SKILL.md

name	description	allowed-tools	compatibility	metadata
router	Deterministic routing rules for delegating work to subagents. Use for triage, delegation decisions, or routing requests to specialized subagents.	read list glob grep line_view webfetch find_symbol get_symbols_overview gitingest ast_grep todowrite todoread skill task	opencode	audience category maintainers routing

What I do

• Route user requests to the best-fit agent(s) based on intent and capabilities.
• Follow deterministic, priority-based routing decisions.
• Encourage short, safe clarification when requests are ambiguous.
• Operate in read-only mode—never write, edit, or execute commands directly.

When to use me

Use this skill when you need to:

• Triage a user request and decide which agent should handle it.
• Make delegation decisions for complex tasks requiring multiple specialists.
• Ensure consistent routing behavior across multiple sessions.
• Maintain clear separation between planning/analysis and implementation.

Keywords

Routing, delegate, triage, dispatch, route, assign, which agent, who should, orchestrate, coordinate, subagent, specialist, expert.

Routing policy

Ground rules

• Prefer an explicit user request (e.g. "use @dev") over heuristics.
• If the request is ambiguous, ask up to 3 targeted clarifying questions and stop.
• If the request is clear but complex, choose a short chain (max 3 agents) rather than one overloaded agent.
• Use parallel delegation only for independent workstreams that don't share dependencies. Maximum 4 parallel agents per delegation.
• When multiple tasks are available, explicitly evaluate whether they can execute in parallel before choosing serial or parallel routing.
• Never modify files directly—always delegate to agents with appropriate tool access.

Capability map (keywords → agent)

• Strategy/architecture/tradeoffs → @oracle
• Local codebase search (“where is”, “find file”, “locate”) → @explorer
• Implementation (“implement”, “fix bug”, “refactor”) → @dev
• Code review/security/performance critique → @code-review
• Docs writing/README/API docs → @writer
• UI/UX/design/CSS/components → @ux
• Frontend interface design → @frontend-design
• External research/docs/GitHub URLs → @librarian
• Git commit message/commit help → @commits
• Git fixup/autosquash history cleanup → @fixup
• Tailwind theme/config/colors/dark mode → @tailwind-theme
• Find similar patterns across codebase → @code-pattern-analyst
• Mutation testing/test quality → @mutation-testing
• Redundant tests/coverage delta pruning → @test-drop
• Prompt injection/safety review → @prompt-safety-review

Priority order (stop at first match)

1. Explicit agent request from user (e.g. “use @dev”).
2. Git workflows (@commits, @fixup).
3. Tailwind theme/config (@tailwind-theme).
4. Prompt safety (@prompt-safety-review).
5. External research (@librarian).
6. Local discovery (@explorer).
7. Documentation chain (@explorer → @writer).
8. UI/UX chain (@explorer → @ux or @frontend-design).
9. Code review (@code-review).
10. Implementation chain (@explorer → @dev).
11. Strategy/architecture (@oracle).
12. Test quality (@mutation-testing, @test-drop).
13. Fallback: ask clarifying questions.

Clarification protocol (max 3 questions)

Ask targeted questions that unblock routing:

• “Which file or area should I focus on?”
• “Do you want a quick fix or a refactor?”
• “Can you paste the exact error message or output?”
• “Are there specific constraints or requirements I should consider?”
• “Is this for a new feature, bug fix, or investigation?”

Delegation prompt template

When delegating, include:

• Goal
• Constraints (read-only vs write, tests required, no new deps, etc.)
• Acceptance criteria
• Relevant files/paths, if known
• Context or background that will help the agent succeed

Examples

User: “Add a dark mode toggle to the navbar.” Routing: @explorer → @ux Rationale: Needs to find the navbar component first, then apply UI changes.

User: “Research how Stripe handles idempotency and tell me how we should implement it in this repo.” Routing: @librarian → @oracle → @dev Rationale: External research first, then strategy, then implementation.

User: “Why is the build failing? Here is the error…” Routing: @explorer → @dev Rationale: Needs to find relevant code matching the error, then fix it.

User: “Find all places where we use console.log.” Routing: @explorer Rationale: Pure search task—no implementation needed yet.

User: “Write a commit message for my changes.” Routing: @commits Rationale: Explicit meta workflow—specialized agent for commit messages.

User: “Fix login bug in auth.ts AND update API docs to reflect the new endpoint changes.” Routing: @dev (parallel) @writer Rationale: Two independent tasks—bug fix and documentation update can run simultaneously.

Guidelines

• Be decisive. If you can route confidently, delegate immediately.
• If you cannot route safely, ask clarifying questions—never guess.
• Keep chains short (max 3 agents) to maintain focus and traceability.
• Always include sufficient context when delegating to reduce back-and-forth.
• Document your routing decision briefly when confidence is low or the user asks "why".
• Use parallel delegation only when tasks are truly independent—avoid coordination overhead.
• When multiple tasks exist, perform a parallelization check: assess independence (different files/modules), absence of shared dependencies, and no conflicts on shared resources (tests, build, git). Only parallelize when safe.
• Respect agent capabilities: don't delegate tasks that require tools the agent doesn't have.
• When in doubt, prefer chaining through @explorer to gather context before implementation.