ralphinho-rfc-pipeline

A structured pattern for decomposing large features into independently verifiable work units, executing them in parallel across multiple agents, and integrating them through quality gates and merge queues. Use this when a feature is too complex for a single agent pass and requires coordinated, staged delivery.

• Decomposes features into a directed acyclic graph (DAG) of work units with explicit dependencies
• Assigns units to agents with scope, acceptance tests, risk level, and rollback plans
• Enforces a quality pipeline: research → plan → implement → test → review → merge-ready
• Manages a merge queue that prevents merging units with unresolved dependencies and re-runs integration tests after each merge
• Handles unit stalls by evicting, snapshotting findings, and regenerating narrowed scope
• Produces RFC execution logs, unit scorecards, dependency graphs, and integration risk summaries

1. 1.Define an RFC describing the feature and its constraints
2. 2.Decompose the RFC into work units using the unit spec template (id, depends_on, scope, acceptance_tests, risk_level, rollback_plan)
3. 3.Classify units by complexity tier (1: isolated edits; 2: multi-file changes; 3: schema/auth/perf/security)
4. 4.Assign units to agents and execute the quality pipeline for each: research → implementation plan → implementation → tests → review → merge-ready report
5. 5.Queue units for merge respecting dependency order; rebase on latest integration branch before each merge
6. 6.Re-run integration tests after each queued merge; evict stalled units and regenerate narrowed scope if needed
7. 7.Collect outputs: RFC execution log, unit scorecards, dependency graph snapshot, and integration risk summary