# skills.sh registry check

All 9 exact listings and all 27 detailed audit pages returned HTTP 200; all published verdicts PASS. Public page coverage is complete; exact Git snapshot correspondence is unverified.

Target commit: `80d5f7b0127c7572ed9e9339937adbfd7240ffeb` (2026-09-24T18:21:08Z). Retrieved UTC: 2026-10-01T00:25:27.148214+00:00 through 2026-10-01T00:25:29.959873+00:00.

| Skill | Gen Agent Trust Hub | Socket | Snyk | Displayed audit date |
|---|---|---|---|---|
| [build-review-interface](https://skills.sh/ai-evals-course/evals-skills/build-review-interface) | [PASS / SAFE](https://www.skills.sh/ai-evals-course/evals-skills/build-review-interface/security/agent-trust-hub) | [PASS](https://www.skills.sh/ai-evals-course/evals-skills/build-review-interface/security/socket) | [PASS / LOW](https://www.skills.sh/ai-evals-course/evals-skills/build-review-interface/security/snyk) | Sep 1, 2026 |
| [error-discovery](https://skills.sh/ai-evals-course/evals-skills/error-discovery) | [PASS / SAFE](https://www.skills.sh/ai-evals-course/evals-skills/error-discovery/security/agent-trust-hub) | [PASS](https://www.skills.sh/ai-evals-course/evals-skills/error-discovery/security/socket) | [PASS / LOW](https://www.skills.sh/ai-evals-course/evals-skills/error-discovery/security/snyk) | Sep 1, 2026 |
| [eval-audit](https://skills.sh/ai-evals-course/evals-skills/eval-audit) | [PASS / SAFE](https://www.skills.sh/ai-evals-course/evals-skills/eval-audit/security/agent-trust-hub) | [PASS](https://www.skills.sh/ai-evals-course/evals-skills/eval-audit/security/socket) | [PASS / LOW](https://www.skills.sh/ai-evals-course/evals-skills/eval-audit/security/snyk) | Sep 25, 2026 |
| [evals-start](https://skills.sh/ai-evals-course/evals-skills/evals-start) | [PASS / SAFE](https://www.skills.sh/ai-evals-course/evals-skills/evals-start/security/agent-trust-hub) | [PASS](https://www.skills.sh/ai-evals-course/evals-skills/evals-start/security/socket) | [PASS / LOW](https://www.skills.sh/ai-evals-course/evals-skills/evals-start/security/snyk) | Sep 25, 2026 |
| [evaluate-rag](https://skills.sh/ai-evals-course/evals-skills/evaluate-rag) | [PASS / SAFE](https://www.skills.sh/ai-evals-course/evals-skills/evaluate-rag/security/agent-trust-hub) | [PASS](https://www.skills.sh/ai-evals-course/evals-skills/evaluate-rag/security/socket) | [PASS / LOW](https://www.skills.sh/ai-evals-course/evals-skills/evaluate-rag/security/snyk) | Sep 1, 2026 |
| [generate-synthetic-data](https://skills.sh/ai-evals-course/evals-skills/generate-synthetic-data) | [PASS / SAFE](https://www.skills.sh/ai-evals-course/evals-skills/generate-synthetic-data/security/agent-trust-hub) | [PASS](https://www.skills.sh/ai-evals-course/evals-skills/generate-synthetic-data/security/socket) | [PASS / LOW](https://www.skills.sh/ai-evals-course/evals-skills/generate-synthetic-data/security/snyk) | Sep 1, 2026 |
| [validate-evaluator](https://skills.sh/ai-evals-course/evals-skills/validate-evaluator) | [PASS / SAFE](https://www.skills.sh/ai-evals-course/evals-skills/validate-evaluator/security/agent-trust-hub) | [PASS](https://www.skills.sh/ai-evals-course/evals-skills/validate-evaluator/security/socket) | [PASS / LOW](https://www.skills.sh/ai-evals-course/evals-skills/validate-evaluator/security/snyk) | Sep 25, 2026 |
| [write-code-eval](https://skills.sh/ai-evals-course/evals-skills/write-code-eval) | [PASS / SAFE](https://www.skills.sh/ai-evals-course/evals-skills/write-code-eval/security/agent-trust-hub) | [PASS](https://www.skills.sh/ai-evals-course/evals-skills/write-code-eval/security/socket) | [PASS / LOW](https://www.skills.sh/ai-evals-course/evals-skills/write-code-eval/security/snyk) | Sep 24, 2026, Sep 25, 2026 |
| [write-judge-prompt](https://skills.sh/ai-evals-course/evals-skills/write-judge-prompt) | [PASS / SAFE](https://www.skills.sh/ai-evals-course/evals-skills/write-judge-prompt/security/agent-trust-hub) | [PASS](https://www.skills.sh/ai-evals-course/evals-skills/write-judge-prompt/security/socket) | [PASS / LOW](https://www.skills.sh/ai-evals-course/evals-skills/write-judge-prompt/security/snyk) | Sep 25, 2026 |

