execution-plan

Trigger: execution plan, executable plan, deep plan. Draft one neutral plan for direct or SDD execution.

SKILL.md

GitHub

Contract

Create exactly one plan for the requested outcome. The plan may be small or large. Use assets/plan-template.md.

Rules

  • Use short complete sentences. Keep paths, symbols, commands, URLs, technical terms, numbers, and Markdown exact.
  • Omit empty sections and filler. Keep the artifact clear to a human; do not use cryptic A2A fragments inside it.
  • Before writing, check the exact destination. Update an existing plan only when the user supplied its exact path or the active conversation already created or selected it. On any other slug collision, ask one closed choice: reuse the existing plan or generate a new slug. Never overwrite implicitly.
  • Qualify behavior identifiers as <capability>/<requirement> when canonical specs matter. Use observable WHEN and THEN scenarios.
  • Only for explicit delivery intent, write one Delivery: line immediately after the title: no commits means working-tree, commits means commit-per-unit, and TCR means tcr. Otherwise omit it; execution defaults to working-tree.
  • Contradictory, duplicated, or unknown delivery requests block plan creation until resolved. Never infer delivery from a PR request, route, work group, or skill name.
  • Use ordered work groups with explicit dependencies. Keep one file even when groups span sessions.
  • Every work group requires Files: and Skills:. Select names with implementation-skill-routing; use names, never paths.
  • Keep delivery skills (tcr, work-unit-commits, or other Git skills) out of Work groups → Skills:.
  • Include one Execution guidance recommendation: direct or SDD. State the reason and show ejecuta el plan <path>.
  • Write artifacts in English. Only the plan file may change; Delivery authorizes later execution. Never edit production code, stage, commit, or push during planning.

Self-check

  • One file contains every required section from the template.
  • Every claim is evidenced or explicitly marked as an open question.
  • Every task is actionable and verifiable; behavior changes do not hide in tasks.
  • Every work group names its implementation skills or none.
  • Delivery, when present, appears once immediately after the title and reflects explicit user language.
  • Dependencies are explicit and acyclic.
  • Risks contain only real rollback conditions or execution hazards.

Resource

  • assets/plan-template.md