evaluate

name	evaluate
description	Quick one-shot evaluation of work against quality criteria. Generates criteria, runs the round-evaluator, and presents the verdict. Lighter than /refine — does not auto-improve, just evaluates. Use for pre-PR quality checks or getting a critical assessment.
user-invocable	true
argument-hint	<path-to-deliverable> [--criteria 'custom criteria']
allowed-tools	Read, Grep, Glob, Bash, Agent

Quick Evaluation

Run a one-shot quality evaluation of a deliverable. No improvement loop — just the verdict.

Steps

1. Accept the deliverable path and optional custom criteria.
2. Ensure the workspace is a git repository (required for agent worktree isolation). If not, warn the user to relaunch via launch-claude-tmux.sh.
3. If no criteria provided, call generate_eval_criteria (quality-tools MCP).
4. Present criteria to the user for confirmation.
5. Spawn the round-evaluator agent with:
   • Deliverable path(s)
   • Evaluation criteria
6. Wait for the round-evaluator to complete.
7. Read and present:
   • verdict.json — scores and verdict (iterate/converged)
   • critique_packet.md — the Evaluation Summary section
   • next_tasks.json — top improvement tasks (if iterate)
8. Summarize: What's strong, what needs work, and the recommended next steps.

When to Use

• Pre-PR quality check: "Is this ready to submit?"
• Quick assessment: "How good is this, honestly?"
• Decision point: "Should I keep iterating or ship it?"

For full iterative improvement, use /refine instead.