tool-discovery
Recommend the right agents and skills for any task. Covers both heavyweight agents (Task tool) and lightweight skills (Skill tool). Triggers on: which agent, which skill, what tool should I use, help me choose, recommend agent, find the right tool.
When & Why to Use This Skill
The Tool Discovery skill serves as an intelligent orchestration layer designed to optimize Claude's workflow by recommending the most efficient agents and skills for any specific task. It streamlines the decision-making process by distinguishing between lightweight, pattern-based 'Skills' and heavyweight, reasoning-focused 'Agents,' ensuring users always utilize the most cost-effective and capable tool for their requirements.
Use Cases
- Architectural Decision Making: Helping users decide whether a task requires a specialized sub-agent (like a Python Expert) for complex reasoning or a simple skill (like file-search) for quick lookups.
- Workflow Efficiency: Providing a quick reference guide to match specific developer tasks—such as SQL window functions, Git rebasing, or Tailwind styling—with the precise tool needed to execute them.
- Resource and Cost Optimization: Reducing token consumption and latency by recommending lightweight skills for reference-heavy tasks instead of spawning resource-intensive heavyweight agents.
- Tool Catalog Navigation: Acting as a central directory for discovering available capabilities within a complex agentic environment, triggered by natural language queries like 'which tool should I use?'
| name | tool-discovery |
|---|---|
| description | "Recommend the right agents and skills for any task. Covers both heavyweight agents (Task tool) and lightweight skills (Skill tool). Triggers on: which agent, which skill, what tool should I use, help me choose, recommend agent, find the right tool." |
| allowed-tools | "Read Glob" |
| depends-on | [] |
| related-skills | [claude-code-templates, claude-code-debug] |
Tool Discovery
Recommend the right agents and skills for any task.
Decision Flowchart
Is this a reference/lookup task?
├── YES → Use a SKILL (lightweight, auto-injects)
└── NO → Does it require reasoning/decisions?
├── YES → Use an AGENT (heavyweight, spawns subagent)
└── MAYBE → Check catalogs below
Rule: Skills = patterns/reference. Agents = decisions/expertise.
Quick Skill Reference
| Skill | Triggers |
|---|---|
| file-search | fd, rg, fzf, find files |
| find-replace | sd, batch replace |
| code-stats | tokei, difft, line counts |
| data-processing | jq, yq, json, yaml |
| structural-search | ast-grep, sg, ast pattern |
| git-workflow | lazygit, gh, delta, rebase |
| python-env | uv, venv, pyproject |
| rest-patterns | http methods, status codes |
| sql-patterns | cte, window functions |
| sqlite-ops | sqlite, aiosqlite |
| tailwind-patterns | tailwind, tw classes |
| mcp-patterns | mcp server, protocol |
Quick Agent Reference
| Agent | Triggers |
|---|---|
| python-expert | Python, async, pytest |
| typescript-expert | TypeScript, types, generics |
| react-expert | React, hooks, state |
| postgres-expert | PostgreSQL, query optimization |
| cloudflare-expert | Workers, KV, D1, R2 |
| Explore | "where is", "find" |
| Plan | design, architect |
How to Launch
Skills:
Skill tool → skill: "file-search"
Agents:
Task tool → subagent_type: "python-expert"
→ prompt: "Your task"
Match by Task Type
| Task | Skill First | Agent If Needed |
|---|---|---|
| "How to write a CTE?" | sql-patterns | sql-expert |
| "Optimize this query" | — | postgres-expert |
| "Find files named X" | file-search | Explore |
| "Set up Python project" | python-env | python-expert |
| "What HTTP status for X?" | rest-patterns | — |
Tips
- Skills are cheaper - Use for lookups, patterns
- Agents are powerful - Use for decisions, optimization
- Don't over-recommend - Max 2-3 tools per task
Additional Resources
For complete catalogs, load:
./references/agents-catalog.md- All agents with capabilities./references/skills-catalog.md- All skills with details