Cluster Assignment

Cluster Assignment

Custom skill created by Maggie Lerman.

Overview

Use this skill to convert a prioritized backlog into parallel execution clusters, each with a clean branch/worktree, tracking issue, and a concrete prompt for an agent to execute.

When To Use

• User asks to run multiple agents in parallel on one repo.
• User asks for branch/worktree setup for project clusters.
• User asks for "what should run simultaneously" across a roadmap.
• User asks for per-cluster GitHub issues and agent-ready prompts.

Required Behavior

• Read repo planning sources first (typically ROADMAP.md, CHANGELOG.md, and active/in-review project docs).
• Cluster by dependency boundaries first, not by equal sizing.
• Avoid placing streams that edit the same files in parallel unless the user explicitly accepts merge risk.
• Use branch names prefixed with codex/.
• Never run destructive git operations (reset --hard, forced cleanups) as part of setup.
• If non-obvious tradeoffs exist (for example, sequence-sensitive migrations), pause and confirm before bootstrapping.

Workflow

1. Build the candidate stream list.

• Extract active, in-review, and near-term stale backlog streams.
• Record hard dependencies and shared-file collision risk.

2. Design the cluster map.

• Group work into 2-4 dependency-safe clusters.
• For each cluster define: objective, project IDs, branch, worktree path, merge risk, and handoff gate.
• Use references/cluster-plan-template.md for output shape.

3. Bootstrap branches and worktrees.

• Prefer deterministic setup with scripts/bootstrap_worktrees.sh.
• Feed a CSV plan using references/cluster-plan.csv.
• Default base branch is main unless user specifies otherwise.

4. Create tracking issues (if gh is available and user wants GitHub tracking).

• Open one issue per cluster with scope, deliverables, and acceptance criteria.
• Include branch/worktree paths and explicit out-of-scope notes.

5. Generate per-agent prompts.

• Use references/agent-prompt-template.md.
• Keep each prompt scoped to one cluster.
• Include required checkpoints, validation commands, and docs/governance expectations.

6. Return execution-ready status.

• Provide created branches/worktrees/issues.
• List blockers and the exact first command each agent should run.

Output Contract

Return sections in this order:

1. Cluster Plan
2. Bootstrap Status
3. Tracking Issues
4. Agent Prompts
5. Next Coordination Checkpoint

References

• Plan template: references/cluster-plan-template.md
• CSV input template: references/cluster-plan.csv
• Prompt template: references/agent-prompt-template.md
• Setup helper: scripts/bootstrap_worktrees.sh