Back to skill
Skillv1.0.0
ClawScan security
Website Auditor · ClawHub's context-aware review of the artifact, metadata, and declared behavior.
Scanner verdict
SuspiciousMar 3, 2026, 3:26 AM
- Verdict
- suspicious
- Confidence
- high
- Model
- gpt-5-mini
- Summary
- The skill's code and runtime instructions mostly match a website-auditor, but the SKILL.md declares required packages and a PAGESPEED_API_KEY while the registry metadata says no env or installs — this mismatch and the inclusion/forwarding of raw HTML are notable concerns.
- Guidance
- This skill appears to implement a reasonable website auditor, but there are a few practical and privacy issues you should resolve before installing: - Metadata mismatch: SKILL.md declares a required environment variable (PAGESPEED_API_KEY) and lists Python packages, yet the registry entry lists no required env or install steps. Ask the publisher to correct the registry metadata or provide a clear install spec so you know what the skill needs and why. - Dependencies: The skill imports non-standard Python packages (Wappalyzer, python-whois, lxml). If your agent environment doesn't already have these, the skill will fail. Prefer an install spec (pip/packaging) or run the skill in a sandboxed environment. - Raw HTML disclosure: The skill returns raw_html and documents it will be consumed by another skill. That can leak sensitive content (forms, tokens in page markup, private data). Confirm how downstream skills handle that data and whether you are comfortable with the coupling. - Credential handling: Only provide a PAGESPEED_API_KEY if you trust the skill. The key is appropriate for the feature, but the registry mismatch means the platform may not prompt you to supply it. - Network activity: The skill performs active network probes (HTTP fetches, SSL cert checks, whois, third-party API calls). If you need to limit outbound network access, run this skill in a restricted/sandboxed context. If you cannot get clarification from the publisher, consider treating the skill as untrusted: run it in an isolated environment, do not supply privileged credentials, and audit any downstream consumers that will receive raw_html.
Review Dimensions
- Purpose & Capability
- noteThe declared runtime actions (HTTP checks, Wappalyzer tech detection, PageSpeed API, whois/SSL checks) are coherent with a website audit. However the skill's SKILL.md metadata lists required packages and an environment variable (PAGESPEED_API_KEY) while the registry entry shows no required env vars or install steps — that's an inconsistency that could cause the skill to fail or hide needed permissions.
- Instruction Scope
- noteInstructions stay within the stated purpose (fetch page, analyze HTML/headers, detect tech, query PageSpeed). Two things to flag: (1) the skill returns raw_html in the output and explicitly notes it's used by another skill (contact-enrichment), which means page content may be forwarded to other components — a privacy/exfiltration risk depending on downstream handling; (2) the SKILL.md contains network/socket operations (requests, Wappalyzer, whois, SSL checks) which are expected but should be understood as active network probes.
- Install Mechanism
- concernThis is instruction-only with no install spec. SKILL.md lists Python packages (requests, beautifulsoup4, lxml, python-Wappalyzer, python-whois) but there is no platform-level install instruction. That means the runtime must already provide these packages or the skill will fail. Absence of an install spec is a practical/operational risk and increases likelihood of silent failures.
- Credentials
- noteThe only credential referenced in SKILL.md is PAGESPEED_API_KEY, which is proportionate for calling Google PageSpeed API. The registry metadata, however, does not declare this required env var — the mismatch is concerning because users/platforms won't be warned to supply the key, and the skill may behave differently if the key isn't present.
- Persistence & Privilege
- okNo elevated persistence requested (always:false). The skill is user-invocable and can be invoked autonomously (platform default) but it does not request system-level config changes or cross-skill configuration edits in the provided instructions.