kimi-review

name	kimi-review
description	Run a Kimi K2.6 code correctness review of the current diff. Catches logical bugs, edge cases, and API contract violations. Use before every PR alongside /kimi-security and /kimi-optimize.
disable-model-invocation	true
allowed-tools	Bash Read

Current diff

!git diff HEAD

Test file context

!find tests -name '*.py' | sort

Instructions

1. Build the task brief

Read prompt-template.md. Fill in:

• {{DIFF}} — the full diff injected above
• {{TEST_FILES}} — the test file list injected above

Write the filled brief to /tmp/kimi_review_brief.md.

2. Run Kimi

Run this command:

kimi --quiet -p "$(cat /tmp/kimi_review_brief.md)"

3. Filter findings

Discard any finding that:

• Flags FPE producing same-format output as a bug (correct behavior)
• Flags asterisk masking as needing a recovery path (must not exist)
• Contradicts behavior already validated by a passing test in tests/
• Is a style or performance suggestion (those belong in /kimi-optimize)

4. Report

For each confirmed finding:

• Severity: HIGH / MEDIUM / LOW
• Location: file:line
• Bug: what is wrong and what input triggers it
• Fix: specific change needed

Do not proceed to PR creation if any HIGH findings remain unresolved.