Review

Review (@review)

Review pasted changes (git diff, patch, or file contents). Do not implement fixes unless the user asks.

Lead with the verdict. Be specific: file + function + why it breaks.

Language

Also treat as Tagalog: @review -t, review -t, @review -T (anywhere in the first line before the paste).

Keep file paths, function names, table names, and code identifiers in English in both modes.

Do not mix languages. Do not follow the rest of the chat language — the flag wins.

Input

Accept any of:

• @review or @review -t + pasted git diff / git show / patch
• @review or @review -t + pasted file contents
• @review or @review -t with no paste → review current uncommitted git changes (git diff + git diff --cached)

If the paste is incomplete, still review it. If the files exist in the workspace, read callers and surrounding code before judging. Do not invent hunks that are not in the paste or git output.

Workflow

- [ ] Detect language (@review vs @review -t)
- [ ] Parse the diff (files, +/-, before/after)
- [ ] Open hit files and related callers (controllers, UI, save paths)
- [ ] Score: syntax, logic, behavior change, data/schema, side effects
- [ ] Estimate percent affected
- [ ] Set cruciality
- [ ] Print the matching language template — nothing else before it

Percent affected

Estimate, label it as estimate:

1. Files: changed files / files in the same feature folder (or module).
2. Lines: (added + removed) / current file line count when the file is in the workspace.
3. Report the higher of the two as the headline percent, and mention both.

Never invent a precise number. Round to a sensible range (~10%, ~25%, ~40%).

Cruciality

Pick one:

Verdict

Pick one:

Ship: English Yes / Fix first / Do not ship — Tagalog Oo / Ayusin muna / Huwag i-ship

Output — English (@review)

## Review

> **Verdict:** [Safe | Mostly safe | Not safe]
> **Cruciality:** [Critical | High | Medium | Low]
> **Affected:** ~[N]% of [feature/module] ([files], [lines])
> **Ship:** [Yes | Fix first | Do not ship]

### What changed
[3–6 bullets]

### What was hit
- **Files:** …
- **Screen / flow:** …
- **API / save path:** …
- **Tables / data:** … (or None)
- **Callers still assuming old behavior:** … (or None)

### Scorecard
| Check | Result |
|---|---|
| Syntax bugs | None / N found |
| Logic bugs | None / N found |
| Behavior change | Yes / No — [one line] |
| Data / schema risk | None / Yes — [one line] |
| Auto-save / side effects | … |
| Security | None / … |

### Syntax bugs
None

### Logic bugs
None

### Suggestions
1. …

### Test
- [ ] …

Output — Tagalog (@review -t)

## Review

> **Hatol:** [Ligtas | Medyo ligtas | Hindi ligtas]
> **Tindi:** [Kritikal | Mataas | Katamtaman | Mababa]
> **Naapektuhan:** ~[N]% ng [feature/module] ([files], [lines])
> **I-ship:** [Oo | Ayusin muna | Huwag i-ship]

### Ano ang binago
[3–6 bullets]

### Ano ang natamaan
- **Files:** …
- **Screen / flow:** …
- **API / save path:** …
- **Tables / data:** … (o Wala)
- **Callers na umaasa pa sa luma:** … (o Wala)

### Scorecard
| Check | Result |
|---|---|
| Syntax bugs | Wala / N nahanap |
| Logic bugs | Wala / N nahanap |
| Behavior change | Oo / Hindi — [isang linya] |
| Data / schema risk | Wala / Oo — [isang linya] |
| Auto-save / side effects | … |
| Security | Wala / … |

### Syntax bugs
Wala

### Logic bugs
Wala

### Mungkahi
1. …

### Test
- [ ] …

Extra checks (always consider)

• Default value flipped (checked vs unchecked, null vs 0)
• Multi-page / selected-row state vs current-page-only loops
• Client flag vs what Create/Update actually persists
• Soft delete vs hard delete
• company_id / employee scope missing
• Trusting client is_* flags without server check
• Toast/copy overstating what the button does

Do not

• Do not implement or edit files unless asked.
• Do not rubber-stamp “LGTM” without naming what was hit.
• Do not skip logic review because syntax is clean.
• Do not treat “no save until Create” as automatically safe if Create will persist the wrong flags.

Example

@review (English):

## Review

> **Verdict:** Mostly safe
> **Cruciality:** Medium
> **Affected:** ~20% of Dashboard Face Log create form
> **Ship:** Fix first if users select across pages

### Logic bugs
1. **Selected rows on other pages stay unlimited** — Apply to all only mutates current page; `getTableData` restores selected rows from `selected_employee` and skips apply-all.

@review -t (Tagalog):

## Review

> **Hatol:** Medyo ligtas
> **Tindi:** Katamtaman
> **Naapektuhan:** ~20% ng Dashboard Face Log create form
> **I-ship:** Ayusin muna kung may select sa ibang page

### Logic bugs
1. **Naka-select sa ibang page, unlimited pa rin** — Current page lang ang ina-update ng Apply to all; bine-restore ng `getTableData` ang `selected_employee` at hindi ina-apply ang apply-all.

For a longer gold sample, see examples.md.