Writing Plans

Settle every implementation decision before handoff. The executor writes the code; the planner specifies the contracts, steps, patterns, and proof that remove the need to choose an approach. Keep production code and complete test bodies out of the plan; include signatures, data examples or concise algorithms only when needed to settle ambiguity.

Inputs: requirements, target repository, the current issue or document, and scope constraints. Output: one current implementation plan bound to requirements and an inspected revision, with resolved decisions, exact paths/symbols, ordered steps, verification mapping and a blocking readiness result.

Writes: the ## Implementation Plan comment on the requirements issue, or the plan section in the single canonical local document, within the authorized publication scope. No implementation files and no autonomous executor dispatch. Reads repository and tracker. Ask only for missing authority or consequential intent; authorization to plan this issue already covers that scope, so skip repeated approval prompts. If a coordinating process owns publication, return a draft to it.

Optional companions: prd-quality-gate (canonical plan template and execution-readiness check) and executing-plans (scripts/plan-header.mjs). Neither ships with this skill. If unavailable, use the checklist below as the readiness check and record the base commit and a hash of the requirements text by hand.

Contract

Inputs:

• Requirements, target repository, current issue/document, and scope constraints.
• Harness-selected planner and authorized publication scope.