1696b92bb8
Sprints 46-58 implement the cross-language porting foundation: language-to-IR adapters, equivalence checking, gate validation, legacy ingestion, managed/dynamic families, low-level/logic-actor semantics, debug workflow tooling, AST-native family tools, Rust/CPP raising tools, system-level orchestration, query family, and porting gates. Sprint 59 adds the governance layer (policy packs, review boards, waiver packets, ambiguity triage, decision ledger) with the whetstone_review_porting_decision MCP tool. Also includes: sprint plans 46-130, MCP taskitem pipeline scripts, CLAUDE.md, docs, and full test matrix (steps 689-828). Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
63 lines
1.6 KiB
Markdown
63 lines
1.6 KiB
Markdown
# 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
|
|
|
|
## 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/`
|
|
|
|
## 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
|