execute

name	execute
description	Execute a prepared Codex kit task-by-task with dependency checks, independent verification, and phase review. Use after architect creates tasks/kits/<slug>, or when asked to run or continue a kit.

Execute a Codex execution kit

Resolve the plugin root before running commands

Set POLYTROPOS_ROOT from this file's real location: in plugin mode, this file is <root>/codex/skills/execute/SKILL.md, so ascend to <root>; in a managed copied install, use the installer-resolved POLYTROPOS_ROOT="{{POLYTROPOS_ROOT}}". Reject a literal placeholder. Before shelling out, verify $POLYTROPOS_ROOT/data/pricing.codex.json and $POLYTROPOS_ROOT/bin/codex_execute.py exist. If proof fails, stop and direct the user to python3 bin/harness_select.py doctor --harness codex; never run a guessed or stale path.

Consume only a kit under tasks/kits/<slug>/. Read PLAN.md, TASKS.md, and any kit guardrails before execution, especially constraints and out-of-scope fences. Honor task dependencies, the exact pending | in-progress | done | blocked status vocabulary, and each task's runtime-resolved model pin. Do not rewrite a brief to fit an unexpected repository state.

Safe progression

Inspect without dispatching:

python3 "$POLYTROPOS_ROOT/bin/codex_execute.py" status --kit tasks/kits/<slug>
python3 "$POLYTROPOS_ROOT/bin/codex_execute.py" run --kit tasks/kits/<slug> --dry-run

The dry run prints the next dispatch and verify plan; it does not launch Codex. A real run does:

python3 "$POLYTROPOS_ROOT/bin/codex_execute.py" run --kit tasks/kits/<slug> --task <id>
python3 "$POLYTROPOS_ROOT/bin/codex_execute.py" review --kit tasks/kits/<slug> --phase <N>

A real run or non-dry review launches headless Codex and spends subscription usage or API-metered funds. Obtain the user's authority before starting it. Never represent a dry run as a real dispatch.

When interactive delegation is available, prefer the canonical kit-implementer, kit-verifier, and phase-reviewer custom agents and keep their roles separate. Plugin install does not install those agents; the headless driver remains the functional fallback. Regardless of path, rerun the task verify command independently before accepting completion and review each completed phase for plan drift.