Skip to content

execution.md

Latest commit

 

History

History
60 lines (35 loc) · 2.67 KB

execution.md

File metadata and controls

60 lines (35 loc) · 2.67 KB
Raw

DOCS.md > Execution

Execution

One command handles all execution for a set, with built-in crash recovery and re-entry.

/rapid:execute-set <set-id> [--gaps]

Runs one rapid-executor agent per wave, processing waves sequentially. Within each wave, the executor implements all tasks from the wave's PLAN.md file, committing atomically per task.

Execution flow

  1. The skill reads all PLAN.md files to determine which waves exist and their order.
  2. For each wave, one rapid-executor agent is spawned with the wave's PLAN.md as input.
  3. The executor implements tasks in order, committing each completed task.
  4. A WAVE-COMPLETE.md marker is created after each wave finishes.
  5. After all waves complete, a lean rapid-verifier agent checks that set objectives are met.

Artifact-based completion detection

The executor determines what work is done by reading planning artifacts -- not from stored state. It checks which PLAN.md files have corresponding implementation commits via WAVE-COMPLETE.md markers and git log. This design makes execution fully re-entrant.

Re-entry after crash

On every invocation, the skill scans planning artifacts to classify each wave:

  • Waves with WAVE-COMPLETE.md markers are skipped (already done)
  • The first incomplete wave resumes from its last committed task
  • Subsequent waves execute normally

Re-running /rapid:execute-set after an interruption picks up exactly where things left off. No manual state recovery is needed.

Pause and resume

For intentional interruptions, use /rapid:pause to save execution state to HANDOFF.md. This records the current wave/job progress and any user notes. Later, /rapid:resume restores context and continues execution from the checkpoint.

Solo mode

When a set is configured for solo mode (no worktree), execution works directly on main. The diff base is the startCommit hash recorded when the set was started. After all waves complete and verification passes, solo sets auto-transition to merged since there is no branch to merge separately.

Gap-closure mode

The --gaps flag enables execution of gap-closure waves on merged sets. Gap-closure waves are numbered after existing waves and never re-execute previously completed work. The set remains in merged status throughout.

State transitions

  • planned -> executed (when execution starts)
  • executed -> complete (after verification passes)
  • complete -> merged (automatic for solo sets)

Agents spawned

  • 1 rapid-executor per wave (sequential)
  • 1 rapid-verifier after all waves complete

See skills/execute-set/SKILL.md for full details.

Next: Review