Pipeline skills reference
Pipeline skills execute sequentially to transform requirements into running code. Each skill produces artifacts consumed by the next stage.Pipeline flow
requirements-engineer
Professional requirements engineering based on SWEBOK v4 Chapter 1.Invocation
Modes
Operation modeOptions:
elicit— Interactive requirements gatheringaudit— Quality assessment of existing requirementsformat— Convert between requirement formats
Input
Path to existing requirements document (audit/format modes only)
Output artifacts
requirements/REQUIREMENTS.md— Structured requirements in EARS format- Sections: Functional, Nonfunctional, Constraints, Traceability
EARS syntax
All requirements use Easy Approach to Requirements Syntax:THE <system> SHALL <behavior>Example: THE system SHALL store all data encrypted at restWHEN <trigger> THE <system> SHALL <behavior>Example: WHEN a user submits login credentials THE system SHALL validate them within 2 secondsWHILE <state> THE <system> SHALL <behavior>Example: WHILE the system is in maintenance mode THE system SHALL reject all write operationsIF <condition> THEN THE <system> SHALL <behavior>Example: IF the database connection fails THEN THE system SHALL retry 3 times with exponential backoffQuality gates
Each requirement is validated against:- Unambiguous — Single interpretation only
- Testable — Concrete acceptance criteria
- Atomic — Single decision per requirement
- Binding — Stakeholder-confirmed necessity
- True — Represents actual need, not solution
specifications-engineer
Transforms requirements into formal technical specifications.Invocation
Modes
Operation modeOptions:
analyze— Analyze requirements for specification readinesscreate— Generate specifications from requirementsvalidate— Validate existing specificationsbrownfield— Incremental specification for existing code
Input
Path to requirements document
Output artifacts
Generates completespec/ directory:
Domain model and business rulesFiles:
01-GLOSSARY.md— Ubiquitous language02-ENTITIES.md— Domain entities03-VALUE-OBJECTS.md— Value objects and enums04-STATES.md— State machines05-INVARIANTS.md— Business invariants (INV-XXX-NNN)
Use case specificationsFormat:
UC-NNN-{slug}.md per use caseSections: Actor, Preconditions, Main Flow, Alternative Flows, Exception Flows, PostconditionsMulti-step process flowsFormat:
WF-NNN-{slug}.md per workflowAPI and event contractsFiles:
API-{module}.md— REST/GraphQL contracts per moduleEVENTS-{module}.md— Domain eventsPERMISSIONS-MATRIX.md— RBAC matrix
Test specificationsFiles:
BDD-UC-NNN.md— BDD scenarios per use casePROPERTY-TESTS.md— Property-based test specs
Non-functional requirementsFiles:
PERFORMANCE.md— Performance targets (p99, throughput)LIMITS.md— Rate limits, quotas, thresholdsSECURITY.md— Security requirementsOBSERVABILITY.md— Logging, metrics, alerting
Architecture Decision RecordsFormat:
ADR-NNN-{slug}.md per decisionClarification markers
When ambiguities cannot be resolved during generation:spec/CLARIFICATIONS-PENDING.md until resolved.
spec-auditor
Systematic specification quality assurance and defect detection.Invocation
Modes
Operation modeOptions:
audit— Detect specification defectsfix— Apply corrections from answered auditfocused— Audit only changed documents
Defect categories
Ambiguities — Terms admitting multiple interpretationsSignals: “appropriate”, “reasonable”, “normally”, missing quantifiers
Implicit rules — Behavior given as understood but not documentedSignals: Validations not specified, order assumed
Dangerous silences — Cases not covered causing undefined behaviorSignals: Flows without error handling, states without exit transitions
Semantic ambiguities — Same term with different meaningsSignals: Synonyms not controlled, glossary violations
Contradictions — Specifications conflicting between documentsSignals: Different values for same parameter, incompatible flows
Incomplete specifications — Missing documents or placeholder sectionsSignals: TODOs, TBD sections, empty sections
Weak invariants — Critical rules without formal invariantSignals: Restrictions in text without INV-ID, invariants without validation
Evolution risks — Designs hindering future changesSignals: Hardcoded values, closed enums, missing versioning
Missing ADRs — Architectural decisions without documentationSignals: Technologies mentioned without justification
3C verification protocol
Spec-level verification dimensions:Spec coverage checks
- Every REQ traces to at least one spec
- No orphan specs without REQ
- All spec subdirectories populated
- No placeholder sections remain
- Traceability chain intact
Spec-requirement alignment
- Semantic match with REQ intent
- No contradictions between specs
- All INV codes valid
- State transitions consistent
- Permission alignment
Cross-spec consistency
- Glossary adherence
- Terminology uniformity
- Valid cross-references
- Value consistency
- Format consistency
Output artifacts
Baseline of accepted/deferred findings
Audit report with findings by category and severity
(Fix mode) Corrections plan with before/after
plan-architect
Generates FASE files and implementation plans from specifications.Invocation
Modes
Operation scopeOptions:
global— Generate all plan artifactsper-fase— Generate plan for single FASEskip-clarify— Use existing clarificationsregenerate-fases— Force FASE regenerationaudit-fases— Read-only coverage check
FASE files
FASE (implementation phase) files are navigation indices mapping specs to incremental phases.Implementation phase indexSections:
