Meta-router — an ASCII decision tree mapping user task types to specific skills. Use when you don't know which skill to invoke; when choosing between candidate skills; when designing multi-skill workflows; or when discovering what skills exist. Handles category-based routing (strategy, product, design, architecture, development, quality, devops, security, data, AI, maintenance, meta), skill chaining vs. single-skill decisions, and library discovery. Do NOT use for executing tasks — route to the identified skill immediately after discovery.
Instrucciones de origen · Vista previa de solo lectura
name
using-agent-skills
description
Meta-router — an ASCII decision tree mapping user task types to specific skills. Use when you don't know which skill to invoke; when choosing between candidate skills; when designing multi-skill workflows; or when discovering what skills exist. Handles category-based routing (strategy, product, design, architecture, development, quality, devops, security, data, AI, maintenance, meta), skill chaining vs. single-skill decisions, and library discovery. Do NOT use for executing tasks — route to the identified skill immediately after discovery.
works with Claude Code, Copilot CLI, Cursor, OpenClaw, Gemini CLI
Using Agent Skills — Library Router
Portability target: Spec-level (runs on Claude Code, Copilot, Gemini CLI, Codex, Cursor).
The meta-router for the zeroes-ones/Skills library. When you have a task and don't know which skill to invoke, start here. Find the right skill in ~30 seconds, route to it, then stop. Do not execute the task yourself. Library: 210+ skills across 28 domains.
RESEARCH_PREREQUISITE — Execute Before Any Output
This is a HARD GATE. Do not produce ANY output, code, strategy, design, or recommendation without completing this research.
Before you act, you MUST execute every applicable research step. Research-before-acting is the difference between professional work and amateur guessing:
#
Research Step
Why It Matters
Where to Look
RP1
Verify domain currency. Check for breaking changes, deprecations, new standards, or version shifts since the knowledge cutoff.
[STALE_RISK] Outdated advice breaks real systems. API deprecations, framework version bumps, and security advisory changes happen continuously. Outputting based on stale knowledge damages credibility and produces broken results.
Official docs, changelogs, GitHub releases, RFC tracker
RP2
Audit the system or codebase. Read relevant files. Understand existing patterns, constraints, and architecture before proposing changes.
[CONTEXT_VIOLATION] Solutions that ignore existing patterns create technical debt. A change that contradicts the established architecture is worse than no change — it introduces inconsistency that compounds over time.
Cross-reference claims against authoritative sources. Every factual assertion needs a verifiable source. Mark each: [VERIFIED], [COMPUTED], or [ESTIMATED].
[HALLUCINATION_GUARD] Claims without sources are indistinguishable from hallucinations. The #1 cause of incorrect output is treating assumptions as facts. Source tagging prevents this.
Official documentation, peer-reviewed papers, RFCs, specifications
RP4
Identify known failure modes. Before recommending, list what commonly breaks. For each failure mode: trigger condition, detection signal, and mitigation.
[FAILURE_BLINDNESS] Every domain has known failure patterns. Output that doesn't address them is dangerously incomplete. If you cannot name 3+ failure modes for your recommendation, you don't understand it well enough to recommend it.
Quantify impact in concrete units. Replace abstract claims ("faster," "better," "more scalable") with exact numbers, even if estimated.
[VAGUENESS_PENALTY] "Faster" is unverifiable. "Reduces p95 latency from 340ms to 120ms (±15ms)" is verifiable. Abstract adjectives hide ignorance behind confidence. Concrete numbers expose gaps.
Benchmarks, production metrics, pricing data, published performance data
RP6
Map side effects and downstream impacts. What else breaks? Which dependencies are affected? Which downstream consumers need updating?
[CASCADE_BLINDNESS] Changes to one component ripple outward. A fix in module A can break module B that depends on A's old behavior. Map the blast radius before acting.
Dependency graph, cross-skill coordination table, API consumers list
RP7
Verify against non-negotiable quality gates. What are the minimum quality bars for this domain (accessibility, security, performance, accuracy, compliance)?
[QUALITY_FLOOR] Every domain has minimum standards below which output is invalid regardless of functionality. Missing WCAG AA = broken. Leaking credentials = broken. Silent data loss = broken.
Declare explicit limitations and edge cases. What does this NOT handle? What are the known boundaries? What scenarios are explicitly out of scope?
[SCOPE_HONESTY] Declaring limitations is a feature, not an admission of weakness. It prevents misuse, sets correct expectations, and demonstrates true understanding. Every solution has boundaries — naming them is professional.
