Skillv1.0.1

ClawScan security

Ai Agent Helper Moss · ClawHub's context-aware review of the artifact, metadata, and declared behavior.

BenignMar 14, 2026, 3:40 AM

Verdict: benign
Confidence: high
Model: gpt-5-mini
Summary: Instruction-only prompt/agent-engineering helper that asks for no installs, credentials, or filesystem access and appears to do what its description says.
Guidance: This is an instruction-only skill for prompt engineering and agent-loop design and does not request secrets or install code, so it's internally coherent and low-risk. Before installing, consider: 1) the skill's source and author are unknown—prefer skills with a verifiable homepage or trusted publisher; 2) the _meta.json metadata differs from the registry metadata (ownerId/slug/version mismatch), which is likely a packaging mistake but worth noting; 3) never paste sensitive credentials or private data into prompts or examples provided to the skill; and 4) if you need stronger assurance, ask the publisher for a canonical source or a signed release.

Purpose & Capability: noteThe skill's name, description, and runtime instructions all describe prompt engineering and agent-loop design; no extra capabilities or credentials are requested. Minor packaging inconsistencies exist (different ownerId/slug/version in _meta.json vs registry metadata), which look like a metadata error rather than malicious behavior.
Instruction Scope: okSKILL.md contains bounded guidance for prompt structure, decomposition, and agent loop patterns. It does not instruct the agent to read files, access environment variables, contact external endpoints, or exfiltrate data.
Install Mechanism: okNo install spec and no code files — instruction-only. This minimizes disk/network risk.
Credentials: okNo environment variables, credentials, or config paths are requested. This is proportional to a prompt-engineering helper.
Persistence & Privilege: okNo 'always: true' or other elevated persistence requested; default agent invocation settings are used.