plan-to-launch

Use a workflow that requires a PRD and design doc before implementation starts. This Skill locks scope with a fingerprint of intended files and effort, then only ships tasks that map to the accepted plan, kicking back to design when code diverges.

It supports lite, default, and deep modes. Lite handles small accepted changes with a scope lock and focused plan, default covers normal features with lightweight docs, and deep applies full gates and councils for high-risk or leadership-visible work.

Human approval is required at final launch signoff and whenever discovery materially expands scope, such as doubling expected files or effort, adding trust boundaries, or exceeding two implementation PRs. The run pauses until approval, rejection, or revision is recorded.

Council levels range from none to full-council based on risk. Low-risk work records a skipped or single-lens reason, while minimal or full councils are required for multi-domain, high-risk, or conflicting-review situations and must pass before planning or merge.

The workflow stops implementation and returns to the design doc instead of patching around the problem. Implementation learnings are preserved as evidence, and material scope deltas pause the run pending human approval.

It depends on external primitives like agent-fleet council and ship, grounding-brief, and run-telemetry, so required councils block if agent-fleet preflight fails. It also adds process overhead that may be unnecessary for trivial changes beyond lite mode.