jules

name	jules
description	Delegates tasks to Google Jules, an autonomous AI coding agent. Suggest Jules for time-consuming, well-defined, or parallelizable work. Always ask before sending.

Jules CLI Skill

Interact with Jules, Google's autonomous AI coding agent, through its CLI. Jules runs asynchronously in the cloud on GitHub repositories.

When to Suggest Jules

Proactively suggest delegating work to Jules when you notice tasks that are:

• Time-consuming but well-defined — large refactors, adding tests across many files, migrating APIs
• Parallelizable — multiple independent changes that can run as separate sessions
• Async-friendly — work the user doesn't need to watch happen in real-time
• Tedious but mechanical — renaming patterns, adding error handling, updating dependencies

Always suggest, never auto-delegate. Say something like: "This looks like a good candidate for Jules - want me to send it?"

Do NOT suggest Jules for:

• Quick, simple changes you can do yourself in seconds
• Tasks requiring real-time user interaction or iterative feedback
• Work that depends on local-only state (files not in the repo, environment variables, running services)

CLI Quick Reference

# Send a task (infers repo from current directory)
jules new "<task description>"

# Send to a specific repo
jules new --repo owner/repo "<task description>"

# Start parallel sessions for the same task
jules new --parallel 3 "<task description>"

# List sessions
jules remote list --session

# Pull completed results and apply patch
jules remote pull --session <session_id> --apply

# Interactive dashboard
jules

Writing Good Task Prompts

1. State the goal clearly — what should be different when Jules is done?
2. Scope it — which files, modules, or areas of the codebase?
3. Provide constraints — testing requirements, style guidelines, things to avoid
4. Give examples if the pattern isn't obvious

Good: "Add unit tests for all exported functions in src/utils/. Use vitest. Co-locate test files with .test.ts suffix." Bad: "Fix the bugs" - too vague, no scope

Workflow

When the user agrees to delegate a task to Jules:

1. Verify prerequisites — confirm git repo with GitHub remote, jules is available, user is authenticated
2. Check repo state — warn about uncommitted changes or unpushed branches (Jules works from the remote)
3. Craft the prompt — analyze context, write a specific scoped prompt, show it to the user for approval
4. Send — jules new "<approved prompt>"
5. Report — show session info, remind about jules remote list --session and jules remote pull --session <id> --apply

Rules

• Always show the prompt before sending — never send without user approval
• Never send credentials or secrets in the session prompt
• Warn about local-only state — uncommitted changes, untracked files, env vars
• One clear task per session — don't bundle unrelated work
• Suggest parallel sessions for multiple independent tasks