Jules Dispatch Skill

Purpose: Automate the handoff of Beads issues to Jules agents using the

jules

Activation

Triggers:

• "assign this to jules"
• "dispatch bd-123 to jules"
• "start jules session for X"
• "run these in the cloud"

User provides: Issue IDs (bd-xyz) OR prompt text.

Core Workflow

1. Context Generation (The "Rich Prompt")

Jules needs the same context guidance as Claude Code Web. We will reuse/adapt the

parallelize-cloud-work

Prompt Structure:

TASK: {issue_title} ({issue_id})
CONTEXT:
- Repo: {repo_name}
- Branch: feature-{issue_id}-jules

🚨 INSTRUCTIONS:
1. INVOKE SKILLS: {context_skills}
2. EXPLORE: Check {files_of_interest}
3. PLAN: Don't reimplement existing logic.
4. EXECUTE:
   - Checkout branch feature-{issue_id}-jules
   - Commit with Feature-Key: {issue_id}
   - Push and create PR

2. Dispatch Logic

The skill will wrap the

jules remote new

Command Template:

jules remote new \
  --repo . \
  --session "{RICH_PROMPT}"

3. Loop & Parallelize

If multiple issues are provided:

# pseudocode
for issue_id in provided_issues:
  prompt = generate_rich_prompt(issue_id)
  jules remote new --repo . --session "$prompt"

Workflow

1. Execute Dispatch Script

# Auto-discover from current repo
python ~/.agent/skills/jules-dispatch/dispatch.py

# Dry-run (print commands without executing)
python ~/.agent/skills/jules-dispatch/dispatch.py --dry-run

# Force dispatch specific issue (ignores label check)
python ~/.agent/skills/jules-dispatch/dispatch.py --action dispatch --issue bd-xyz --force

3. Monitor

Check status of active sessions:

python ~/.agent/skills/jules-dispatch/dispatch.py --action list

4. Pull & Apply (The Pull Pattern)

When a session is complete, retrieve the code into a local feature branch:

python ~/.agent/skills/jules-dispatch/dispatch.py --action pull --session <SESSION_ID>

2. Mega-Prompt Construction

For each candidate, the skill constructs a rich prompt:

TASK: {title} (ID: {issue_id})

DESCRIPTION:
{description}

DESIGN SPEC:
{design field from Beads issue}

----------
TECH PLAN / DOCS:
{Contents of docs/{issue_id}/TECH_PLAN.md if exists}
----------

CRITICAL INSTRUCTIONS:
1. Implement exactly per the DESIGN SPEC above.
2. If the Spec is ambiguous, PAUSE and ask key questions (do not guess).

DEFINITION OF DONE (REQUIRED):
1. Create a reproduction test case (or new unit test).
2. Run `make ci-lite` (or standard test suite) and fix ALL failures.
3. If this is a UI feature, verify no console errors.
4. Your PR description must include a "Verification" section with test logs.

3. Dispatch to Jules

The skill calls the Jules CLI:

jules remote new \
  --repo <owner>/<repo> \
  --session "<mega-prompt>"

Usage Examples

Example 1: Dispatch All Ready Tasks

# From any repo with Beads
python ~/.agent/skills/jules-dispatch/dispatch.py

Output:

Found 2 candidates.
🔍 Analyzing bd-4mot: Add Jules Health Check Endpoint...
🚀 Dispatching to Jules...
✅ Dispatched bd-4mot successfully.
🔍 Analyzing bd-ijk: Implement Rate Limiting...
🚀 Dispatching to Jules...
✅ Dispatched bd-ijk successfully.

Example 2: Dry Run

python ~/.agent/skills/jules-dispatch/dispatch.py --dry-run

Output:

Found 1 candidates.
🔍 Analyzing bd-4mot: Add Jules Health Check Endpoint...
  [DRY RUN] Would execute:
  jules remote new --repo stars-end/prime-radiant-ai --session "..."
  [Prompt Length]: 842 chars

Example 3: Force Dispatch Specific Issue

python ~/.agent/skills/jules-dispatch/dispatch.py --action dispatch --issue bd-xyz --force

Example 4: Monitor Progress

# List active Jules sessions
python ~/.agent/skills/jules-dispatch/dispatch.py --action list

# Pull code from completed session
python ~/.agent/skills/jules-dispatch/dispatch.py --action pull --session 123456

Preparing Issues for Jules

1. Add Design Spec

Use Beads to add a design spec:

mcp__plugin_beads_beads__update(
  issue_id="bd-xyz",
  design="### Endpoint\nGET /health/jules\n### Response\nJSON: {'status': 'ok'}"
)

2. Add TECH_PLAN (Optional)

Create detailed context in

docs/{issue_id}/TECH_PLAN.md

# Tech Plan: bd-xyz

## Overview
...

## Implementation Details
...

## Testing Strategy
...

3. Apply Label

mcp__plugin_beads_beads__update(
  issue_id="bd-xyz",
  labels=["jules-ready"]
)

Environment Setup (Per-Repo)

Each repo needs a

scripts/jules_setup.sh

#!/bin/bash
# scripts/jules_setup.sh - Bootstraps repo toolchain for Jules

# Install toolchain via mise
curl https://mise.run | sh
export PATH="$HOME/.local/bin:$PATH"
eval "$(mise activate bash)"
mise trust
mise install --yes

# Install dependencies
cd backend && poetry install --no-interaction
cd ../frontend && pnpm install

# Generate mock .env
cat <<EOF > .env
DB_HOST=localhost
USE_MOCK_DATA=true
EOF

echo "✅ Environment ready"

Jules configuration in the Jules UI:

• Setup Script:

  bash scripts/jules_setup.sh

• Environment Variables: Add

  RAILWAY_TOKEN

  (for database access if needed)

Integration with Beads

Before dispatch:

• Issue must exist in Beads
• Recommended: Add

  design

  field and/or TECH_PLAN doc
• Apply

  jules-ready

  label

After dispatch:

• Jules creates feature branch:

  feature-{issue_id}-jules-{session_id}

• Jules commits with

  Feature-Key: {issue_id}

  trailer
• Jules creates PR on completion

Post-merge cleanup:

mcp__plugin_beads_beads__close(
  issue_id="bd-xyz",
  reason="Completed via Jules: PR#123 merged"
)

CLI Options

--dry-run

--issue ID

--force

jules-ready

--repo OWNER/NAME

Troubleshooting

"No 'jules-ready' tasks found"

Either no issues have the label, or you're not in a Beads-enabled repo.

# Check Beads is initialized
ls .beads/issues.jsonl

# Check for labeled issues
grep "jules-ready" .beads/issues.jsonl

"'jules' CLI not found"

Install the Jules CLI:

# Via npm (if packaged)
npm install -g @google/jules-cli

# Or via mise
mise use -g npm:@google/jules-cli

"Failed to dispatch"

Check Jules authentication:

jules auth status
jules auth login

"Session started but no PR created"

Check Jules session status:

python ~/.agent/skills/jules-dispatch/dispatch.py --action list

Version History

• v2.0.0 (2025-12-15): Major upgrade

  • Cross-repo support (works from any Beads-enabled repo)
  • JSONL auto-discovery for jules-ready tasks
  • Rich mega-prompt with TECH_PLAN support
  • Per-repo setup script pattern
  • Centralized in agent-skills

• v1.0.0 (2025-12-10): Initial implementation

  • Basic dispatch via bd CLI
  • Manual issue ID arguments