multi-codex-orchestrator

# Multi Codex Orchestrator Use this skill when the user explicitly wants multiple Codex workers running as independent agents, rather than relying on built-in sub-agent behavior. ## When To Use - The task can be split into mostly independent workstreams. - Each worker can own a bounded path set or responsibility. - The coordinator can validate and integrate outputs at the end. ## When Not To Use - The task is small enough for one agent. - Several workers would need to edit the same core files. - The task is highly sequential and cannot benefit from parallelism. ## Default Execution Model 1. Create a plan file from `assets/templates/plan.yaml`. 2. Give each worker a clear `id`, `role`, `paths`, and `goal`. 3. Use `scripts/spawn_agents.sh` to validate the plan and create one git worktree per worker under `/tmp/`. 4. Run each worker with `codex exec` in its own worktree. 5. Require each worker to write `result.json` in its run directory. 6. Use `scripts/collect_results.sh` to gather artifacts and diffs. 7. Integrate centrally. Do not let workers merge each other's changes. ## Hard Rules - One worker owns one write scope. - Default to isolated git worktrees, not one shared writable tree. - Workers must produce structured outputs, not just free-form prose. - The coordinator must re-validate before integrating. - If two workers need the same file, redesign the split or serialize the work. ## File Layout - `scripts/plan_tasks.sh`: copy a starter plan into a target project...