$crank — Autonomous Epic Execution (Codex Tailoring)
This override captures the Codex-native execution model for multi-issue epic work.
Core Model
- The lead agent owns wave planning, validation, and issue state transitions.
- Workers are spawned with
spawn_agent(...). - Each worker gets a bounded task, explicit owned files, and validation steps.
- The lead agent uses
wait_agent(...)only when integration is blocked on a worker result.
Codex-Native Wave Loop
Step 1: Select the wave
Use bd ready --json when available, or the current plan artifact when beads is unavailable. Only include unblocked work with disjoint file ownership in the same wave.
Step 2: Prepare worker packets
For each issue, include:
- issue id and subject
- short implementation brief
- owned files
- validation commands
- reminder that the worker is not alone in the repo
Step 3: Spawn workers
Use one spawn_agent(...) call per issue with agent_type="worker".
Step 4: Monitor and correct
Collect worker results with wait_agent(...). If a worker needs rework, use send_input(...) with a focused correction or launch a new wave for the remaining work.
Step 5: Validate centrally
The lead agent reviews the combined diff, runs repo-level validation, and only then closes issues or advances the wave.
Step 6: Repeat
Continue until all children are closed or the epic is genuinely blocked.
Constraints
- Do not rely on batch-spawn, background-report, or timeout-specific wait APIs that are not present in Codex.
- Do not let workers commit, push, or mutate issue state unless the task explicitly requires it.
- Serialize overlapping file ownership into later waves instead of risking merge conflicts.