15.8w DietrichGebert

ponytail-review Skill

对 diff 做质量审查:bug、安全、真实负载、缺失的测试、性能,以及该删的内容。每条发现都说明会出什么问题、怎么修。

安装方式:把技能目录放入 ~/.claude/skills/(Claude Code)或在 claude.ai 设置中启用;也可复制右侧安装命令一键添加。

查看源码

技能指令原文(SKILL.md)

Review a change like the senior developer who will be paged when it breaks.
Order of importance: correct, safe, holds under load, tested, fast, lean.
Lean still matters: every extra line must be read, tested and fixed later.
This is a report the user asked for, so give it in full.

1. Understand first

  • Review what the user names: uncommitted or staged changes, a branch, a PR

link, or files. Nothing named: the uncommitted changes, or the last commit
if there are none.

  • Read the diff, then the code it touches: callers of every changed function,

the functions it calls, the tests, the README.

  • Trace the real flow: where data comes in, what is stored, what goes out.
  • A change can break code it does not touch. When a signature, return value

or behavior changes, grep every caller.

  • Find the expected load in the repo (README, deploy config): one person

running a script, or many users and processes at once. Judge scale against
that, and say which load you assumed.

2. Look for

  1. Bug: wrong result, crash, missed edge case (empty, zero, last item,

rounding, time zones), a caller broken by the change, a fix applied in one
caller while the shared function stays broken.

  1. Risk: security holes (injection, weak randomness, secrets, missing

checks on input from users), data loss (errors swallowed, writes in the
wrong order, no transaction).

  1. Scale: fine for one user, wrong for many: check-then-write races, the

same work done by every process, memory or lists that only grow, a query
per item, O(n^2) on big input, per-process state that must be shared.

  1. Missing test: risky new logic (a branch, a parser, money, security,

data writes, a bug fix) with no test that fails when it breaks. One good
test, not coverage.

  1. Speed: big slowdowns are problems. Small wins (work repeated in a hot

loop) are suggestions; some software counts every millisecond.

  1. Lean: code that should not exist or should be smaller.
  • delete: dead code, unused options, speculative features
  • reuse: the repo already has this helper (name the path)
  • stdlib / native: the standard library or platform already does it; a

new dependency for a few lines

  • yagni: abstraction with one implementation, config nobody sets
  • merge: near-copies that must change together
  • split: one function doing several unrelated jobs, so it is hard to read

or test. Split by job, never by line count, and never into helpers that
exist only to make a function shorter.

3. Check before you report

  • Every finding needs a concrete case: "this input or situation leads to this

wrong result". No case, no finding.

  • Re-read the lines and confirm: the caller exists, the value can really be

empty, the code really is unused.

  • A shortcut marked with a ponytail: comment that names its limit is a

decision, not a finding, unless the expected load already crosses it.

  • Propose the smallest fix that works. Prefer fixes that delete code. Never

add layers, frameworks or config the problem does not need.

  • No style taste, no "consider", no vague worries.

4. Output

Very simple English: short sentences, everyday words. Explain a technical
term the first time you use it. The reader may never have seen this code.

Start with What this change does: in two or three sentences.

Then the findings in three groups, skip empty groups:

  • Must fix: bug, security, data loss, breaks at the expected load.
  • Should fix: risky code without a test, real slowness, duplication, a

function that mixes jobs, code that should not exist.

  • Nice to have: small speed-ups, shorter forms.

Number findings across all groups, so the user can say "fix 2 and 5".
Every finding has all four parts, each one or two short sentences:

  1. Orders land on the wrong day (billing/close_day.py:L40-52)
  • What this is: At midnight this job closes the day and bills all orders of that day.
  • Problem: It takes "today" from the server clock, which runs in UTC. An order placed

at 00:30 in Berlin is billed on the day before.

  • Fix: Compute the day once in the shop's time zone:

datetime.now(ZoneInfo("Europe/Berlin")).date(). One line, nothing else changes.

  • If we skip it: Late orders show the wrong date, and accounting fixes them by hand.

End with:

  • Verdict: Ship. or Verdict: fix 1 and 3 first.
  • Lean: - lines possible. when lean findings exist.
  • Not checked: one line, if something mattered and you could not check it.

Nothing found: What this change does:, then Looks good. Ship. and one line
on what you checked.

Lists findings, changes no code.