# Dw Check

> A second opinion on the diff so far, mid-build: what is wrong with it, and does it fit this repo? The pass goes to an outside reviewer — codex — or to a quick self-review when the diff is trivial or none is installed, and every mistake it names is verified at a real file:line before it counts, then fixed in-session on your approval.

- Skill: `dominikwozniak/dw-check` (Agent Skill)
- Install (CLI): `npx skillmds@latest add dominikwozniak/dw-check`
- Raw SKILL.md: https://api.skillmd.com/api/skills/dominikwozniak/dw-check/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: dominikwozniak (https://skillmd.com/u/dominikwozniak)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/dominikwozniak/dw-check

---


# dw-check — a fast look, then fixes

**The review step, repeatable mid-build.** Run it while the diff is small.

## What it reads

The diff against the default branch — the ref `bash "${CLAUDE_PLUGIN_ROOT}/scripts/base-ref.sh"` prints — plus
the branch's `CHANGE.md` goal, so findings are judged against what the change is trying to do, not
against taste. The argument is read two ways: the single word `codex` forces the delegated pass
past the triviality floor, and anything else narrows the focus to a path or a topic. It writes no
`.ai/` artifact — approved fixes land as code commits.

## Workflow

### 1. Establish the diff

`git diff <base>...HEAD` plus `git log --oneline <base>..HEAD`, narrowed by the focus when given.

### 2. Delegate by default, above a floor

Bare hands the diff to `codex:rescue` whenever the codex plugin is installed, without asking — the
second model is what this gate adds. Two things send it to step 3 unassisted, each costing one
line saying so, never a silent skip:

- **A trivial diff** — 2 files or fewer **and** under 50 lines from `git diff --shortstat`.
  `codex` overrides this floor, and only this floor.
- **No reviewer installed** — self-review, naming the fix in the same line:
  `/plugin marketplace add openai/codex-plugin-cc`, then `/plugin install codex@openai-codex` —
  or `/codex:setup` when it's installed but not ready.

`/codex:review --wait` is the richer, user-typed pass — name it when the diff deserves it. Either
way, quote delegated findings verbatim, then **verify each against the file before it counts** —
line numbers are the first thing to check.

### 3. Two axes, judged separately — never merged into one score

- **Correct?** — does the diff do what the goal says: edge cases, error paths, the empty input.
- **Does it fit?** — compared against the neighbouring code, not best practice in the abstract.

Every finding points at a real `file:line` you opened — if you can't ground it, don't report it.
"No findings on either axis" is a normal, useful answer; say it plainly and stop. Then filter, and
name what you dismissed, one line each:

- All nits means the diff is probably fine — lead with that conclusion, nits after it.
- "I would have done it differently" is not a finding until it names a concrete problem with this.
- More than five things to act on means the filter is too loose — except correctness and security,
  which earn more scrutiny before dismissal, not less.

### 4. Present, wait, then fix

List the findings with a severity-ordered recommendation and **stop — nothing is fixed without
approval.** On approval, fix in-session and commit per `## Git conventions` — related fixes together,
unrelated apart. Then offer to run again; the second look is the point of a gate this cheap.

**Next:** `dw-next` for the next task, `dw-land` when nothing is pending, or `dw-grain` where
`dw-solo-extras` is installed.

$ARGUMENTS