agent-hunter is a Claude Code SKILL that discovers and ranks the best skills and MCP servers for your projects.

New to Claude Code? agent-hunter is designed as a first-install skill. Install it right after Claude Code itself.

git clone --depth 1 https://github.com/indhra/agent-hunter.git ~/.claude/skills/agent-hunter
cd ~/.claude/skills/agent-hunter

./setup

The setup script will:

• [YES] Check Python 3.10+
• [YES] Create a local virtual environment
• [YES] Install dependencies (PyYAML, requests, rich)
• [YES] Register /agent-hunter in your global ~/.claude/CLAUDE.md
• [YES] Symlink sub-skills
• [YES] Create ~/.local/bin/agent-hunter for CLI access

**** /agent-hunter is now available in every Claude Code session.

Open any Claude Code session (brand new project or existing repo) and type:

/agent-hunter

Or ask Claude directly:

What skills or tools should I install for this project?

agent-hunter will automatically:

1. Read your project's tech stack (framework names only, no code)
2. Hunt curated skills + GitHub results
3. Security-scan every result
4. Show top 3 recommendations
5. Guide you through installation if useful

By default, agent-hunter runs when you call /agent-hunter or ask Claude. To make it automatically activate when you start work on a new project:

Add this to your project's .claude/settings.json:

{
 "autoActivateSkills": ["agent-hunter"],
 "onSessionStart": {
 "if": "currentProject !== lastProject",
 "then": "/agent-hunter"
 }
}

What this does:

• Detects when you open a new project (by comparing current directory path with last session)
• Automatically runs /agent-hunter on session start
• Shows top 3 recommendations without waiting for you to ask

For team-wide activation, add to your shell profile (~/.zshrc or ~/.bash_profile):

export AGENT_HUNTER_AUTO=1

Then in Claude Code, agent-hunter will auto-run once per session per unique project path.

If you're hunting on existing projects (not just new ones), GitHub token is highly recommended:

export GITHUB_TOKEN=ghp_xxxxxxxxxxxx

Generate at https://github.com/settings/tokens (no scopes needed, public access only).

Make permanent:

echo 'export GITHUB_TOKEN=ghp_xxxxxxxxxxxx' >> ~/.zshrc
source ~/.zshrc

Brownfield projects have very specific needs. With GitHub token:

• [YES] Searches 5,000+ skill repositories
• [YES] Finds niche tools for refactoring, optimization, testing
• [YES] Better matches for "existing project enhancement"
• [YES] Avoids GitHub API rate limit hits

Recommendation:

• Without token: Works fine, uses verified index + web search
• With token (recommended): Full access to all 5,000+ GitHub skills, best for brownfield projects

/agent-hunter
# or
agent-hunter hunt .

Discovers the top 3 skills/MCPs relevant to your current project.

/agent-hunter audit
# or
agent-hunter audit

Checks all installed skills for:

• Security issues
• Tampering (SHA mismatch)
• Dependency conflicts

/agent-hunter rollback
# or
agent-hunter rollback

Restores your skill registry to the last known good snapshot (created before the last audit/update).

# Enable automatic hunting on new projects
export AGENT_HUNTER_AUTO=1

# GitHub API access (for broader discovery)
export GITHUB_TOKEN=ghp_...

# Debug output
export DEBUG=1

Create or edit .claude/CLAUDE.md in your project:

## agent-hunter

Proactively discover, security-scan, and install SKILL.md files and MCP servers
relevant to this project. Invoke /agent-hunter when:
- Starting a new feature that might have existing tools
- Asking "what should I install"
- Need to audit installed skills

Available skills: /agent-hunter, /agent-hunter-update

Then add to .claude/settings.json:

{
 "autoActivateSkills": ["agent-hunter"],
 "proactiveTriggers": {
 "newProject": true,
 "beforeFeatureBuild": true
 }
}

Check 1: Verify setup completed

ls ~/.claude/skills/agent-hunter/

Check 2: Reload your shell

source ~/.zshrc # or ~/.bash_profile

Check 3: Verify global CLAUDE.md

grep -i "agent-hunter" ~/.claude/CLAUDE.md

Reason 1: GitHub rate limit hit

• Set GITHUB_TOKEN for higher quota

Reason 2: Tech stack not detected

• Try again and agent-hunter will show what was detected
• File an issue if detection is missing for your stack

Reason 3: Your stack is very niche

• Use web search: agent-hunter will offer to search broader when < 3 results

If agent-hunter flags a skill as [REVIEW] (review before installing):

1. Read the SKILL.md file from the recommended repo
2. Check the flagged patterns (usually filesystem access, network calls)
3. Decide if it fits your use case
4. Install manually if appropriate

RED ([BLOCKED]) flagged skills are never shown - they're blocked entirely.

cd ~/.claude/skills/agent-hunter && git pull && ./setup

rm -rf ~/.claude/skills/agent-hunter

Then remove the ## agent-hunter section from ~/.claude/CLAUDE.md.

[YES] Only tech keywords (framework names, library names) are extracted [YES] No code, no paths, no secrets ever leave your machine [YES] All results security-scanned before showing [YES] RED results blocked entirely - never shown to you

See CONTRIBUTING.md for full details.

1. Install: ~/.claude/skills/agent-hunter && ./setup
2. First hunt: Open a project and type /agent-hunter
3. Token (optional): Set GITHUB_TOKEN for broader discovery
4. Proactive (optional): Enable AGENT_HUNTER_AUTO=1 for auto-activation

Questions?

• [Issue] GitHub Issues
• 💬 Discussions