   - project-c, project-d, ...
   
   ### Minimal (.claude/ directory only)
   - project-e, project-f, ...
   
   ### Recommendations
   - N projects missing CLAUDE.md — add project instructions
   - N projects have empty CLAUDE.md — add stack, conventions, gotchas
   - N projects missing .claude/settings.json — configure permissions
   

4. Offer to scaffold — ask if user wants to generate template CLAUDE.md files for projects that are missing them

/inventory dashboard

Generate an AI collaboration portfolio dashboard — a self-contained HTML file with embedded metrics, charts, and interactive features. Requires a claude-code-projects.json from /inventory report as input.

Steps

1. Locate or generate project JSON — check if claude-code-projects.json exists in the output directory. If not, run /inventory report first.

2. Ask user for output directory — where to save the dashboard files

3. Optionally expand project list from GitHub orgs — ask user if they want to scan GitHub organizations for additional repos not in the local scan. Use gh repo list <org> --limit 200 --json name,description,url,primaryLanguage,isArchived,isFork,pushedAt to discover repos, filter by pushedAt date (e.g., active since 2025), exclude archived and forked repos, merge with existing claude-code-projects.json, and deduplicate by URL and name.

4. Generate data collection script (collect-ai-metrics.sh):

   Create a bash script following the template in references/collect-metrics-template.md. The script:

   • Reads claude-code-projects.json as input
   • For each project with a local .git/ directory, extracts metrics via git log
   • For projects with a GitHub URL but no local repo, falls back to gh api
   • Detects AI tools by email pattern in Co-Authored-By trailers
   • Counts per-tool commits (Claude, OpenCode, Copilot)
   • Handles duplicate GitHub repos (e.g., two projects pointing to same repo)
   • Deduplicates projects that exist both locally and remotely (prefers local)
   • Skips projects with neither local repo nor GitHub URL
   • Outputs ai-portfolio-data.json with per-project and aggregate metrics

   Per-project metrics:

   Metric	Source
   total_commits	git rev-list --all --count or GitHub API
   co_authored_commits	Commits with Co-Authored-By matching AI tool emails (case-insensitive)
   co_author_percentage	Calculated
   first_commit_date / last_commit_date	Earliest & latest commit dates
   first_ai_commit_date	Earliest co-authored commit date
   claude_models_used	Parsed from Co-Authored-By: Claude <model> trailer
   ai_tools	Array of detected AI tools (Claude, OpenCode, Copilot)
   claude_commits / opencode_commits / copilot_commits	Per-tool commit counts
   authors_count	Unique committer emails
   tech_stack	Detected from project files or GitHub languages API
   monthly_commits / monthly_ai_commits	{month, count} arrays

   AI tool detection by email pattern:

   Email	Tool
   noreply@anthropic.com	Claude (Claude Code)
   opencode@anthropic.com	OpenCode
   Copilot@users.noreply.github.com	GitHub Copilot

   Aggregate metrics:

   • Totals (projects, commits, AI commits, percentage)
   • Projects by organization and category
   • Tech stack distribution
   • Monthly timeline (total + AI commits per month)
   • AI adoption timeline (sorted by first AI commit date)
   • Claude model usage with project lists
   • AI tool breakdown (per-tool commits and project lists)

   Edge cases:

   • Skip missing directories gracefully
   • Detect AI tools by Co-Authored-By email pattern (not name)
   • Count unique commit hashes (not trailer lines — a commit can have multiple trailers)
   • Deduplicate projects that exist both locally and on GitHub (prefer local)
   • GitHub search API has 30/min rate limit — use single combined search call per repo with 3s delay
   • Handle API rate limit errors gracefully (default to 0)

5. Run the collection script — execute and verify output is valid JSON

6. Generate HTML dashboard (ai-portfolio.html):

   Create a self-contained HTML file following the template in references/dashboard-template.md. Embed the JSON data inline via <script>const DATA = {...};</script>.

   Dashboard sections:

   #	Section	Visualization
   1	Hero Header	Gradient bg, dynamic count-up stats from DATA
   2	Executive Summary	6 stat cards in 3x2 grid
   3	AI Activity Heatmap	GitHub-style 365-day calendar heatmap
   4	AI Adoption Trend	Dual-line chart (AI commits + cumulative projects)
   5	AI Adoption Timeline	Swim-lane chart (project bars from first AI commit to last)
   6	Monthly Activity	Stacked bar chart (AI vs human commits)
   7	Projects by Category	Grouped card grids
   8	Organization Breakdown	Horizontal bar chart
   9	Tech Stack Distribution	Horizontal bars + CSS donut chart
   10	AI Collaboration Leaderboard	Ranked table with AI tool badges, progress bars, medals
   11	AI Tools Breakdown	Per-tool cards with commit counts + CSS donut chart
   12	Claude Model Usage	Cards per model with project tag lists
   13	Project Detail Grid	Searchable, filterable, sortable table with AI Tools column
   14	Footer	Generated date, "Powered by Claude Code" badge

   Important implementation notes:

   • Hero stats MUST be dynamic — read values from DATA.aggregates via JS, never hardcode
   • All data-count attributes should be set from DATA at runtime
   • Project Detail Grid section should have class="reveal visible" (always visible, not scroll-dependent)
   • Table elements need explicit color:var(--fg) to prevent dark-on-dark in dark mode
   • Select dropdowns need option { background:var(--bg2); color:var(--fg) } for dark mode

   Design system:

   • Dark-first theme (#0B1120), light mode toggle with localStorage
   • Fonts: Space Grotesk (display), JetBrains Mono (mono), Inter (body) via Google Fonts
   • Accent colors: Amber #F59E0B, Green #10B981, Blue #3B82F6, Purple #8B5CF6, Cyan #06B6D4, Pink #EC4899
   • Layout: max-width: 1200px, CSS Grid, responsive at 768px/480px
   • Charts: CSS-only (calculated widths, conic-gradient donut, grid swim-lanes)
   • Heatmap: Full 365-day calendar grid with 5-level green intensity scale
   • Animations: IntersectionObserver scroll reveal, count-up hero numbers
   • Print: @media print forces light theme, hides interactive controls
   • Zero external JS dependencies — all CSS/JS inline

   JS features (all inline):

   • Theme toggle with localStorage persistence
   • Count-up animation on hero stats
   • Search/filter/sort on project grid
   • IntersectionObserver scroll reveal
   • Tooltip on heatmap cells

7. Verify dashboard:

   • Open in browser — all sections render
   • Dark/light toggle works
   • Search/filter in project grid works
   • Heatmap shows full year
   • Print preview looks clean

Output Files

Tech Stack Detection

Use these file indicators to detect tech stack:

Organization Extraction

Extract organization from git remote URL:

https://github.com/ORG/repo.git → ORG
git@github.com:ORG/repo.git → ORG
https://gitlab.com/ORG/repo.git → ORG
https://custom.domain/ORG/repo → ORG

If no remote is configured, set org to null.

Notes

• Always ask the user which directories to scan — never assume paths
• Never hardcode personal information (names, emails, paths) — keep everything dynamic
• Projects without any AI signals are excluded from the inventory
• The scan is non-destructive — it only reads files, never modifies them
• For large directory trees, show progress updates during scanning
• Group projects logically by location, then by category/subfolder
• Include both CLAUDE.md and .claude/ directory projects in the count
• JSON output should be valid and parseable by standard tools
• Dashboard HTML must be self-contained — no external JS/CSS dependencies
• Embed data inline in the HTML so it works offline and can be shared as a single file
• All paths in generated files should use variables, not hardcoded absolute paths

Reference Files