Skills Overview

What Are Skills?

Skills are the internal workflow assets beneath the 20 public instruction tools. Consumers never call skills directly — they invoke an instruction tool which internally orchestrates the right skills.

72 skills · 16 domains · 0 orphans (verified by scripts/verify_matrix.py)

Model class glossary

Every skill carries a model-class hint. These describe task complexity, not pricing — the class tells the orchestrator what tier of model can handle this task well:

Badge	Internal key	What it means
Zero-Cost	free	Solved by zero-cost models — GPT-4.1, GPT-5 mini
Efficient	cheap	Low-cost models — Claude Haiku 4.5, GPT-5.4 mini
Advanced	strong	Full-capability models — Claude Sonnet 4.6, Claude Opus 4.6, GPT-5.4, GPT-5.3-Codex
Cross-Model	reviewer	Independent cross-model review — Gemini 3.1 Pro, Claude Opus 4.6

See Model Routing for the full class resolution logic · Artificial Analysis

Skill Domains

Core domains

Requirements (4) req-* — Analysis, scope, ambiguity detection, acceptance criteria

Architecture (4) arch-* — System, reliability, scalability, security design

Quality (5) qual-* — Code analysis, review, performance, security, refactoring priority

Debugging (4) debug-* — Assistant, root cause, reproduction, postmortem

Documentation (4) doc-* — Generator, README, API reference, runbook

Evaluation (5) eval-* — Prompt, output grading, variance, design, benchmark

Benchmarking (3) bench-* — Analyzer, blind comparison, eval suite

Workflows (3) flow-* — Orchestrator, context handoff, mode switching

Governance (7) gov-* — Policy, compliance, guardrails, prompt-injection hardening

Orchestration (4) orch-* — Agent orchestrator, delegation, multi-agent, synthesis

Prompting (4) prompt-* — Engineering, chaining, refinement, hierarchy

Research (4) synth-* — Comparative, research, recommendation, engine

Strategy (4) strat-* — Advisor, roadmap, prioritization, tradeoff

Resilience (5) resil-* — Homeostatic, membrane, voter, replay, clone-mutate

Leadership (7) lead-* — Capability mapping, roadmap, exec briefing, mentoring

Feature-gated domains

Feature gates required

The domains below must be explicitly enabled via environment variables. They are off by default to prevent non-deterministic behavior in production.

Adaptive (5) adapt-* — ACO router, annealing, Hebbian, Physarum, quorum. Requires ENABLE_ADAPTIVE_ROUTING=true

Instruction–Skill Coverage Graph

The graph below shows which instructions invoke which skill domains. Auto-generated from src/workflows/workflow-spec.ts.

flowchart LR

    subgraph INSTRUCTIONS["Instructions"]
        adapt["adapt"]
        bootstrap["bootstrap"]
        debug["debug"]
        design["design"]
        document["document"]
        enterprise["enterprise"]
        evaluate["evaluate"]
        govern["govern"]
        implement["implement"]
        meta_routing["meta-routing"]
        orchestrate["orchestrate"]
        plan["plan"]
        prompt_engineering["prompt-engineering"]
        refactor["refactor"]
        research["research"]
        resilience["resilience"]
        review["review"]
        testing["testing"]
    end

    subgraph D_req["Requirements"]
        req_analysis["req-analysis"]
        req_scope["req-scope"]
        req_ambiguity["req-ambiguity-detection"]
        req_acceptance["req-acceptance-criteria"]
    end

    subgraph D_arch["Architecture"]
        arch_system["arch-system"]
        arch_reliability["arch-reliability"]
        arch_scalability["arch-scalability"]
        arch_security["arch-security"]
    end

    subgraph D_qual["Quality"]
        qual_review["qual-review"]
        qual_code["qual-code-analysis"]
        qual_perf["qual-performance"]
        qual_security["qual-security"]
        qual_refactor["qual-refactoring-priority"]
    end

    subgraph D_debug["Debug"]
        debug_assistant["debug-assistant"]
        debug_root["debug-root-cause"]
        debug_repro["debug-reproduction"]
        debug_post["debug-postmortem"]
    end

    subgraph D_gov["Governance"]
        gov_injection["gov-prompt-injection-hardening"]
        gov_policy["gov-policy-validation"]
        gov_data["gov-data-guardrails"]
        gov_model["gov-model-governance"]
        gov_compat["gov-model-compatibility"]
        gov_regulated["gov-regulated-workflow-design"]
        gov_compliance["gov-workflow-compliance"]
    end

    subgraph D_orch["Orchestration"]
        orch_agent["orch-agent-orchestrator"]
        orch_delegation["orch-delegation"]
        orch_multi["orch-multi-agent"]
        orch_synthesis["orch-result-synthesis"]
    end

    subgraph D_resil["Resilience"]
        resil_homeostatic["resil-homeostatic"]
        resil_membrane["resil-membrane"]
        resil_voter["resil-redundant-voter"]
        resil_replay["resil-replay"]
        resil_clone["resil-clone-mutate"]
    end

    subgraph D_adapt["Adaptive (gated)"]
        adapt_aco["adapt-aco-router"]
        adapt_anneal["adapt-annealing"]
        adapt_hebbian["adapt-hebbian-router"]
        adapt_physarum["adapt-physarum-router"]
        adapt_quorum["adapt-quorum"]
    end



    design --> D_req
    design --> D_arch
    implement --> D_req
    implement --> D_arch
    research --> D_qual
    review --> D_qual
    review --> D_gov
    govern --> D_gov
    resilience --> D_resil
    orchestrate --> D_orch
    adapt --> D_adapt
    refactor --> D_qual
    debug --> D_debug
    bootstrap --> D_req
    plan --> D_arch
    evaluate --> D_qual
    enterprise --> D_gov
    testing --> D_qual
    prompt_engineering --> D_qual

How Skills Are Invoked

Skills are dispatched by tier in src/tools/skill-handler.ts:

• Governance tier (gov-*): Optional human-in-the-loop via ENABLE_GOVERNANCE_STRICT
• Adaptive tier (adapt-*): Requires ENABLE_ADAPTIVE_ROUTING=true
• Core tier: Everything else, always available

See The Skill System for the full architecture.