Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →ivanwng97/two-lens-review
Run pixtuoid's review protocol at either scope — the mandatory pre-merge DIFF gate (2+ differentiated-lens agents on the diff) or a whole-codebase AUDIT (subsystem × factor fan-out over the whole tree). Both draw ONE shared factor taxonomy + verify contract + disposition; they differ only in population and orchestration. Use before merging ANY PR, on 'review this PR/branch' / 'is this ready to merge' (diff scope), or on 'whole-codebase review' / pre-release / periodic audit (whole-codebase scope
Passed 17 of 20 test cases with the skill, and 9 of 20 without it.
and 9 more in the full report
9 fixed10 passed either way3 still failing0 broke
Tokens, turns and time describe resource use, not a dollar price.
This benchmark measured an earlier version of the skill.
A newer attempt on gemini-3.6-flash on 10/2/2026 did not produce a publishable comparison (not comparable). The snapshot above is the previously measured run.
One of 9 test cases the skill fixed. Same model, same prompt, with and without the skill.
During a whole-codebase audit sweep, finder agent Alpha identifies 15 potential memory leak locations and self-certifies all 15 as confirmed HIGH severity bugs without further checks. What verification step must take place before generating the final report?
| Checked | Without skill | With skill |
|---|---|---|
| Prohibits finder agents from self-certifying their own candidate findings. | Passed | Passed |
| Requires a separate adversarial skeptic pass to attempt to refute each candidate finding. | Missed | Passed |
The instructions your model reads when the skill loads.
Every rule is REVIEW.md's; fill briefs from it, never a
paraphrase here.
Isolate the branch in a worktree (two sessions on one tree race on
HEAD). Note path, branch, base sha.
Dispatch both lenses in parallel, in the background, plus one lens per matching escalation row, each briefed:
You are the <correctness|design|row> lens for <PR/branch> on pixtuoid.
Worktree: <path> (branch <name>, base <sha>), read-only.
Diff: git -C <path> diff <base>..HEAD.
Read AGENTS.md, then apply REVIEW.md's <lens or row>, its Lenses preamble
and Do not flag.
<change-specific claims (from the PR body's impl-plan answers) or design
questions, one per line>
Run the applicable gates; report each exit code as observed, never through
a pipe. Each finding carries an integer confidence 0–100. The comment audit
reports N items each with a disposition, never "passed", plus the diff's
net added comment lines and every sentence deletable with nothing lost.
Your final message is the report, ending in one verdict: APPROVE or
REQUEST-CHANGES.
The filled slots are the quality lever; a lazily filled one turns every lens generic.
Collect and verify. A one-word or placeholder return is a STUB, not a review: re-run that lens alone (#455). Verify every finding's premise yourself before coding a fix — read the comments on the item it names
test cases passed, vs 9 / 20 without
More skills from the same source and nearby in the registry, with their headline benchmark lift.