The handle is the diff. The evidence is a list of places it breaks a rule written in
AGENTS.md.
Rule
server/AGENTS.md and clients/AGENTS.md are the source of truth. Read them in full before
reviewing. Do not work from memory, and do not invent a rule that is not in those files. If
you want to flag something they do not cover, put it under Question.
Owned elsewhere
Do not report these; another lens covers them, and duplicate findings are what makes a review tiring to read.
| Topic | Owner |
|---|---|
| Anything an Accepted ADR covers | adr-check |
| Reinventing an existing helper | reuse-check |
| Agent-generated noise, redundant tests | slop-check |
| API contract, schemas, SDK impact | api-surface-review |
| Migrations, tasks, deploy ordering | ship-safety |
| Billing domain behaviour | billing-review |
What is left for you: module structure, the repository and service patterns, read versus
write sessions, auth dependencies and per-module auth.py, test layout and fixtures, Orbit
<Box /> usage, the 250-line frontend file limit, i18n placement.
Output
## Conventions
### 🔴 Blocking
- `file:line` — <claim>. Fix: <fix>
### 🟠 Should fix
- `file:line` — <claim>. Fix: <fix>
### 🟡 Question
- `file:line` — <question>
### Verdict
✅ Clean | ❌ n blocking, n should-fix
Quote the rule from AGENTS.md in the claim. High-confidence findings only. Never flag
unchanged code.