Loading the registry…
Audit the evaluation layer of an AI product against design-completeness questions: offline criteria, ground-truth quality, online signal, cohort/disparate-impact, adversarial + robustness coverage, and drift detection. Eval-side companion to `ai-ux-review`. Produces an editable Markdown artifact plus a self-contained HTML report under `docs/ai-ux/`. Use when the user asks to "review my AI eval setup", "audit my eval design", "is my AI eval rigorous enough", "responsible-AI eval review", "fairness eval check", or "drift detection design", or ships an LLM/ML feature and wants an eval-rigor check. Invoke even if only one block is named — the others stress-test it. Does NOT trigger for: human-AI UX review (`ai-ux-review`); lean-canvas work (`validation-canvas`); adversarial pre-mortem with a verdict (`startup-grill`); SKILL.md audits (`skill-evaluator`); implementing eval pipelines, writing eval code, or labeling datasets (this skill names gaps, it does not build them).
Claude Skills are markdown + scripts loaded into Claude Code; scanned with the same pipeline.
We found this by crawling public sources and nobody has attested to it. That is how most of the registry gets here, and it is not a mark against the artifact — it only means no authenticated act of publication is on record. An artifact can be the vendor's own and still appear here.
307 critical findings reported by agent-audit-kit, osv-scanner, 1 of them in the installed artifact. The composite score alone does not reflect this — read the per-scanner reports below.
Claude Skills install to a single target, no runtime picker.
nerlo install -- ai-eval-review
Manual alternative: copy the skill folder into ~/.claude/skills/ so Claude Code loads it on next launch.
cp -r -- ai-eval-review ~/.claude/skills/
Install respects the composite badge: Clean proceeds, Caution prompts for confirmation, and Flagged is refused. Write operations require an API token.
5 of 11 scanners examined this package. The rest do not apply to it — language, ecosystem and packaging decide which analyzers can say anything, and a scanner that cannot examine an artifact reports nothing rather than passing it.
v2.0.11
164 findings · 20.0s
v0.3.26
354 findings · 4.9s
v0.1.0
0 findings · 5.1s
v0.1.0
5 findings · 5.1s
v2.3.8
18 findings · 1.5s
v1.4.0
None of the entry points this scanner reads were found in this artifact. What each scanner covers
0 findings · 0.3s
vv0.3.2
No Go packages were found in this artifact. What each scanner covers
0 findings · 0.1s
v0.71.0
No dependency manifest or lockfile was found in this artifact. What each scanner covers
0 findings · 0.3s
v0.71.0
No container image was published for this artifact. What each scanner covers
0 findings · 0.0s
vv1.6.0
No Go modules were found in this artifact. What each scanner covers
0 findings · 0.1s
No report in this scan: nerlo-multi-source.
1 scan on record. Every scan's full results are retained immutably for 24 months.
| Completed | Composite | Change | Scanners | Status |
|---|---|---|---|---|
| Aug 14, 2026 | Flagged25 | — | agentshield ·n/a: not applicablecisco-skill-scanner: completeagent-audit-kit: completenerlo-behavioral: completenerlo-install-instruction: completecapslock ·n/a: not applicabletrivy ·n/a: not applicableosv-scanner: completetrivy_image ·n/a: not applicablegovulncheck ·n/a: not applicable | completed |
Findings in files this artifact installs stand as reported; a model re-read the rest and said which ones it believes are false positives. This is a second opinion published beside the evidence, not a correction to it: the per-scanner reports above are unchanged, every dismissed finding is still listed there at its original severity, and the score and the Flagged badge are computed from those raw severities alone. Nothing below moved them.
This review has not run for this scan yet. It runs once per scan, in the background, and is kept; it appears here when it has. Until then all 541 findings above stand exactly as their scanners reported them, at full severity.