ADR 002: Sprint context bus files

Status

Accepted

CodyMaster already had a shared context bus (context-bus.json) for multi-skill chains. Feature work needed:

A named pipeline (brainstorm → … → retro) with explicit current step and artifacts on disk.
Human escape hatches (skip a step, reset sprint) without corrupting audit trails.
MCP and CLI reading the same state.

Persist sprint state under .cm/sprint/:

state.json — pipeline definition, current_index, completed[], skipped[] (v2 schema), timestamps, artifacts_dir.
events.jsonl — append-only events (e.g. skip, step transitions) for forensics.
artifacts/ — per-step markdown or other files produced by the workflow.

Implementation: src/sprint-pipeline.ts. CLI: cm sprint init|dry-run|skip|reset|… (see src/cli/commands/engineering.ts).

Skipping a step writes a stub artifact, advances index, appends a skip event — does not silently delete history.

Positive: Reproducible handoffs between agents and humans; state is grep-friendly.
Positive: cm suggest and skills can infer “where we are” from state.json.
Negative: Two concepts (root bus vs sprint folder) — docs must point authors to the right file for their use case.