sprintpilot-plan-sprintlisted

## STOP — read this entire file before doing anything This skill writes the authoritative sprint plan at `_bmad-output/implementation-artifacts/sprint-plan.yaml`. The plan controls which story Sprintpilot runs next and in what order, with dependency-aware priorities and optional external-tracker links. Follow **`./workflow.md`** verbatim. The 14 steps it lists are not suggestions — each one shells out to a primitive (`infer-dependencies.js`, `resolve-dag.js`, `sprint-plan.js`) that validates inputs and updates the plan atomically. Skipping a step or reordering them produces an inconsistent plan that downstream phases (autopilot queue resolver, reorder validator) will reject. ### Never improvise - Never infer dependencies in chat — always pipe LLM JSON through `node _Sprintpilot/scripts/infer-dependencies.js dry-run --epic <id>` (or `--cross-epic`) first. The script enforces schema, unknown-key, self-dep, cross-epic-isolation, missing-rationale, and cycle checks. Hand-edited envelopes that bypass dry-run leave a plan that fails validation on the next read. - Never write directly to `sprint-plan.yaml` — route every mutation through `node _Sprintpilot/scripts/sprint-plan.js` so atomic write, schema validation, and `generated`/`auto_inferred_at` stamping happen. - Never call an LLM "to plan the sprint" outside this skill. The skill IS the LLM-driving layer; scripts never call models. ### Invocation modes The skill responds to three callers: 1. **User-direct