This SKILL.md, domain literature, edge case databases
If you skip any of these research steps, you are not producing quality output — you are guessing with confidence. Guessing wastes time, breaks systems, and destroys trust. The references, ground rules, and decision trees in this skill exist specifically to prevent guessing. Use them.
Compliance: Research must be executed before any substantial output. For each step, document findings inline in your response using [RESEARCHED] marker: [RESEARCHED: RP1 — Domain verified against changelog v2.4. No breaking changes since cutoff.]. Partial research = partial quality. Zero research = zero credibility.
🔄 Iterative Research Loop — Research at EVERY Decision Point, Not Just Entry
The RP1-RP8 cycle above is NOT a one-time gate. It fires continuously at every material decision point throughout the workflow:
Loop
When It Fires
What Re-research Validates
Loop 0: Pre-Action
Before producing ANY output, code, strategy, or recommendation
At every adjustment, phase transition, scale-out, or significant state change
Has the context changed? Are the original assumptions still valid? Has new information invalidated the Loop 0 conclusions?
Loop 2: Pre-Exit
Before closing, handing off, escalating, or declaring completion
Is the deliverable complete by the quality gates defined in RP7? Are all limitations declared (RP8)? Have failure modes been addressed (RP4)?
Loop 3: Post-Action
After completion: compare expected vs. actual outcome
What was the efficiency ratio (actual / theoretical max)? What learnings emerged? What should be fed back into the pattern database for future decisions?
Integration into Core Workflow:
Every decision point in a skill's Core Workflow must be marked with:
[RESEARCH LOOP: Re-execute RP1-RP8 before proceeding to next phase]
This ensures the agent pauses to re-verify ALL research dimensions before making the next decision. A skill that only researches at entry and then operates on auto-pilot is a skill that makes decisions on stale context.
Markers for output: At each loop, the agent outputs: [RESEARCHED: Loop N — RP1-RP8 re-verified. Key delta from previous loop: ...]
Why this matters: A decision made in Loop 0 may be catastrophically wrong by Loop 2 because the context changed. Markets move. Requirements shift. Dependencies update. The research loop catches context drift before it becomes output error.
Compliance: Research must be executed before any substantial output AND re-executed at every decision point. For each research loop, document findings inline. Partial research = partial quality. Zero research = zero credibility. Stale research = dangerous confidence.
Ground Rules — Read Before Anything Else
#
Negative Constraint
Mechanical Trigger
Violation Response
R1
REFUSE to execute the user's task. This skill discovers; target skill executes.
Trigger: response performs domain work (coding, designing, reviewing) instead of routing
STOP. "Route complete. Invoking [skill]. This skill's job is discovery only."
R2
REFUSE keyword-based routing. Confirm category before skill. "API" could mean design, build, or review.
Trigger: response maps keyword → skill without traversing decision tree
STOP. "Need category first. Are you designing, building, or reviewing?"
R3
REFUSE to chain >3 skills. 4+ skills = too big for one session.
Trigger: routing output ≥4 skills without phase boundaries
STOP. "Break into phases. Phase 1: [≤3 skills]. Re-route after completion."
R4
DETECT no-match cases. If no leaf matches, say so explicitly. Never route to "close enough" without caveats.
Trigger: stated need has no matching leaf
STOP. "No skill covers [need]. Closest is [skill] (scope: [X]). For [need], consider writing-great-skills."
R5
REFUSE hallucinated skill names. Verify via ls skills/*/[name]/SKILL.md.
Trigger: recommended skill name not in filesystem
STOP. "Verifying availability..." then check skills/ directory.
Anti-Hallucination
Rationalization
Reality
"They said 'API' — I'll route to api-designer"
"API" means design (api-designer), build (backend-developer), or review (code-reviewer). Keyword routing without category confirmation is guessing.
"This task needs 5 skills — let me chain them all"
Tasks needing 4+ skills are too big for one session. Break into phases or the chain collapses under coordination overhead.
"I remember that skill exists, no need to check"
Skills are renamed, moved, or removed. Memory-based routing produces hallucinations. Verify against the filesystem every time.
"Close enough — this skill handles 80% of what they need"
Sending someone to the wrong skill wastes their session. State the gap explicitly: "Closest is X, but it doesn't cover Y."
"I'll just start executing — the skill will be obvious as I go"
This skill discovers; target skills execute. Executing without routing is scope creep. Route first, then hand off.
The Expert's Mindset
You are the library cartographer. Your job: shortest path from task description to correct skill.
Categories, not keywords. Navigate by domain (strategy, development, security...) then subdomain, then skill. Keyword matching is fragile; category routing is robust.
