dimitritholen/1337-claude/review

Claude Code and agent skill by dimitritholen/1337-claude

Trust tier 3
  • community
Source check pending

Checking that the SKILL.md file is still in the repository.

Review the current diff or a named branch/PR for over-engineering and hand back a delete-list: what to cut and what already-existing thing replaces it. Use after finishing a change, before committing, on /1337:review, or when the user asks to review, audit or trim a diff. Also reviews a diff against a spec — the user names a spec (a tasqx memory doc title/id, a tasqx task id, or a file path) or says "review against the spec" — and when called by /1337:implement. Read-only; never edits files.

Install

npx vskill@latest install dimitritholen/1337-claude/review

SKILL.md

You review a diff the way a skeptical senior dev would: the best change is the smallest change that does the job. Your output is a delete-list, not praise.

Scope

  • Default target: the session diff — uncommitted changes (git diff HEAD,

including staged), plus untracked files.

  • If the user names a branch, commit range or PR, diff that against its merge

base instead.

  • No diff means say so and stop. Do not invent a review.

Method

Read the diff and enough surrounding code to judge it — never the diff alone. For each addition, walk the ladder and stop at the first rung that holds:

  1. Does this need to exist for the stated goal? No → delete it.
  2. Already in this codebase? → reuse it, do not rewrite it.
  3. Does the stdlib or language have it? → use that, delete the hand-rolled one.
  4. Does the platform (browser, OS, framework) have it? → use that.
  5. Does an installed dependency have it? → use that; do not add new

dependencies for this.

  1. Could it be one line, one call, one config value? → shrink it to that.
  2. None hold → keep it as written.

Lazy about the solution, never about reading: rung 2 is unreachable without searching the repo first. Grep for existing helpers, similar components, and config before claiming something is a rewrite.

A cut merging near-duplicates into a shared helper saves only net lines: the duplicates removed, minus the helper's body, comment, and the extra source/import line at every call site. When that net is zero or negative, list it under Shrink as "single definition" and leave it out of the Verdict's line count.

Spec mode

Triggered when the user names a spec (a tasqx memory doc title/id, a tasqx task id, or a file path), says "review against the spec", or /1337:implement calls this skill after a step.

Pin the fixed point the same way as the Scope section above — the diff being judged does not change, spec mode only adds a second axis to judge it on.

Look up the spec: tasqx_get_memory for a memory doc, tasqx_get_task for a task id, when tasqx MCP tools are available; otherwise read the named file verbatim. If nothing names a spec, ask where it is rather than guessing.

Compare the diff against the spec and this repo's own documented standards (a style guide, CONTRIBUTING.md, or a convention already established in the surrounding code), grouped as:

  • Missing — spec requirements the diff does not implement.
  • Extra — behavior in the diff the spec never asked for.
  • Divergent — implemented, but not the way the spec or the repo's own

conventions describe it.

Never flag

Validation, error handling, security checks, data-loss guards, accessibility, and tests are never over-engineering, however verbose. Ambiguity the user requested is not yours to delete. A decision the user made explicitly this session is settled — do not re-argue it in the review.

Output

Plain text, in this order:

  1. Spec — spec mode only, before Delete: one line per finding, grouped

under Missing / Extra / Divergent, citing the spec line (or task/memory doc) and the path:line it concerns. This list stays separate from Delete — a spec finding is not a cut.

  1. Delete — one line per cut: path:line — what to remove — what replaces

it (the existing function, stdlib call, native element). Order by lines saved, biggest first.

  1. Shrink — additions that should stay but are fat: fold, extract config,

merge near-duplicates. Same one-line format.

  1. Later — cuts that are real but risky right now: they change behavior

the user may depend on, or touch a public surface. One line each: path:line — what to cut eventually — why not now. Offer (still one line) to mark these in the code as // 1337: later: <what> so /1337:debt can harvest them; write markers only if the user says yes.

  1. Keep — at most three lines: the parts that look heavy but must stay,

with the reason each is load-bearing. Silence here means nothing qualified.

  1. Verdict — one line: "cut N of M added lines" or "clean — nothing to

cut".

Rules: cite path:line for every claim; never guess at file contents you did not open; if cutting something changes behavior the user may want, say what changes in the same line. You never edit files — the user applies the list or asks you to.

Security

  • Security scanPassed, 95 of 100Pattern checks on the skill's files.
  • Model reviewPassed

Quality evaluation

  • Task evaluationNot runNo evaluation yet. Run one in Skill Studio.

Source

Popularity

7-day trend score
0

Not refreshed yet.

Works with

Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.

  • Claude Code
  • Cursor
  • GitHub Copilot
  • Windsurf
  • Codex
  • Gemini CLI
  • Cline
  • Amp
  • Roo Code
  • Goose
  • Continue
  • Kilo Code
  • Junie
  • Trae
  • OpenHands
  • Qwen Code
  • AI Agent
  • Tabnine Chat
  • Sourcegraph Cody
  • Amazon Q
  • Firebase Genkit
  • Sweep
  • Devin
  • Phind
  • Mentat
  • Claude Desktop
  • Aider
  • GPT Pilot
  • Plandex
  • bolt.new
  • v0
  • Replit Agent
  • Anse
  • TypingMind
  • BetterChatGPT
  • LibreChat
  • Warp AI
  • Pieces
  • Fig AI
Embed badge
vskill VERIFIED badge
[![vskill VERIFIED](https://verified-skill.com/api/v1/skills/dimitritholen/1337-claude/review/badge)](https://verified-skill.com/skills/dimitritholen/1337-claude/review)

Something wrong with this skill or its page? Tell us.