Jules MCP + Skills

[TheRealAshik]

---

name: jules-delegator
description: Use this skill to delegate complex coding tasks by creating Jules sessions. Jules is an AI coding agent that can autonomously implement features, fix bugs, and make code changes across repositories.

Jules Delegator

Delegate coding tasks to Jules (autonomous AI coding agent by Google Labs, powered by Gemini) via MCP tools.

When to Use

• User asks to delegate a task to Jules
• Task requires autonomous multi-file implementation
• User wants a Jules session for a repo

Workflow

1. Ask Session Type

Always ask first: "Repoless session or a specific repository?"

• Repoless: context provided in prompt, no repo needed
• Repo session: use current or user-specified repo

2. Get Repo Info (if needed)

git remote get-url origin   # get repo URL
git branch --show-current   # get current branch

Extract owner/repo from the URL.

3. Create Session (create_session)

Param	Notes
repo	owner/repo — omit for repoless
branch	target branch — omit for repoless
prompt	clear, detailed task description
title	short descriptive title
interactive	false = auto-run, true = review plan first
autoPr	true = auto-create PR on completion (repo sessions only)

4. Monitor

• get_session_state — check progress
• get_code_review_context — review changes
• send_reply_to_session — send feedback if Jules is blocked

Prompt Template

## Role
You are **[Agent Name]** – a **[domain]** agent.
**Mission:** [goal]

## Boundaries
✅ Always: [safe actions]
⚠️ Ask: [actions needing approval]
🚫 Never: [prohibited actions]

## Process
1. 🔍 Observe – [what to look for]
2. 🎯 Select – [how to choose task]
3. 🛠️ Implement – [how to execute]
4. ✅ Verify – [checks to run]
5. 🎁 Present – [output format]

Jules Reference

How Jules Works

1. Clones repo into a secure Google Cloud VM
2. Reads and understands the full codebase
3. Forms a plan → you approve/modify
4. Executes asynchronously; internal Critique Agent reviews for intent, security, edge cases, and test coverage
5. Opens a PR when done → you review & merge

Planning Modes

Mode	Behavior
Interactive Plan	Asks clarifying questions before executing
Review Plan	Shows plan for approval before any code changes
Plan and Go	Forms plan and executes immediately

Agent Tools

Tool	Purpose
initiate_memory_recording	Start/load persistent session memory
read_file	Read any file from cloned repo
edit_file	Modify files with diff-based changes
google_search	Search docs, changelogs, library info
request_plan_review	Submit plan for approval
replace_with_git_merge_diff	Apply changes via git merge diff
pre_commit_instructions	Define pre-commit hooks/validation
run_in_bash_session	Run shell commands (persistent session)
run_terminal_command	Run single shell commands
submit	Complete task, generate PR, request review

🌐 jules.google · Docs · API