STARTER_CHARACTER = 💎

Prerequisites

Agent teams must be enabled in settings:

{
  "env": {
    "CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS": "1"
  }
}

If not set, offer to add it before proceeding.

Setup

If $ARGUMENTS provided, use as target path. Otherwise ask for:

Target path (files or folder to refactor)
Test command to verify changes

Verify the target path exists and tests pass before proceeding.

Launch the Team

Generate a short random ID: head -c 3 /dev/urandom | xxd -p | head -c 3

Use it to name the teammates:

Worker: worker-ID (e.g. worker-a3f)
Reviewer: reviewer-ID (e.g. reviewer-a3f)

Read the spawn prompts:

Worker: references/worker-prompt.md
Reviewer: references/reviewer-prompt.md

Before spawning, replace these placeholders in both prompts:

TARGET_PATH → actual target path
TEST_COMMAND → actual test command
LENSES_DIR → ${CLAUDE_SKILL_DIR}/references/lenses
GUIDES_DIR → ${CLAUDE_SKILL_DIR}/references/reviewer-guides
WORKER_NAME → the worker's name (e.g. worker-a3f)
REVIEWER_NAME → the reviewer's name (e.g. reviewer-a3f)

After Launch

Tell the user:

Shift+Down cycles between worker and reviewer
For split panes: set teammateMode: "tmux" in settings

Monitor Progress

When a worker goes idle, read .refactoring-state to check the lens number. If it skipped a value, message the reviewer to go back and complete the skipped lens first.

Third-party checks

No alerts6d3 checks · Risk SAFE

Gen Agent Trust Hub6d
The skill facilitates an automated, iterative refactoring workflow using a worker-reviewer agent team. It processes local source code and executes user-provided test commands to verify changes. The skill operates using local scripts and standard development tools without any detected malicious behaviors or obfuscation.
Socket6d
No alerts
Snyk6d
Risk: LOW · No issues

/refactoring-team

SKILL.md