- Objetivo — What the phase enables
- Criterios de Éxito — Acceptance checklist
- Specs a Leer — Organized by type (UCs, ADRs, Domain, Contracts)
- Invariantes Aplicables — Relevant invariants
- Contratos Resultantes — Endpoints and events to implement
- Verificación — Validation commands
- Alcance — Incluye/Excluye boundary table
Clarification taxonomy
10 categories of implementation gaps requiring clarification:Technology stack — Missing language/framework choices
Physical data model — Logical-to-physical mapping gaps
Architecture topology — Undefined deployment strategy
Security implementation — Specs without library/pattern
Integration patterns — Undefined external protocols
Performance strategy — NFR targets without strategy
Test implementation — Missing test framework
Build & deploy — Missing CI/CD definition
Observability — Missing monitoring/logging
Cost & resources — Missing infrastructure cost estimate
Output artifacts
FASE navigation indicesFiles:
README.md— Coverage matrices and dependency graphFASE-N-{slug}.md— One per implementation phase
Master implementation planSections:
- Technical Context
- Component Decomposition
- Cross-FASE Concerns
- Risk Assessment
- Developer Quickstart
- Validation & Traceability
Interactive clarification session log
Architecture viewsViews:
- C4 System Context (L1)
- C4 Container Diagram (L2)
- C4 Component Diagram (L3)
- Deployment View
- Physical Data Model
- Integration Map
- Security Architecture
Per-FASE implementation detailsSections:
- FASE-specific technical decisions
- Component implementation details
- API implementation notes
- Data changes (migrations)
- Test strategy
- Test Coverage Map (source → test mapping)
task-generator
Decomposes implementation plans into atomic, reversible tasks.Invocation
Task ID convention
TASK-F{N}-{SEQ}Examples:TASK-F0-001— First task of FASE 0TASK-F3-015— 15th task of FASE 3
Task markers
Parallelizable — Can execute concurrently with other [P] tasks
Tasks that depend on this task completing
Tasks that must complete before this task
Task documents already-implemented code (from reverse-engineer)
Task generated by req-change cascade
Task structure
Each task entry includes:Pre-generated conventional commit messageFormat:
{type}({scope}): {description}Types: feat, fix, refactor, test, chore, docs, ci, perfConcrete acceptance criteria from spec
Traceability references (FASE, UC, ADR, INV, REQ)
Revert strategy and impact assessmentCategories: SAFE, COUPLED, MIGRATION, CONFIG
Review checklist for human reviewers
Output artifacts
Global index of all tasks across all FASEs
Implementation order with dependency graph and critical path
Tasks for FASE N organized by phase:
- Setup — Project structure, dependencies
- Foundation — Shared infrastructure
- Domain — Entities, value objects, services
- Contracts — API routes, handlers, events
- Integration — Wiring, event handlers
- Tests — Unit, integration, BDD scenarios
- Verification — End-to-end validation
task-implementer
Implements code from task documents with test-first development.Invocation
Execution phases
Context loadingLoads: CLAUDE.md, glossary, entities, states, invariants, FASE file, plan, task document
Readiness gatesValidates: task document exists, plan exists, FASE exists, glossary exists, specs exist, git clean, tools available
Task selection & orderingDetermines execution order respecting dependencies and [P] markers
Pre-implementation designReads specs, extracts contracts, identifies invariants, analyzes integration
Test-first constructionWrites failing tests for acceptance criteria, exception flows, and invariants (RED phase)
ImplementationWrites code to pass tests (GREEN phase), then refactors (REFACTOR phase)
Quality verificationExecutes review checklist, runs full test suite, verifies build and lint
Atomic commit & SHA captureStages task files, commits with pre-generated message, captures SHA for traceability
Progress trackingUpdates task checkbox, reports progress with SHA, evaluates next task
FASE checkpoint & verificationCreates git tag, verifies Criterios de Éxito, runs full test suite, checks coverage
Pause conditions
Implementation pauses and asks user when:Spec does not define behavior clearly
[DECISION PENDIENTE] marker found in specSpec and plan contradict
Build fails after task implementation
Previously passing test now fails
Implementation feedback
When spec-level issues are discovered during implementation:Spec-level issues found during implementationIssue types: AMBIGUITY, CONFLICT, MISSING-BEHAVIOR, INCORRECT-CONTRACT, STALE-DECISIONSeverities: BLOCKER, WARNINGFed to
req-change for spec correctionsOutput artifacts
Implementation code following project structure
Unit tests, integration tests, BDD scenarios
Updated with completed task checkboxes (
- [x])(if spec issues found) Spec-level feedback for req-change
Checkpoint tags:
fase-N-setup, fase-N-domain, fase-N-verifiedtest-planner
Generates comprehensive test plans and coverage maps.Invocation
Output artifacts
Master test planSections:
- Test strategy by type (unit, integration, e2e, performance, security)
- Coverage targets per FASE
- Test environment requirements
- Test data management
- CI/CD integration
Test coverage matrix per domain areaColumns: Requirement, Use Case, Test Type, Test File, Coverage Status
(if applicable) Performance test scenarios with load profiles and acceptance criteria