completely-run

name	completely:run
description	Drive the Beads queue (`bd ready`) with one engine in two autonomy modes — supervised (GSD wave subagents, human gates at phase boundaries) or unattended (Ralph-style fresh-context loop, stops when the queue is empty). Quality gates + the default-FAIL evaluator run underneath both. Use to execute planned work. Backed by `cmpl run`.
version	0.2.0
user-invocable	true
argument-hint	[--mode supervised|unattended] [--max N] [--dry-run]

Execute planned work from Beads with the right autonomy level.

Pick the mode (the dial): run unattended only when the spec is unambiguous, tasks are atomic, write-zones are disjoint, and verification is automated. Otherwise run supervised.

• Unattended (cmpl run / cmpl run --max N): fresh claude -p per iteration reading the Beads-aware overlay prompt; ONE task per iteration; stops when bd ready is empty. No --dangerously-skip-permissions — the guard hook + allowlist gate dangerous ops.
• Supervised (cmpl run --mode supervised): hand off to /gsd:execute-phase <phase> (wave subagents, checkpoints). Either way, status lives in Beads and evidence in bd comment.

Use --dry-run first to see how many tasks are ready and what would execute, without running. Each iteration: claim one task → TDD → gates → evidence comment → bd close → next. Blocked → bd ... --status blocked + comment, never a silent stub.