mattmre/sparc-methodology

SPARC Methodology Skill

A 5-phase structured development approach that ensures correctness at each stage before proceeding.

Trigger

Use this skill when:

• Starting a complex feature with unclear requirements
• Refactoring a large system component
• Building multi-service integrations
• Onboarding to an unfamiliar codebase

The 5 Phases

Phase 1: Specification (S)

Output: phase_1_{slug}.md
Produces: Functional Requirements (FR-01..FR-N), Non-Functional Requirements (NFR-01..NFR-N), Gherkin scenarios (Given/When/Then), acceptance criteria.

Done criteria:

• All FRs are testable
• NFRs have measurable thresholds
• Gherkin scenarios cover happy path + top 3 edge cases

Phase 2: Pseudocode (P)

Output: phase_2_{slug}.md
Produces: Language-agnostic algorithm representation, data flow diagram, error handling strategy.

Done criteria:

• Pseudocode covers every FR
• Error states enumerated
• No ambiguous "do magic here" blocks

Phase 3: Architecture (A)

Output: phase_3_{slug}.md
Produces: Component diagram, interface contracts, dependency graph, ADR for key decisions.

Done criteria:

• All interfaces defined
• No circular dependencies
• ADR written for any non-obvious choice

Phase 4: Refinement (R)

Output: Working code + tests
Produces: Implementation iterating from pseudocode, test cases from Gherkin, TypeScript type-safe.

Done criteria:

• All Gherkin scenarios pass
• TypeScript compiles clean
• Truth score >= 0.90

Phase 5: Completion (C)

Output: PR + docs
Produces: PR with all sections filled, docs updated, CLAUDE.md updated if patterns changed.

Done criteria:

• CI green
• PR review checklist complete
• next-session.md updated

Usage

/sparc-spec "{feature name}"     → Runs Phase 1 only
/sparc-pipeline "{feature name}" → Runs all 5 phases sequentially with gate checks