---

name: Agent Communications Skill description: Communicate with the STAR orchestrator agent. Use this to report progress, send results, or request input when working on delegated tasks.

Purpose

Enable bidirectional communication between Claude Code and the STAR orchestrator agent. When you are spawned by the orchestrator to perform a task, use this skill to:

1. Report Progress - Send status updates during long-running tasks
2. Send Results - Return your final output (code, explanations, etc.) to be displayed in the conversation
3. Request Input - Ask the orchestrator (and user) for clarification if needed

Environment Variables

The orchestrator passes these when spawning you:

• STAR_SESSION_ID - The orchestrator's session ID (for linking messages back)
• STAR_AGENT_ID - Your unique agent ID
• STAR_SERVER_URL - The server endpoint (default: http://localhost:3001)

Instructions

When to Use This Skill

• ALWAYS use report_result when you complete a delegated task
• Use report_progress for tasks that take more than a few seconds
• Use request_input if you need clarification before proceeding

Workflow

1. Check if STAR_SESSION_ID is set (indicates you were spawned by orchestrator)
2. Execute the requested task
3. Use the tools below to communicate back

Tools

Report Progress

Send a progress update to the orchestrator and user.

Copypython .claude/skills/agent-comms/tools/report.py progress "Working on visualization..."

Report Result

Send your final result. This will be displayed in the conversation.

Copypython .claude/skills/agent-comms/tools/report.py result "Here is the completed code:" --code "$(cat output.html)"

Or for multiline content:

Copypython .claude/skills/agent-comms/tools/report.py result "Task complete. Created visualization." --file output.html

Request Input

Ask for user input (the orchestrator will relay this to the user):

Copypython .claude/skills/agent-comms/tools/report.py input "Should I use Canvas or SVG for the visualization?"

Message Types

Example Session

When spawned with prompt: "Create a double-slit visualization"

Copy# 1. Acknowledge the task
python .claude/skills/agent-comms/tools/report.py progress "Starting work on double-slit visualization..."

# 2. Work on the task...
# (create files, write code, etc.)

# 3. Report the result
python .claude/skills/agent-comms/tools/report.py result \
  "Created an interactive double-slit experiment visualization. The HTML file demonstrates wave-particle duality with controls for observation mode." \
  --file double-slit.html \
  --type html

Important Notes

• Messages are delivered in real-time to both the Agent Monitor and the main conversation
• Large code blocks should use --file to avoid shell escaping issues
• The orchestrator can see all your tool usage, but explicit reports ensure results appear in the conversation
• If STAR_SESSION_ID is not set, the tools will still work but messages won't be linked to a session