Skill package manager for Claude Code. Discovers relevant skills and MCP servers for your projects.
Given your repository, agent-hunter identifies the top 3 relevant skills or MCPs, explains why they fit, and identifies unsafe ones to avoid.
$ agent-hunter hunt .
Found: FastAPI, PostgreSQL, pytest, Docker
Top 3 recommendations:
1. fastapi-backend-testing
Status: SAFE
Reason: Your repo uses FastAPI, pytest, and Docker. This skill provides
test and deployment workflow helpers.
2. postgres-mcp-server
Status: SAFE
Reason: PostgreSQL detected. Provides database query and schema tools.
3. api-security-scanner
Status: REVIEW
Reason: REST API security audits. Flagged for filesystem access.
Review SKILL.md before installing.
Blocked (security): 2 results
Requirements: Claude Code, Git, Python 3.10+
git clone --depth 1 https://github.com/indhra/agent-hunter.git ~/.claude/skills/agent-hunter
cd ~/.claude/skills/agent-hunter
./setupThe setup script:
- Verifies Python 3.10+
- Creates isolated virtual environment
- Registers agent-hunter in global ~/.claude/CLAUDE.md
- Makes /agent-hunter available in all Claude Code sessions
/agent-hunterOr:
agent-hunter hunt /path/to/projectEnable automatic hunting when opening new projects:
export AGENT_HUNTER_AUTO=1Add to ~/.zshrc or ~/.bash_profile for permanence.
See INSTALL.md for full proactive setup options.
Open any repo and type:
/agent-hunter
Or ask Claude:
What skills or tools should I install for this project?
agent-hunter will:
- Read your project's tech stack (framework names only - no code, no paths)
- Hunt curated skills + GitHub results
- Security-scan every result
- Show top 3 recommendations with clear explanations
- Guide you through install if useful
Source: ~100 verified skills (offline) Speed: Instant Trust: Highest (human-reviewed) Setup: None required
Manually vetted skills. Cryptographically signed.
Source: GitHub Code Search - 5,000+ repositories with SKILL.md Speed: 2-5 seconds Trust: Medium (automated filtering) Setup: Requires GITHUB_TOKEN
export GITHUB_TOKEN=your_tokenGenerate at https://github.com/settings/tokens (no scopes required).
Source: Web search across GitHub, marketplaces, documentation Speed: 5-15 seconds Trust: Variable (security-scanned before showing) Activation: User request or auto-trigger when < 3 results
# Main command - hunt for skills/MCPs relevant to your project
agent-hunter hunt .
# Health check - audit all installed skills
agent-hunter audit
# Restore registry to last known good state
agent-hunter rollbackRun from Claude Code: /agent-hunter
Update:
cd ~/.claude/skills/agent-hunter && git pull && ./setupUninstall:
rm -rf ~/.claude/skills/agent-hunterThen remove the ## agent-hunter section from ~/.claude/CLAUDE.md.
Extracts tech keywords only (framework names, library names). No code, paths, or project-specific strings leave your machine.
See CONTRIBUTING.md for details.
See CONTRIBUTING.md for details.
MIT © 2026 Indhra Kiranu N A