repoloom

Analyze your project and install the right AI agent skills and plugins — automatically, from skillsllm.com.

npx repoloom analyze

What it does

repoloom fingerprints your project (stack, frameworks, test tools, CI), fetches the live skills catalog from skillsllm.com, and uses Claude to score each skill on incremental value — what gap does this fill given what you already have? Only skills that genuinely improve your setup get recommended.

Works with Claude Code, Cursor, GitHub Copilot, Codex CLI, Gemini CLI, and more. Each platform gets skills installed in the right location automatically.

Installation

No install required — use via npx:

npx repoloom <command>

Or install globally for faster runs:

npm install -g repoloom

Commands

| Command | Description | |---------|-------------| | repoloom analyze | Fingerprint project, score skills, install selected ones | | repoloom sync | Re-install all skills from repoloom.lock | | repoloom list | Show installed skills (local and global) | | repoloom remove <skill> | Uninstall a skill by slug | | repoloom publish | Scaffold or publish a skill to npm |

analyze

Primary command. Detects your stack, fetches skills from skillsllm.com, scores each by incremental value, and installs what you choose.

npx repoloom analyze

Example output:

Platform: claude-code | Project: javascript+typescript react+express
Coverage: Has senior engineering and readthru skills. Gaps: E2E testing,
security hardening, React UI patterns, CI/CD.

8 skill(s) add real value (incremental ≥6/10):

  1. playwright-skill  ★2.9k  [testing]
     Fills complete E2E testing gap — vitest/jest alone can't cover browser flows.
     Incremental: █████████░ 9/10  Overall: 8.9/10
     R:9 U:9 Q:8

  2. agent-skills  ★77.7k  [ai-agents]
     24-skill bundle covering performance, security, code review, CI/CD, frontend.
     Incremental: █████████░ 9/10  Overall: 8.9/10
     R:9 U:9 Q:8

? Which skills to install? (→ .claude/skills/) › playwright-skill, agent-skills

✔ playwright-skill [plugin] → .claude/plugins/playwright-skill/
✔ agent-skills [bundle: 24 skills] → api-and-interface-design, security-and-hardening, ...

If your project is already well-covered, repoloom says so and exits:

Coverage: All major areas covered — no high-value additions found.
Your skill coverage is solid.

Set ANTHROPIC_API_KEY for LLM-ranked scores. Without it, falls back to tag-based matching.

Local vs Global install

By default, skills install locally to the current project:

.claude/
  skills/       ← SKILL.md-based skills
  plugins/      ← full plugin repos (have .claude-plugin/ manifest)

To install globally (available in all projects):

# Not yet a flag — install globally by pointing at your home dir
# Recommended: use local install per project and commit repoloom.lock

Skill vs Plugin distinction:

• Skill — repo contains SKILL.md. Installed to .claude/skills/{slug}/.
• Plugin — repo contains .claude-plugin/. Installed to .claude/plugins/{slug}/ (full repo copied — includes hooks, slash commands, settings).
• Bundle — repo has skills/*/SKILL.md. Each sub-skill extracted to .claude/skills/.

Team workflow with repoloom.lock

One teammate sets up skills, commits the lock. Everyone else syncs.

# Teammate 1 — runs analyze, installs, commits lock
npx repoloom analyze
git add repoloom.lock
git commit -m "chore: add repoloom skills"
git push

# Teammate 2 (Claude Code) — syncs from lock
npx repoloom sync

# Teammate 3 (Cursor) — syncs, gets skills in .cursor/rules/
npx repoloom sync

# Teammate 4 (GitHub Copilot) — syncs, appends to .github/copilot-instructions.md
npx repoloom sync

repoloom.lock format:

{
  "version": 2,
  "skills": {
    "playwright-skill": "https://github.com/lackeyjb/playwright-skill",
    "agent-skills": "https://github.com/addyosmani/agent-skills"
  }
}

Lock is platform-agnostic — each teammate's sync detects their tool and installs to the right location.

Platform install locations

| Platform | Skills location | Plugins location | |----------|----------------|-----------------| | Claude Code | .claude/skills/{slug}/SKILL.md | .claude/plugins/{slug}/ | | Cursor | .cursor/rules/{slug}.md | — | | GitHub Copilot | .github/copilot-instructions.md (appended) | — | | Codex CLI | .codex/skills/{slug}/SKILL.md | — | | Others | Manual — prints skillsllm.com link | — |

How it works

1. Fingerprint — detects languages, frameworks, test tools, CI, package managers, already-installed skills
2. Fetch — scrapes relevant categories from skillsllm.com (ai-agents, cli-tools, testing, etc.) — always fresh per run
3. Rank — Claude evaluates what's missing given installed skills. Scores: relevance, usefulness, quality, incremental_value. Only shows skills scoring ≥6/10 on incremental value.
4. Install — detects repo type (skill / plugin / bundle), installs to correct location for your platform, writes repoloom.lock

Setting up ANTHROPIC_API_KEY

Works without a key (tag-based matching), but LLM ranking is significantly smarter.

# Current session
export ANTHROPIC_API_KEY=sk-ant-your-key-here

# Permanent (zsh)
echo 'export ANTHROPIC_API_KEY=sk-ant-your-key-here' >> ~/.zshrc && source ~/.zshrc

# Permanent (bash)
echo 'export ANTHROPIC_API_KEY=sk-ant-your-key-here' >> ~/.bashrc && source ~/.bashrc

Requirements

• Node.js 20+
• git (for cloning skills from GitHub)
• ANTHROPIC_API_KEY (optional — enables LLM-ranked recommendations)

License

MIT