Published PASS observations:

- **build-review-interface**: Published analysis describes Markdown design instructions with no executable scripts or command snippets. Rendered traces are untrusted; the skill calls for sanitizing content, removing raw HTML, and disabling Markdown images. Mentions Playwright interface testing; the provider found no malicious commands.
- **error-discovery**: JSONL, CSV, and JSON records are ingestion points. Provider reports no explicit data/instruction boundaries or content sanitization before analysis or subagent use. Instructions include a Python standard-library HTTP server and background subagents for corpus scans. Recommends frontend libraries from CDNs: marked.js and highlight.js or Prism.
- **eval-audit**: Evaluation traces, evaluator definitions, judge prompts, and labels can contain adversarial instructions. Provider reports absent data boundaries and sanitization guidance. It describes the provided skill as instructional, without declared shell, network, or write tools.
- **evals-start**: Routing instructions only; published analysis reports no executable logic, filesystem or network actions, or untrusted-data processing.
- **evaluate-rag**: RAG evaluation methodology, with no scripts, command execution, or network operations identified. Prompt templates and evaluation grids are treated as examples; provider reports no executable files or remote dependencies.
- **generate-synthetic-data**: Application descriptions and dimension values enter prompt templates without explicit delimiters or validation. Malicious inputs could alter query generation.
- **validate-evaluator**: Human labels and production traces enter judge evaluation without stated delimiters or filtering. Provider names labeled_data, human_labels, and prod_eval_labels; statistical examples use scikit-learn and numpy. Recommends pip install judgy for statistical bias correction; that package is a separate dependency outside this registry check.
- **write-code-eval**: Provider describes instructions for writing evaluators; it identifies no executable scripts, command-line operations, network requests, remote dependencies, or malicious patterns.
- **write-judge-prompt**: Markdown and YAML guidance for evaluator prompt design; provider reports no detected security issues. No scripts, binaries, or command execution patterns identified. Processes user-provided failure modes and examples. Provider considers risk negligible in conversational use and attributes no file, network, or shell capabilities to the skill.

No displayed audit date predates its target skill directory’s latest change. Dates omit timezone; this is not proof of content correspondence.

Third-party audits of skills.sh snapshots do not verify publisher identity, exact local artifact, dependencies at install time, or runtime behavior. No target skills were installed or executed.

All exact owner/repo/slug matches were verified from the canonical listing. Some listings record a previous publisher path, hamelsmu/evals-skills; the checked paths are ai-evals-course/evals-skills. Socket package snapshot hashes are saved in [normalized evidence](skills-sh-evidence.json); no Git-commit mapping was established.

Provider observations, retrieval times, snapshot hashes, history metadata, and exact URLs are preserved in [normalized evidence](skills-sh-evidence.json). Raw HTML is retained in the local review archive. Missing confidence, identifiers, issue counts, or timestamps were recorded as unavailable rather than invented.