agent-router

A Claude Code skill + agent that discovers every installed skill and subagent, ranks them for the task at hand, and routes to the best one — so you don't have to remember which of your dozens of skills/agents to use.

• Skill (/agent-router): can recommend and dispatch to the chosen tool (run by the main agent, which can call other agents/skills).
• Agent (subagent_type: agent-router): recommends a ranked shortlist only (read-only advisor).
• Discovery (/skill-finder): for new users — browses the marketplace and ranks uninstalled plugins/skills/agents by rating so you know what to install.

How it works

Given a task, it:

1. Inventories candidates from the session list, ~/.claude/, ./.claude/, and plugins.
2. Scores each 0-100 = relevance (60) + specificity (20) + reputation (20, from data/registry.json).
3. Shows a ranked table and routes to / recommends the top pick.

Install / register — two ways

A) Claude Code plugin (no npm needed)

Push this repo to GitHub, then anyone runs:

/plugin marketplace add RamanaNani/agent-router
/plugin install agent-router@agent-router

(The .claude-plugin/marketplace.json and plugin.json make this work.)

B) npx installer (published to npm)

Publish once:

npm login
npm publish --access public

Then anyone installs the skill + agent into their Claude config with:

npx agent-router            # installs to ~/.claude (user scope)
npx agent-router --project  # installs to ./.claude (project scope)

Use it

/agent-router

or just ask: "which agent should I use to review this Go file?"

Discover new skills (for new users)

Empty ~/.claude/? Use the companion skill-finder skill to browse the marketplace and find highly-rated tools to install:

/skill-finder

or ask: "what skills are available for testing?" It checks what you already have, then ranks uninstalled candidates by relevance + rating (GitHub stars or curated score), shows each source link, and gives you the exact /plugin install (or npx) command. Candidates come from a live web/GitHub search plus the curated catalog in data/marketplace.json — it never invents repos.

Use it internally first (dogfood with logs)

Both skills append a JSON line to ~/.claude/agent-router/logs/decisions.jsonl every time they run — the task, what they picked, and (for skill-finder) what was missing. After a week or two of real use:

node scripts/review-logs.js

prints a summary: most-routed tools, misroutes you flagged, and the capabilities you keep needing but don't have. Use it to tune data/registry.json scores and decide which skills to build before publishing. The log lives in ~/.claude (personal) — it is not committed to the repo.

Customize the reputation scores

Edit data/registry.json — add { name, type, score, source, notes } entries for tools you trust. Unknown tools default to a neutral score of 10.

Layout

agent-router/
├── package.json                 # npm metadata + bin -> scripts/install.js
├── scripts/install.js           # npx installer (copies into ~/.claude)
├── .claude-plugin/
│   ├── plugin.json              # plugin manifest (skills + agents)
│   └── marketplace.json         # marketplace entry for /plugin install
├── skills/agent-router/SKILL.md # the routing skill (recommend + dispatch)
├── skills/skill-finder/SKILL.md # marketplace discovery (browse + rank by rating)
├── agents/agent-router.md       # the advisor subagent (recommend only)
├── data/registry.json           # curated reputation of installed tools
└── data/marketplace.json        # curated discovery catalog (skill-finder)

License

MIT