| name | jules-executor |
| description | Execute implementation plans by dispatching tasks to Jules (Google's AI agent). Use when user invokes /jules, has a plan file with tasks for Jules, or wants Jules-first development. Handles plan parsing, session creation, polling, question handling, and auto-merging PRs. |
Jules Executor
Dispatches tasks from a markdown implementation plan to Jules AI sessions. Tasks marked executor: claude are handed off via superpowers:subagent-driven-development.
Steps
Follow these steps when the user invokes /jules <plan-file>:
-
Verify prerequisites
- Confirm
<plan-file> argument was provided. If not, ask the user for the plan file path.
- Run
<skill-path>/scripts/jules-run jules_cli.py auth to verify Vault and API access.
-
Launch the plan
- Run
<skill-path>/scripts/jules-run run_plan.py <plan-file> in interactive mode.
- This parses the plan, creates Jules sessions respecting
depends: ordering and concurrency limits, polls for progress, and auto-merges completed PRs.
- Present the dashboard output to the user after each poll cycle.
-
Handle executor: claude tasks
- Tasks marked
executor: claude are not sent to Jules.
- When their dependencies are met, dispatch them using
superpowers:subagent-driven-development with the task body as the prompt.
-
Handle Jules questions
- When a task shows
NEEDS_INPUT, Jules has asked a question.
- If the answer is in the plan body, AGENTS.md, or codebase context, use
<skill-path>/scripts/jules-run jules_cli.py chat --session-id <id> to answer it.
- If it requires a judgment call or credentials, escalate to the user.
-
Report results
- When all tasks reach a terminal state, present the final summary to the user.
- Flag any failed tasks with their error messages.
Troubleshooting
See README.md for CLI usage and references/troubleshooting.md for common issues.
