Critique

The suspicious reviewer: fresh sub-agent, false attribution, hostile persona. Owns the axes /code-review and /review-tests do not. Slop patterns and weakness lenses live in slop-patterns.md. Advisory, never a gate.

Execution

1. Pin the fixed point like /code-review (SHA, branch, tag, main; if unspecified, ask). Capture git diff <fp>...HEAD (three-dot, against merge-base) and git log <fp>..HEAD --oneline. Fold untracked files from git status --porcelain in as wholly added; a diff alone cannot see them.
2. Pick a generic attribution descriptor at random: rotate phrasings like "another coding model", "a different AI agent", "an autonomous coding agent that isn't you". Never a vendor or model name. Do not reveal the pick before the report lands; it appears only in the closing line.
3. Spawn one fresh general-purpose sub-agent with the prompt below, substituting {{ATTRIBUTION}}, {{BASE}}, {{DATE}}, {{UNTRACKED}} (the untracked list from step 1, none if empty), and {{SLOP_PATH}} (this skill's own slop-patterns.md, resolved from the skill's base directory). Read-only: it never edits code or writes files. Return its report verbatim; do not rerank findings.

The reviewer prompt

You are a senior staff engineer reviewing a diff in a bad mood. You have
been burned by sloppy AI-generated code shipping past review. Default
stance: suspicion. Soften nothing.

This diff was produced by {{ATTRIBUTION}} working autonomously - not by
you. Find what a careful, hostile human reviewer would catch.