Ponytail Review — Testing & QA agent for Claude Code
Cold over-engineering reviewer.
How to install Ponytail Review
Installs to ~/.claude/agents/raditia-craftkit-ponytail-review.md
mkdir -p ~/.claude/agents && curl -fsSL https://raw.githubusercontent.com/raditia/CraftKit/HEAD/agents/ponytail-review.md -o ~/.claude/agents/raditia-craftkit-ponytail-review.md Restart Claude Code, or start a new session, for it to be picked up.
What Ponytail Review does
name: ponytail-review description: Cold over-engineering reviewer. Spawned by parallel workflows when non-test source changes, to find what to delete: reinvented stdlib, speculative abstractions, dead flexibility. Never edits files. Complexity only, not correctness or security. tools: Read, Grep, Glob model: sonnet color: yellow craftkitInject: karpathy-guidelines
You are a cold over-engineering reviewer. You do not flatter. You find what to delete.
Apply the decision ladder and simpli
Alternatives in Testing & QA
- Weed — Weed the Allium garden 3.7k ★
- Fable Verifier — Cold verifier for finished fable deliverables 837 ★
- Post Gate Agent — Review building phase implementation against plan requirements and test coverage 243 ★
Full documentation available on GitHub
View Source RepositoryRelated Agents
Deep Architect
Handles architectural decisions, novel abstractions, breaking contract changes, row-transforming migrations, a
Quick Executor
Executes small, low-risk, mechanical tasks fast — typo fixes, copy/i18n tweaks, config value changes, single-f
Cto Self Eval
CTO-grade codebase self-evaluation. Measures code health, dead code, debt density, test meaningfulness, and ov
Getff Cold Run Prober
One-beat cold-run acceptance probe for the getff-any-stack-trace umbrella (spec §9.3). Hands a fresh subagent
Hs Validator
Executes one QA validation dimension against a merged feature (build/lint/test, acceptance criteria, non-goals
Cold Reviewer
Zero-context code reviewer. Reviews changes cold - no spec, no project knowledge, no domain context. Catches w
Related Skills
Before AI
Before you reach for AI — interrogate one real workflow from first principles, delete the dead weight, then de
Ask Kilo
Get an advisory second opinion from an external non-Claude model via the kilo CLI. Prompt-only — kilo never se
QA Review
Review test automation code for OOP compliance, Allure metadata, proper abstractions, and flaky patterns. Invo