← ClaudeAtlas

odoo-reviewlisted

Reviewing Odoo code before it merges — your own AI-generated patch or someone else's PR. Use after writing or generating an Odoo change and before commit, or when asked to review/audit an Odoo module, diff, or addon. Catches the Odoo-specific defects AI confidently ships that lint and "it ran for me as admin" miss: blanket sudo() / privilege bypass, missing ACL or record rules, N+1 recordset loops, incomplete @api.depends on stored computes, patching core instead of a separate addon, wrong inheritance mode / MRO layer, deprecated v≤16 syntax (attrs/states/<tree>/name_get), guessed field & method names, unwired __init__ / manifest, data-loss-on-upgrade renames, and public routes with the wrong auth. Pairs with odoo-testing (review finds it, tests prove it). Read ground truth from the running instance to confirm a suspicion — don't guess whether a field or rule exists. Targets Odoo 17/18/19.
tuanle96/odoo-ai-skills · ★ 4 · Code & Development · score 62
Install: claude install-skill tuanle96/odoo-ai-skills
# Odoo code review — the second gate `odoo-testing` proves a change does what it claims; **this skill catches what it claims that it shouldn't, and what it silently broke elsewhere.** AI-generated Odoo code is the dangerous case: it's syntactically clean, it "works for admin on my machine", and it fails later — for a non-admin user, on the second company, on a list of records, on `-u`, or on a hot list view. Review against the Odoo-specific contracts a linter can't see. **The rule: every finding is either confirmed against the running instance (via `odoo-introspect`) or written down as an explicit assumption to verify — review never adds new guesses.** **Version floor: Odoo 17/18, through Odoo 19 (current LTS).** Deprecation checks below assume 17/18+; for older targets — and for the v18.1 → 19 renames — see `skills/odoo-introspect/references/version-matrix.md`. ## How to review (in order) 1. **Get the diff and the ground truth.** Read the change; run `odoo-ai all <model>` for every model it touches (fields, MRO, security, depends). Half of "review" is just comparing the patch to what the brief says already exists. 2. **Run the checklist below**, severity-first (security → data loss → silent correctness → performance → style). 3. **Confirm, don't assert.** "This field may not exist" → check the brief. "This override may not run" → read the MRO / trace it. Turn each suspicion into a fact or a flagged assumption. 4. **Hand correctness findings to `odoo-testi