name: workflow-orchestrator description: Orchestrate multiple Codex CLI workflows using done-signal files, optional locks, and simple dependencies. metadata: short-description: Orchestrate CLI workflows with done signals

Workflow Orchestrator

Use this skill when the user wants multiple CLI workflows to coordinate via done-signal files, optionally with file locks.

Core behavior

• Ask the user for a done-signal file naming scheme (default suggestion: /tmp/workflow-<name>.done).
• Ask the user whether any shared resources need locks (files, migration runner, build outputs). If yes, ask for lock file paths.
• Express dependencies by listing each workflow's --wait-for done files (one per dependency).
• Warn if a done file already exists from a prior run; recommend deleting it or using a run-specific name.

Scripts

Use scripts/run-workflow.sh to run a workflow with dependencies and optional locking.

Usage

run-workflow.sh \
  --wait-for /tmp/workflow-A.done \
  --done-file /tmp/workflow-B.done \
  --lock /tmp/workflow-shared.lock \
  -- ./your-workflow-command --with args

• --wait-for can be repeated for multiple dependencies.
• --done-file is touched only if the workflow exits successfully.
• --lock is optional; if provided, the script uses flock when available, otherwise a lock directory.

Recommended dependency pattern

Use sentinel files as dependency edges:

• Workflow A: --done-file /tmp/workflow-A.done
• Workflow B: --wait-for /tmp/workflow-A.done --done-file /tmp/workflow-B.done
• Workflow C: --wait-for /tmp/workflow-B.done

If the user wants more structure, propose a small text config where each workflow lists its done and wait_for files, then translate it into the script calls.