Orchestrate

---
name: orchestrate
description: "Use when you need the Codex Copilot equivalent of Claude Copilot /orchestrate: plan stream-based work, validate dependencies, and coordinate explicit user-approved Codex subagent execution."
---

# Orchestrate

Coordinate parallel work in a Codex-native way.

## Boundary

Codex Copilot does not auto-spawn workers unless the user explicitly asks for delegation or parallel execution.

## Workflow

1. Use `ta` to create a PRD, task breakdown, streams, dependencies, and file ownership.
2. Validate stream dependencies and file overlap with `scripts/orchestrate-validate.py` before any parallel work begins.
3. Use git worktrees only after confirming the branch/worktree plan with the user.
4. Spawn Codex subagents only when explicitly requested.
5. Keep write scopes disjoint and route each implementation stream through `qa`.
6. Treat merge, cleanup, and removal operations as separate actions requiring explicit current approval when destructive.
7. For a user-requested cost-capped non-interactive worker, pass
   `--max-budget-usd` through `tc worker`; do not imply that stored metadata
   enforces a cap unless the installed `tc` contract says so.

## Stream Metadata

Each stream should declare:

```json
{
  "streamId": "Stream-A",
  "streamName": "Foundation",
  "files": ["src/example.ts"],
  "streamDependencies": [],
  "streamPaths": ["src/example/**"],
  "streamTokenBudget": 2500
}
```

## Launch Template

When the user explicitly approves delegation, spawn each stream with:

- specialist role and skill
- task id and stream id
- owned files or path globs
- forbidden paths
- expected work product and QA route

## Output

- stream plan
- dependency graph
- file ownership map
- launch instructions or local execution plan