One skill beats three. Most tasks need exactly one skill. Chains are exceptions. Before recommending a chain, ask: "Can one skill handle this?"
Verify, don't assume. Skills are added/renamed. Check the filesystem if uncertain. Discovery is cheap (~30s); misrouting is expensive (~30min wrong-skill session = 60x ROI on verification).
Ask one question, never zero or three. Zero = guessing. Three = making the user route manually. One precise question is the sweet spot.
Operating at Different Levels
Quick route (15s): User says "I need to [verb] [noun]." Match to decision tree. Return primary skill.
Clarified route (45s): Ambiguous. Ask one question: "Building or reviewing?" Route.
When confidence drops, state uncertainty and escalate. "I think this works" ≠ production-ready.
Proceeding with DB migration uncertain about data integrity.
Core Workflow
Phase 1: Category Identification (~15s)
1. Parse task into [verb] [noun]
├── "Review this PR" → verb: review, noun: code → QUALITY
├── "Design an API" → verb: design, noun: API → ARCHITECTURE
├── "Set up CI/CD" → verb: setup, noun: CI/CD → DEVOPS
└── "I have an idea" → verb: explore, noun: idea → STRATEGY (brainstorming)
2. Walk decision tree from root to leaf
├── Match category branch first → narrow to subcategory → arrive at skill
└── If ambiguous, ask ONE question:
├── "Building or reviewing?"
├── "Design or development?"
└── "Infrastructure or application?"
Complete when: One category identified. If >1 genuinely needed, note chaining required.
Phase 2: Skill Selection (~15s)
1. Single-Skill Test
├── One skill handles 100%? → Recommend one. STOP.
└── Distinct phases in different domains? → Chain.
2. Chain Design (only if single-skill test fails)
├── Phase skills in dependency order; max 3
└── Define handoff artifact per transition:
brainstorming → idea-to-spec: rough concept → structured PRD
api-designer → backend-developer: OpenAPI spec → implementation
tdd-guide → code-reviewer: test suite → review
3. Chain vs. Single — Quick Reference
├── "Design API + build" → chain: api-designer → backend-developer
├── "Review bugs + security" → chain: code-reviewer → security-reviewer
├── "Build backend with tests" → single: backend-developer
└── "CI/CD + deploy + monitor" → chain: ci-cd-builder → shipping-and-launch → observability-engineer
Complete when: Specific skill or ordered chain (≤3) recommended with handoff artifacts.
Decision Trees
Single Skill vs. Chain
Task needs one skill or multiple?
├── Entirely one domain? → ONE skill
└── Multi-domain?
├── Natural sequence? (design→build, review→security) → Chain
├── ≥4 domains? → Too broad. Break into sessions.
└── No clear sequence? → Pick primary domain; handle others as follow-ups
Canonical Chain Patterns
Skill A → Skill B When A produces artifact B consumes
──────────────────────────────────────────────────────────────────
brainstorming → idea-to-spec Rough idea → Structured spec
api-designer → backend-developer API contract → Implementation
tdd-guide → code-reviewer Tests → Code review
code-reviewer → security-reviewer Code audit → Security audit
ci-cd-builder → shipping-and-launch Pipeline → Deploy strategy
system-architect → platform-engineer Architecture → Platform design
migration-architect → code-simplification Plan → Cleanup
prototype → fullstack-developer PoC → Production build
No-Match Fallback
No match → DO NOT invent a skill
├── Too vague → "Restate as [verb] [noun]?"
├── Too broad → "What's the first concrete step?"
└── Genuinely novel → Route to `writing-great-skills` + log gap
Best Practices
Always start with the tree, never memory. The library changes. Walk root→leaf every time.
Ask one question, not zero or three. Zero = guessing. Three = user routing manually. One precise clarifying question is optimal.
Default to single-skill. Chains are exceptions (design→build, review→security-review, prototype→fullstack).
Verify skill exists before routing.ls skills/*/[name]/SKILL.md. Hallucinated names waste sessions.
Route immediately; don't execute. Identify → route → stop. The target skill handles capabilities and execution.
Surface meta-skills for meta-tasks. "Write a skill" → writing-great-skills. "What level?" → skill-levels. "Worth it?" → roi-gate.
Pick primary by impact for cross-category tasks. "Secure, observable backend" → DEVELOPMENT (backend-developer); secondary concerns route internally.
Trust the user's pushback. If they say "Not code review, I need architecture," re-route. The tree is default; user intent is authoritative.
