LaunchKit · 2026
Back to Skills

checkmate

Enforces task completion: turns your goal into pass/fail criteria, runs a worker, judges the output, feeds back.

0
526 downloads
by @insipidpoint

Setup & Installation

openclaw skills install @insipidpoint/checkmate

Or with the ClawHub CLI, for registry-managed skill folders outside a full OpenClaw workspace:

npx clawhub install checkmate

Version History

v2.0.4Feb 22, 2026 - Version note by skill developer:

Redesign intake: goal statement (plain English) replaces checklist. Judge evaluates holistically against goal, not a spec sheet. User reviews a paragraph, not 10 bullet points.

What This Skill Does

Checkmate converts a task description into explicit pass/fail criteria, spawns a worker agent to attempt the task, runs a judge agent against the criteria, and loops with accumulated feedback until every criterion passes. An interactive mode lets you review criteria and approve each checkpoint; batch mode runs fully autonomously.

Unlike a single-shot agent run, it locks criteria upfront and uses a separate judge role each iteration, so the definition of done cannot drift and partial results cannot slip through.

When to use it

  • Running a code generation task until all tests pass
  • Producing a research report that must cover specific required topics
  • Drafting a document that must meet a defined quality checklist
  • Iterating on a data transformation script until output matches a spec
  • Automating QA loops that would otherwise require manual review cycles

Example Workflow

Here's how your AI assistant might use this skill in practice.

INPUT

User asks: checkmate: Write a Python function that validates email addresses and passes all edge cases in the provided test suite

AGENT
  1. 1Resolve session UUID and create a timestamped workspace directory
  2. 2Run intake loop to draft pass/fail criteria from the task description and present them for user approval
  3. 3On approval, spawn a worker agent session that attempts the task and writes output to iter-01/output.md
  4. 4Spawn a judge agent session that evaluates the output against locked criteria and writes a verdict
  5. 5On FAIL, extract gaps, present checkpoint to user, then loop with accumulated feedback until judge returns PASS
OUTPUT

final-output.md containing the validated function, with all criteria marked passed in the final verdict