Checks

A check is a code-review guideline: one topic a review tool (Cursor Bugbot, Kilo Code Reviews, GitLab Duo, Augment, Amp, …) applies to every pull or merge request. You write each check once in .ai-rulez/checks/, and generate writes it into the review file each enabled tool reads.

Example

ai-rulez add check security --severity high --description "Flags common security issues"

.ai-rulez/checks/security.md

---
type: Reference
title: Security
description: Flags common security issues
x-ai-rulez:
  kind: check
  id: security
  metadata:
    severity: high
---

Review the diff for injection vulnerabilities, hardcoded secrets and unsafe
deserialization. Report each finding with a file and line reference.

.ai-rulez/config.toml

presets = ["cursor", "kilo", "amp", "augment"]

After ai-rulez generate, Cursor Bugbot and Kilo get a section in a shared markdown file. If REVIEW.md already holds your own text, the block is appended below it:

REVIEW.md

This repository has no review notes yet.

<!-- ai-rulez:checks:begin -->
<!-- ai-rulez:check:security -->

## security

Review the diff for injection vulnerabilities, hardcoded secrets and unsafe
deserialization. Report each finding with a file and line reference.
<!-- ai-rulez:checks:end -->

Amp gets one file per check, with its metadata in the frontmatter (tools too, when the check sets them):

.agents/checks/security.md

---
description: Flags common security issues
name: security
severity-default: high
---

Review the diff for injection vulnerabilities, hardcoded secrets and unsafe
deserialization. Report each finding with a file and line reference.

Augment gets an area in its YAML file:

.augment/code_review_guidelines.yaml

areas:
  security:
    description: Flags common security issues
    globs:
      - '**'
    rules:
      - description: |-
          Review the diff for injection vulnerabilities, hardcoded secrets and unsafe
          deserialization. Report each finding with a file and line reference.
        id: security
        severity: high

Source files

.ai-rulez/checks/<name>.md
.ai-rulez/domains/<domain>/checks/<name>.md
.ai-rulez/local/checks/<name>.md        machine-local, gitignored

The file name is the check’s identity. It may contain only letters, digits, ., _ and -, since it becomes part of output paths and section markers; validate rejects anything else. As for rules, type is required in a tree created by init; create checks with ai-rulez add check, or run ai-rulez generate after writing one by hand so the index.md listings include it. Two names that differ only in case count as one check, and the higher-precedence source wins.

Root checks are always included; domain checks are included when the domain is in the active profile or role. A root check shadows a domain check of the same name.

Key	Type	Default	Meaning
description	string	none	Short summary. Used as the text when the body is empty.
severity	string	none	low, medium, high or critical. Any other value fails validation.
tools	list	none	Tools the check may use. Only Amp has a place for them.
targets	list	every preset	Presets, paths or globs the check is written to, as for rules. add check and update_check reject a misspelt preset name.

Managing checks

ai-rulez add check security --description "Security issues" --severity high
ai-rulez add check sql-injection --domain backend --targets "src/**/*.go"
ai-rulez add check perf --tools Read,Grep --severity medium
ai-rulez list checks [--domain backend] [--format json]
ai-rulez remove check security [--domain backend] [--yes]

The MCP server has create_check, read_check, update_check, delete_check and list_checks. update_check takes new content, fields (description, severity, tools, targets), or both. Content without frontmatter replaces only the body and keeps the existing frontmatter.

An include can import checks with include = ["checks"].

Where they are written

Preset	Output	Shape
cursor	.cursor/BUGBOT.md	one block, one section per check (Bugbot)
kilo	REVIEW.md	one block, one section per check
qwen	.qwen/review-rules.md	one block, one section per check
rovodev	.rovodev/.review-agent.md	one block, one section per check
factory	.factory/skills/review-guidelines/SKILL.md	one block in a skill with name and description frontmatter
amp	.agents/checks/<name>.md	one file per check
augment	.augment/code_review_guidelines.yaml	one area per check, merged into your YAML
gitlab-duo	.gitlab/duo/mr-review-instructions.yaml	one instructions group per check, merged into your YAML

Shared markdown files

ai-rulez owns only the block between <!-- ai-rulez:checks:begin --> and <!-- ai-rulez:checks:end -->. Each check in it is a <!-- ai-rulez:check:<name> --> marker, a ## <name> heading and the check text:

<!-- ai-rulez:checks:begin -->
<!-- ai-rulez:check:security -->

## security

Review the diff for injection vulnerabilities...
<!-- ai-rulez:checks:end -->

• Text outside the block is yours and survives every generate. An existing file gets the block appended.
• clean, or removing the last check, removes only the block. A file that held nothing else is deleted; a file with your text is never deleted or gitignored.
• Do not edit inside the block (the next generate rewrites it) and do not put the marker lines in a check’s text. generate fails rather than touch a file whose markers do not pair up.
• Severity and tools have no equivalent in these files and are not written.
• For factory, the skill frontmatter is written once, when the file is created. A skill of your own named review-guidelines collides with it, and generate fails until one is renamed.

Augment and GitLab Duo YAML

Both files are documented as hand-written, so ai-rulez merges into them. It owns only the areas (Augment) or groups (GitLab Duo) named after its checks, keeps everything else (file_paths_to_ignore, your own entries, comments and key order), and removes its entries when a check is deleted.

• An entry you wrote under a check’s name is yours: it is never replaced, the check is not written there, and generate warns. Rename one of them.
• An entry ai-rulez wrote becomes yours once you edit it.
• Augment’s severity scale stops at high: critical is written as high, and an unset severity as medium. Every area gets globs: ["**"].
• GitLab Duo groups have name and instructions only, with no file filter.

Tool notes

• Commit the outputs. A hosted reviewer reads only committed files, usually from the base branch of the pull request. Check outputs are never added to the managed .gitignore block, even with gitignore = true.
• Kilo reads REVIEW.md only when Use REVIEW.md is enabled in the Kilo web app, and truncates it at 10,000 characters. generate warns when the file, your text included, is longer.
• Bugbot also reads nested <dir>/.cursor/BUGBOT.md files, which ai-rulez does not generate. generate --user writes no Bugbot file.
• Amp: the format follows the rulesync documentation, since Amp’s own manual does not describe it. Verify it against your Amp version.
• Local checks (.ai-rulez/local/checks/) are written only for presets with one file per check (Amp), as a gitignored file. Shared files never take a local check, and a warning names the presets that skipped it. A local check may not share a name with a shared one.
• ai-rulez doctor and ai-rulez tokens include the check files.

Not generated: Augment per-area globs, GitLab fileFilters, Takt quality gates, Hermes pre-verify specs, and JetBrains AI Assistant (its self-review path is a per-user IDE setting).