Files
whetstone_DSL/docs/SPRINT_TASKITEM_EXECUTION_POLICY.md

2.3 KiB

Sprint Taskitem Execution Policy

This policy defines how agents must execute sprint plans using MCP taskitem tools and how run data must be captured for later LoRA curation.

Scope

  • Applies to sprint plan execution (for example: sprint46_plan.md through sprint100_plan.md)
  • Applies to all future agents working in this repository

For the practical agent workflow on when to trust generated taskitems, when to manually correct them, and when to record generator misses instead of bypassing the planning layer, also see:

docs/AGENT_PLAYBOOK_WHESTONE_FIRST_SPRINT_EXECUTION_2026-03-20.md

Required MCP Pipeline

For each sprint plan, agents must run:

  1. whetstone_architect_intake
  2. whetstone_generate_taskitems
  3. whetstone_queue_ready
  4. whetstone_validate_taskitem

Use:

tools/mcp/run_sprint_taskitem_pipeline.sh <sprint_plan.md>

Required Data Capture

After each successful run, agents must export run artifacts to LoRA capture JSONL:

tools/mcp/export_taskitem_run_for_lora.sh <run_output_dir>

Use:

training_data/lora/taskitem_pipeline_runs.jsonl

Note: Raw logs are capture material, not training-ready data. Curation is required before any fine-tuning.

Batch Execution

For multiple sprints, agents should use:

tools/mcp/run_sprint_range_with_capture.sh <start_sprint> <end_sprint>

Example:

tools/mcp/run_sprint_range_with_capture.sh 46 100

Quality Requirements

  • Keep whetstone_mcp_stable as execution binary during active development
  • Do not use unstable binaries for bulk taskitem generation
  • Preserve all run artifacts under logs/taskitem_runs/
  • Preserve JSONL append-only history under training_data/lora/

Hybrid Contract Gate

To reduce C++-first bias while keeping language-first and AST-first workflows available, run the hybrid contract validator after sprint execution:

tools/mcp/validate_hybrid_pipeline_contract.sh --start <N> --end <M>

Use strict blocking mode for enforcement:

tools/mcp/validate_hybrid_pipeline_contract.sh --start <N> --end <M> --strict --enforce-non-cpp

Reference: docs/HYBRID_PIPELINE_CONTRACT.md

Handoff Requirement

When finishing a batch, agents must report:

  1. sprint range attempted
  2. successful runs count
  3. failed runs count
  4. batch summary file path
  5. LoRA JSONL append confirmation