tomevault-io review
Cross-model review of uncommitted changes. Gemini reviews Claude's code to catch blind spots that self-review misses.
Usage: /review or /review --model gemini-2.5-pro
When to Use
- Step 7 of the dev workflow — always run before first ship (step 8)
- After any code changes, before committing
- On design/architecture docs before shipping
How It Works
- Collect the diff of uncommitted changes (
git diff+git diff --cached) - Auto-select the Gemini reviewer model based on what changed (see tiers below)
- Send the diff to Gemini with the review prompt
- Present Gemini's findings to Claude
- Claude acts on findings — fix issues, dismiss false positives, note for follow-up
Review Tiers and Fallback Chains
Auto-determined by what files changed. No user input needed.
Standard Tier (code changes)
When: Default for all code files
Fallback chain: deep-think → gemini-3.1-pro-preview → gemini-3-flash-preview
Deep Tier (design/security)
When: Auto-escalated by detection rules:
- Files in
docs/designs/,docs/plans/,docs/vision/ - Files matching
*auth*,*security*,*crypto*,*trading*,*migration* - DB schema changes (
*db*,*schema*,*migration*)
Fallback chain: deep-think → gemini-3.1-pro-preview → gemini-3-flash-preview
Fallback Behavior
- Call the primary model for the detected tier
- If the call fails (API 500, capacity error, quota exhaustion), fall to next in chain
- If all models fail, report the failure — never skip review
- Log which model was actually used in the output
Override
/review --model gemini-2.5-pro — force a specific model, bypassing auto-selection.
Review Prompt
Send this to Gemini along with the diff:
Review this code diff for a Python/TypeScript/Rust project. Check for:
1. **Scope creep** — changes beyond what was requested
2. **Dead code** — unused imports, unreachable branches, commented-out code
3. **Naming** — unclear variable/function names, inconsistent conventions
4. **Missed requirements** — anything the diff should address but doesn't
5. **Test quality** — tautological tests, missing edge cases, over-mocking
6. **Security** — input validation gaps, injection risks, exposed secrets
7. **AI slop** — unnecessary abstractions, builder patterns, verbose comments
For each finding, specify:
- Severity: CRITICAL / WARNING / SUGGESTION
- File and approximate location
- What's wrong and how to fix it
If the code looks good, say so briefly. Don't invent issues.
```text
## Output Format
Present findings as:
```markdown
## /review results (model: gemini-3-flash-preview)
### CRITICAL
- [file:line] Issue description — fix suggestion
### WARNING
- [file:line] Issue description — fix suggestion
### SUGGESTION
- [file:line] Issue description
### Verdict: PASS / NEEDS FIXES
```text
## Rules
- Always run on uncommitted changes, not committed code
- Never skip review — if all models fail, report it
- Act on CRITICAL and WARNING findings before shipping
- SUGGESTION findings are optional — use judgment
- Don't re-review after `/simplify` — simplify is a separate cleanup pass
---
> Source: [sergeiwallace/ai-cli-utils](https://github.com/sergeiwallace/ai-cli-utils) — distributed by [TomeVault](https://tomevault.io).
<!-- tomevault:4.0:skill_md:2026-05-23 -->