Skillv1.0.0

ClawScan security

claude-authenticity · ClawHub's context-aware review of the artifact, metadata, and declared behavior.

BenignMar 16, 2026, 9:31 AM

Verdict: Benign
Confidence: high
Model: gpt-5-mini
Summary: The skill is internally coherent: it is an instruction-only Claude-authenticity checker that asks the user for an API endpoint and key and provides a self-contained Python script to run locally.
Guidance: This skill is coherent and works by sending requests you configure to the target endpoint using an API key you provide. Before using it: (1) review the included Python code yourself — do not paste secret or production keys into unknown scripts; (2) only test endpoints you own or have permission to probe; (3) use a temporary or limited-scope API key and run the script in an isolated environment (e.g., a disposable VM or container); (4) be aware that the optional 'extract prompt' behavior intentionally tries to reveal injected system prompts and could cause the endpoint to return hidden content; and (5) expect false positives/negatives — results are heuristic and should be one signal among others.

Purpose & Capability: okSkill name/description match the provided instructions and script: it implements rule-based checks to evaluate whether an endpoint is backed by Claude. The runtime asks for an API endpoint, API key, model names and an API type — inputs that are necessary for its stated purpose.
Instruction Scope: noteInstructions contain a self-contained Python script that will send crafted requests to the user-supplied endpoint and optionally attempt to extract any injected system prompt. This is consistent with the stated goal, but note that prompt-extraction is an active probing behavior (it deliberately tries to coax the provider to reveal hidden prompts) and will transmit the supplied API key and probe payloads to the target endpoint.
Install Mechanism: okNo install spec in registry; SKILL.md simply instructs to pip install httpx. Instruction-only distribution is low-risk compared to downloading arbitrary binaries.
Credentials: noteRegistry metadata lists no required env vars, but the SKILL.md expects the user to provide an API key and endpoint at runtime. That is proportionate to the tool's purpose. Important: the key will be sent to the target endpoint, so the user must supply a key they control and be aware of exposure risk.
Persistence & Privilege: okSkill is instruction-only, has no install step that writes to disk, and does not request persistent privileges. Flags: always is false and model invocation is allowed (platform default).