diff --git a/.github/workflows/skill-review.yml b/.github/workflows/skill-review.yml new file mode 100644 index 0000000..aee21d6 --- /dev/null +++ b/.github/workflows/skill-review.yml @@ -0,0 +1,22 @@ +# Tessl Skill Review — runs on PRs that change any SKILL.md; posts scores as one PR comment. +# Docs: https://github.com/tesslio/skill-review +name: Tessl Skill Review + +on: + pull_request: + branches: [main] + paths: + - "**/SKILL.md" + +jobs: + review: + runs-on: ubuntu-latest + permissions: + pull-requests: write + contents: read + steps: + - uses: actions/checkout@v4 + - uses: tesslio/skill-review@main + # Optional quality gate (off by default — do not enable unless user asked): + # with: + # fail-threshold: 70 diff --git a/platform/skills/audit/SKILL.md b/platform/skills/audit/SKILL.md index 61edc52..36e2f80 100644 --- a/platform/skills/audit/SKILL.md +++ b/platform/skills/audit/SKILL.md @@ -1,13 +1,27 @@ --- -name: quorum:audit -description: "Run a quorum audit manually — trigger consensus review, re-run failed audits, test audit prompts, or force a specific provider. Use when the hook-based auto-trigger didn't fire, or you want explicit control." -model: sonnet +name: quorum-audit +description: "Trigger a manual cross-model consensus audit on code changes. Submits evidence to an independent auditor (Codex, GPT, Claude, or any configured provider) for structural review. Use when hook-based auto-trigger did not fire, you need to re-run a failed audit, force a specific provider, or want explicit control over the consensus cycle. Triggers on: 'audit', 'run audit', 'trigger audit', 're-audit', 'manual review', '감사'." +user-invocable: true +triggers: + - audit + - run audit + - trigger audit + - re-audit + - manual review --- # Audit -Manual cross-model review trigger. +Manually trigger the quorum consensus protocol: submit evidence, route to an independent auditor, and collect a verdict. The author model cannot approve its own code — a second model must review. + +## Workflow + +1. Gather evidence (changed files, context, fitness score) +2. Select auditor provider via tier router (T1 skip / T2 simple / T3 deliberative) +3. Submit evidence to auditor for structural review +4. Collect verdict — approved, rejected, or amendment required +5. If rejected, author fixes and resubmits; cycle repeats until consensus ## Knowledge -Read `agents/knowledge/protocols/audit.md` for the full protocol. +Read `agents/knowledge/protocols/audit.md` for the full protocol — includes evidence format, provider routing, verdict schema, and retry logic. diff --git a/platform/skills/designer/SKILL.md b/platform/skills/designer/SKILL.md index 68759ce..b85d96b 100644 --- a/platform/skills/designer/SKILL.md +++ b/platform/skills/designer/SKILL.md @@ -1,12 +1,29 @@ --- -name: quorum:designer -description: "Generate and validate design documents (Spec, Blueprint, Domain Model, Architecture) with mandatory mermaid diagrams. Use after PRD confirmation when DRM requires Design Phase. Triggers on 'generate design', 'design docs', '설계 문서', '설계 생성'." -model: opus +name: quorum-designer +description: "Generate and validate design documents — Spec, Blueprint, Domain Model, and Architecture diagram — each with mandatory mermaid visualizations. Produces the four artifacts required by the Design Phase gate before implementation can begin. Use after PRD confirmation when the DRM requires a Design Phase, or when creating architecture documentation for a track. Triggers on: 'design', 'generate design', 'design docs', 'blueprint', 'architecture diagram', '설계 문서', '설계 생성'." +user-invocable: true +triggers: + - design + - generate design + - design docs + - blueprint + - architecture diagram --- # Designer -Design document generation — 4 artifacts that define how to build. +Generate the four design artifacts that define how to build — each validated with mandatory mermaid diagrams before the Design gate opens. + +## Workflow + +1. Read confirmed PRD and track scope from planner output +2. Generate Spec document (functional requirements, constraints, acceptance criteria) +3. Generate Blueprint (component hierarchy, data flow, API boundaries) +4. Generate Domain Model (entity relationships, schemas, invariants) +5. Generate Architecture diagram (system context, container, component views) +6. Embed mermaid diagrams in each artifact (sequence, class, flowchart, C4) +7. Validate all four artifacts exist and contain required mermaid blocks +8. Open the Design gate for implementation to proceed ## Knowledge diff --git a/platform/skills/orchestrator/SKILL.md b/platform/skills/orchestrator/SKILL.md index b83e947..79dbad6 100644 --- a/platform/skills/orchestrator/SKILL.md +++ b/platform/skills/orchestrator/SKILL.md @@ -1,12 +1,27 @@ --- -name: quorum:orchestrator -description: "Session orchestrator — reads handoff, picks unblocked tasks, distributes to parallel workers, tracks agent assignments, manages correction cycles." -model: opus +name: quorum-orchestrator +description: "Distribute tasks to parallel worker agents, track assignments, and manage correction cycles across worktrees. Reads session handoff, picks unblocked tasks by dependency graph, assigns to agents by complexity tier, and handles retries on audit rejection. Use when resuming a session, distributing track work, or coordinating multi-agent execution. Triggers on: 'orchestrate', 'distribute tasks', 'resume session', 'run track', '오케스트레이션'." +user-invocable: true +triggers: + - orchestrate + - distribute tasks + - resume session + - run track --- # Orchestrator -Task distribution and agent lifecycle management. +Task distribution and agent lifecycle management. Coordinates parallel worktree execution with claim-based ownership and wave-based scheduling. + +## Workflow + +1. Read session handoff and identify unblocked tasks +2. Classify task complexity (T1 micro / T2 simple / T3 deliberative) +3. Distribute tasks to worker agents via ClaimService (per-file ownership) +4. Execute in waves — parallel within wave, sequential across phases +5. Collect results, trigger audit per wave +6. On rejection: route to Fixer agent for targeted correction, then re-audit +7. On approval: merge worktree, update RTM, advance to next wave ## Knowledge diff --git a/platform/skills/planner/SKILL.md b/platform/skills/planner/SKILL.md index 4e2817a..30dbb08 100644 --- a/platform/skills/planner/SKILL.md +++ b/platform/skills/planner/SKILL.md @@ -1,12 +1,30 @@ --- -name: quorum:planner -description: "Design tasks into tracks with work breakdowns. Writes and maintains PRDs, generates DRM-driven documents. Use whenever the user mentions planning, PRD, tracks, features, requirements, work breakdown, or 설계/기획." -model: opus +name: quorum-planner +description: "Decompose requirements into tracks with MECE work breakdowns and generate DRM-driven planning documents (PRD, ADR, execution order, work catalog). Use when starting a new feature, defining project scope, writing a PRD, creating tracks, or breaking work into tasks. Triggers on: 'plan', 'PRD', 'tracks', 'work breakdown', 'requirements', 'features', '설계', '기획'." +user-invocable: true +triggers: + - plan + - PRD + - tracks + - work breakdown + - requirements + - features --- # Planner -8-phase requirement analysis and document generation protocol. +8-phase requirement analysis and document generation protocol. Produces 10 document types across project and track levels — from PRDs and ADRs down to work breakdowns, API contracts, and data models. + +## Workflow + +1. Clarify requirements via Socratic interview (ambiguity resolution) +2. Generate PRD with goals, features, acceptance criteria +3. Decompose into tracks using MECE decomposition +4. Create work breakdown per track (task-id blocks with depends_on/blocks) +5. Generate supporting documents (ADR, API contract, test strategy, UI spec, data model) +6. Produce execution order based on dependency graph +7. Write track READMEs and work catalog +8. Validate completeness via FDE checklist ## Knowledge diff --git a/platform/skills/verify/SKILL.md b/platform/skills/verify/SKILL.md index 91b71de..cd391bf 100644 --- a/platform/skills/verify/SKILL.md +++ b/platform/skills/verify/SKILL.md @@ -1,12 +1,30 @@ --- -name: quorum:verify -description: "Run all 8 done-criteria checks (CQ/T/CC/CL/S/I/FV/CV) with 20 analysis tools and 5-language registry. Deterministic tools only — zero LLM tokens. Triggers on 'verify', 'self check', 'pre-audit', '검증', '자가 검증'." -model: haiku +name: quorum-verify +description: "Run all 8 done-criteria checks (CQ, T, CC, CL, S, I, FV, CV) using 20 deterministic analysis tools and a 5-language registry. Zero LLM tokens — pure static analysis covering code quality, tests, coverage, compliance, security, integration, functional verification, and convergence. Use when preparing evidence for audit submission, validating implementation completeness, or running a pre-commit quality gate. Triggers on: 'verify', 'self check', 'pre-audit', 'done criteria', '검증', '자가 검증'." +user-invocable: true +triggers: + - verify + - self check + - pre-audit + - done criteria --- # Verify -8-category implementation verification. Deterministic only. +8-category implementation verification using deterministic tools only — no LLM tokens consumed. + +## Workflow + +1. Identify changed files and their language (TypeScript, Python, Go, Rust, Java) +2. Run CQ (code quality): eslint, type-safety scan (`as any` per KLOC) +3. Run T (tests): execute test suite, check line + branch coverage +4. Run CC (code coverage): coverage map per file +5. Run CL (compliance): license scan, PII detection +6. Run S (security): OWASP pattern scan, secrets detection (gitleaks) +7. Run I (integration): dependency graph, blast radius analysis +8. Run FV (functional verification): FVM matrix validation +9. Run CV (convergence): integrity check across all categories +10. Aggregate pass/fail per category and report results ## Knowledge