1.5 KiB
1.5 KiB
Sprint 123 Plan: SLM Debug Readiness and Guarded Automation
Context
Sprint 121-122 established deterministic failure capture, clustering, context, patch proposal, and repro persistence. Sprint 123 adds decision-grade metrics, guarded execution tooling, and readiness scoring so small models can be routed safely into autonomous debug loops.
Goals
- Add regression guard MCP access for autonomous patch flows
- Add repro packet JSONL export for training/eval pipelines
- Add benchmark harness and readiness scoring for SLM debug routing
- Keep all outputs deterministic and machine-readable
Steps
Step 1469: Failure trend tracker model (12 tests)
Step 1470: Patch candidate ranking with history penalties (10 tests)
Step 1471: Repro command runner facade (mockable) (10 tests)
Step 1472: whetstone_regression_guard MCP tool (8 tests)
Step 1473: whetstone_export_repro_jsonl MCP tool (8 tests)
Step 1474: whetstone_get_debug_metrics MCP tool (8 tests)
Step 1475: Debug benchmark harness for fixture suites (8 tests)
Step 1476: SLM debug readiness scoring model (8 tests)
Step 1477: whetstone_get_slm_debug_readiness MCP tool (8 tests)
Step 1478: Sprint 123 integration summary + regression (8 tests)
Architecture Gate
- Routing decisions must be explainable with metric components
- Guard and readiness outputs must be deterministic
- Repro exports must be stable and append-friendly
- Max 600 lines per header