agentsclimarketplace

Ponytail review

Skill wildcard/caro/.claude/skills/ponytail-review

Run the pragmatic-skeptic 'ponytail' reviewer over the current diff to surface over-engineering — code, abstractions, dependencies, files, or process steps the problem does not need. Read-only and advisory; never trims safety, security, accessibility, data-loss handling, or their tests. Use before pushing a PR, when a change feels heavier than the task warranted, or when boilerplate/a new dependency crept in. Triggers: 'ponytail review', 'is this over-engineered', 'what can I delete here', 'did I over-build this'.From its SKILL.md

Install
npx -y skills add wildcard/caro --skill ponytail-review

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

What its file declares

Copied from the file, not written here

The file declares its own license as AGPL-3.0. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.

SKILL.md

3.9 KB, as published. Nobody here has run it

Ponytail Review Skill

What This Skill Does

Spawns the ponytail-reviewer agent against the current change set and returns its 6-section critique. The reviewer is the "lazy senior dev" — it looks for the simplest thing that could work and names what is surplus, without ever touching safety, security, accessibility, data-loss handling, the tests that guard them, or the release/ADR process.

It is the active companion to the passive good-boy-scout rule, and the deliberate opposite of devils-advocate: devils-advocate asks "should this be built?"; ponytail asks "is this bigger than it needs to be?"

This skill is advisory. It proposes reductions; it never applies them. The human or parent agent decides what to act on.

When to Use This Skill

  • Before pushing a PR, as a bloat check.
  • When a change feels heavier than the task warranted.
  • When boilerplate, a new abstraction, or a new dependency crept in.
  • When reviewing whether a process change is ceremony or signal.

Do not use it as a gate that blocks merges, and do not let it override safety-pattern TDD or any constitution carve-out.

How to Run

  1. Determine the change set under review (in order of preference):

    • An explicit diff/PR the user named.
    • git diff (unstaged) and git diff --staged on the working branch.
    • git diff main...HEAD for the whole branch when reviewing a PR.
  2. Spawn the reviewer with the Task tool:

    • subagent_type: "ponytail-reviewer"
    • Prompt: paste the diff (or name the files) plus the stated intent of the change, and ask for the standard 6-section critique.
  3. Return the agent's critique verbatim. Do not soften the verdict.

What You Get Back

The standard 6-section contract:

  1. What I read — the change and its intent.
  2. The simplest thing that could work — the YAGNI baseline.
  3. Reduction objections (numbered) — what could go, which ladder rung it fails, why it's surplus, what would justify keeping it.
  4. Carve-outs respected — the safety/security/a11y/data-loss/test/ release lines it deliberately did not trim.
  5. Net effect — lines/files/deps/steps removable vs. the cost.
  6. Verdict — Keep as-is / Trim / Over-built.

Guardrails

  • Read-only. The skill and the agent never edit, delete, format, or commit. They report.
  • Carve-outs are absolute. If the agent ever suggests trimming a safety pattern, a security check, an accessibility affordance, a data-loss safeguard, a test for new functionality, or a release- checklist step — discard that objection; it violated the contract.
  • Pragmatic, not zealous. A few extra lines that aid clarity are not a finding. Shorter is not always simpler.

See Also

  • .claude/agents/ponytail-reviewer.md — the agent this skill drives.
  • .claude/agents/devils-advocate.md — the opposite critical axis.
  • .claude/rules/good-boy-scout.md — the passive principle this operationalizes.
  • docs/adr/ADR-016-ponytail-pragmatic-reviewer.md — why this exists and what we deliberately did not adopt from the upstream ponytail project.

Keep looking

Skills are one crate of 627,801. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.