# Crit

> Crit is an Agent Skill for testing & quality published in the tomasz-tomczyk/crit repository on GitHub: a SKILL.md instruction file an AI agent can load to do this task.

Review code changes, a plan, a live page (running dev server), or a local HTML file with crit inline comments. Use when asked to review code, a plan, a diff, a running web app, or when you want structured human feedback on your work.

- Skill name: `crit`
- Category: [Testing & Quality](https://www.openparable.com/skill/testing-quality/1)
- Source: https://github.com/tomasz-tomczyk/crit/blob/HEAD/integrations/gemini/skills/crit/SKILL.md
- Repository: tomasz-tomczyk/crit (765 GitHub stars when collected)

## SKILL.md

# Review with Crit

Review and revise code changes, plans, live pages (running dev servers, staging URLs), or local HTML files using `crit` for inline comment review.

## Step 1: Pass arguments to `crit`

The CLI auto-detects the review mode from its arguments. **Do not ask the user which mode to use.** Pass arguments through:

```
crit <arguments>               # file, dir, URL, .html — CLI auto-detects mode
crit --pr <num|url>            # GitHub PR (range mode)
crit --range <base>..<head>    # commit range (range mode)
crit                           # no args → branch diff
```
If no arguments, check conversation context:

1. A plan file was written earlier in this conversation → `crit <plan-file>`
2. Otherwise → bare `crit` (branch diff)

## Step 2: Launch crit and block until review completes

**CRITICAL — you MUST run this step. Do NOT skip it. Do NOT proceed without it.**

Tell the user upfront: **"Opening crit for review — leave inline comments in the browser and click Finish Review when done."**

Then run `crit` via `run_shell_command` — it blocks until the user clicks "Finish Review", then the tool call completes:

```bash
crit <plan-file>   # specific file
crit               # git mode
```

If a crit server is already running from earlier in this conversation, `crit` automatically connects to it. Starting from scratch, it spawns the daemon, opens the browser, and blocks until the user clicks "Finish Review".

**Do NOT proceed until `run_shell_command` returns.** Do NOT ask the user to type anything. Do NOT read the review file early. The tool call completing is how you know the human is done reviewing.

## Step 3: Read the review output

When `crit` completes, read **stdout** and follow its instructions. Check **stderr** for `approved: true` or `approved: false`.

<important if="a comment has a quote, anchor, or drifted field">
- `quote`: the specific text the reviewer selected — focus your changes on the quoted text rather than the entire line range
- `anchor`: use it to locate the current position of the content; line numbers may be stale after edits
- `drifted: true`: original content was removed or heavily rewritten — line numbers are approximate at best
</important>

**Fallback** (mid-round re-entry, plan hooks, or headless workflows): `crit comments` / `crit comments --json`. Use `crit comments --plan <slug>` for plan-mode reviews.

## Step 4: Address each review comment

For each unresolved comment:

1. Understand what the comment asks for
2. If it contains a suggestion block, apply that specific change
3. Revise the referenced file (plan or code file from the diff) using `replace`
4. Reply with what you did: `crit comment --reply-to <id> --author 'Gemini' '<what you did>'` (reply bodies support markdown)
5. **Do not pass `--resolve`.** Resolving is the reviewer's call. Only add `--resolve` if the user explicitly asks.

Editing the plan file triggers Crit's live reload — the user sees changes in the browser immediately.

<important if="you are replying to multiple comments at once">
Use `--json` for a single bulk call instead of one invocation per comment:

```bash
echo '[
  {"reply_to": "c_a1b2c3", "body": "Fixed"},
  {"reply_to": "c_d4e5f6", "body": "Refactored as suggested"}
]' | crit comment --json --author 'Gemini'
```
</important>

## Step 5: Signal completion and start next round

**CRITICAL — you MUST run this step. Do NOT skip it. Do NOT proceed without it.**

The finish prompt on stdout includes the command to run again — use it to start a new round.

On subsequent calls, `crit` automatically signals round-complete first, then blocks until the next "Finish Review" click.

Tell the user: **"Changes applied. Review the diff in your browser and click Finish Review when ready."**

**Do NOT proceed until `run_shell_command` returns.** When it does, return to Step 3. If the user finishes with zero comments, the review is approved — stop the loop and proceed.

<important if="the user asks for a URL, a shareable link, or a QR code for the review">

```bash
crit share <file>
```

**Always relay the full output to the user** — copy the URL (and QR code if `--qr` was used) directly into your response. Don't make them dig through tool output.

To remove a shared review:

```bash
crit unpublish [file...]
```
</important>

<important if="you are about to add --qr to a share command">
Only use `--qr` in real terminal environments with monospace rendering. Skip it in mobile apps or web chat UIs — Unicode block characters won't render.

```bash
crit share --qr <file>
```
</important>

---

Source: https://www.openparable.com/skill/testing-quality/crit (OpenParable). Cite the canonical URL above.
