miudog/agentic-workflow-setup
Intent-routed DEFINE, PLAN, BUILD, VERIFY, and lean-context skills with measurable local evaluation.
Diagnose a reproducible failed code check or unexpected program behavior, then make an authorized bounded fix using Red → Yellow → minimal Green fallback. Use only after a failure exists; skip routine verification, initial implementation, and sensory-quality review.
Implement one accepted, implementation-ready module slice inside its Task Packet write boundary. Use for BUILD after product and architecture decisions are fixed; skip unresolved design, test-author-only work, and failure diagnosis.
Reduce controllable development context through current-truth capsules, bounded retrieval, low-coordination delegation, and evidence-scaled verification. Use for long sessions, handoffs, context pressure, or workflow-cost reduction; skip ordinary small edits that already have sufficient context.
Design or maintain one module's architecture and current-stage slices from an accepted READY specification. Use for architecture or task breakdown; skip unresolved product intent and implementation-ready work.
Define an unresolved product or small-system idea as one accepted current specification before architecture or implementation. Use for new behavior with open product decisions; skip accepted or implementation-ready work.
Author the minimal deterministic code checks for an accepted risky BUILD slice as an independent Test Author. Use for requested tests, regressions, persistence or transaction risk, cross-language contracts, or behavior humans cannot reliably verify; skip sensory evaluation and routine implementation.