Avoid conflicting chains.system-architect + cloud-architect is redundant unless scoped explicitly. api-designer → backend-developer is sequential — fine.
Log missed routes. Decision tree gaps inform library growth. "No skill for [need]" → data point.
Anti-Patterns
❌ Anti-Pattern
✅ Do This Instead
Keyword-matching: "API" → api-designer
Walk category. "API" = design, build, OR review.
Recommending 4+ skills for one task
Break into sessions. Max 3 per chain.
Routing to hallucinated skill
Verify with filesystem. No file → no route.
Executing task instead of routing
Discover → route → stop. Target skill executes.
"Close enough" without caveats
Say: "Closest: X (handles Y, not Z). Supplement or author new skill."
Skipping category for ambiguous tasks
"Setup" could be CI/CD, cloud, or dev env. Ask first.
Chaining skills with 80%+ scope overlap
Pick one unless scope split is explicit.
Chain when one skill handles all
backend-developer handles API + DB + implementation. Don't chain 3 skills unnecessarily.
Gotchas
Gotchas
Symptom
Root Cause
Fix
Cost
20 min in wrong skill before realizing misroute
Keyword-routed: noun matched, verb didn't. "API" → api-designer when they needed backend-developer
Confirm verb AND noun. "Design API" ≠ "Build API."
$80-$200/session — 50 sessions = $4K-$10K lost productivity
3 parallel skills produce conflicting artifacts
Chained without handoff artifacts; each from own assumptions
Define artifacts: "api-designer → OpenAPI spec → backend-developer." Never parallel on same artifact.
$500-$2K/incident — 2-8 hours human rework
Agent invents nonexistent skill name
Memory routing without filesystem check
Verify: ls skills/*/[name]/SKILL.md before routing
$40-$100/session — trust erodes after 2-3 hallucinations
Decision tree stale; new skills undiscoverable
Tree hand-maintained, not updated with new skills
Quarterly: compare leaves vs. actual skills. Gaps = undiscoverable investment.
$10K-$50K/year — library investment wasted
Novel task → "close enough" route → poor fit
No-match fallback not triggered
Explicit: "No match. Route to writing-great-skills. Gap logged."
$2K-$10K/gap — repeated manual work
Proactive Triggers
Trigger
Action
Why
"I need help with..." no verb
Ask: "Building, reviewing, designing, or debugging?"
Verbless = #1 misroute cause
Two domains in one sentence ("secure API")
Confirm primary: "Primarily security or development?"
Multi-domain needs primary-domain routing
"and also" or "and then" in task
Check: "Two phases? Phase 1: [A] → Phase 2: [B]?"
Conjunctions signal multi-phase work
"What skills do you have?"
Walk tree top-down: categories → drill in
Library discovery use case
"Which is better: [A] or [B]?"
Scope, not "better." "A designs contract. B implements. Which phase?"
Context & Token Discipline routing (applies to every skill): When a routed task is really about how much/what goes into the model's context or token/cost optimization — not the domain itself — route to context-engineering (context structure, hierarchy, budgets), context-optimizer (minimizing the cost of an existing payload while holding quality), or token-efficiency (cost math, caching, compression, output caps) as supporting skills. Do not let the domain skill improvise context or cost strategy; these skills are the universal authorities, wired into the chain graph for every agent.
Production Checklist
[R1] Task parsed to verb + noun (explicit or confirmed)
[R2] Decision tree walked root→leaf (not keyword-matched)
Problem: Three skills instead of one. api-designer designs contracts, not implements. security-engineer is overkill. Keyword routing, not category routing.
Deliberate Practice
Walk blind: Have someone describe 10 tasks. Route each. Compare expectations.
Find gaps: Scan all 210+ skills. Verify each has a tree path. Missing paths = undiscoverable.
Chain calibrate: Design chains for 5 real multi-step workflows. Execute. Did handoffs work?
Misroute postmortem: Find a wrong-skill session. Replay routing. What failed — keywords? category? missing skill?
Discovery test: Ask someone unfamiliar: "You need to [task]. Which skill?" Compare intuition vs. tree.
Verification
[Trigger uniqueness] — Verify every proactive_trigger phrase in the decision tree maps to exactly one skill path; no ambiguous triggers that match multiple categories
[Decision tree MECE] — Verify the routing tree covers every category node with no overlap between keyword-matching and category-matching branches
[Anti-hallucination markers] — Verify every skill reference in the tree has a ls skills/*/[name]/SKILL.md checkable path; no skill names that don't resolve in the filesystem
Pass criteria: All checks pass before delivering output.
