Skillv1.0.0

ClawScan security

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

BenignMar 11, 2026, 10:03 AM

Verdict: benign
Confidence: high
Model: gpt-5-mini
Summary: This is an instruction-only
Guidance: This skill is instruction-only and internally consistent with its purpose: it asks the agent to run a structured, candid quality review and produce a fixed-format response. There is no code, no network endpoint, and it requests no secrets. Before installing, consider: (1) how you plan to use it — don't send highly sensitive secrets or private PII as the 'deliverable' to be reviewed; (2) the skill's tone is explicitly blunt (it instructs the agent not to soften criticism), so expect direct feedback; (3) review the example outputs to ensure the format and bluntness fit your workflow; (4) if you prefer manual control, turn off autonomous invocation so the skill runs only when you call it. Overall this skill appears coherent and proportionate to its stated purpose.

Purpose & Capability: okName and description (a structured quality review called 'Caring') match the contents of SKILL.md and the included examples. The skill requests no binaries, env vars, or config paths — which is appropriate for a purely instructional review tool.
Instruction Scope: okSKILL.md defines a clear, narrow runtime behavior: apply six specific review checks and produce a fixed markdown-style report. It does not instruct the agent to read system files, access environment variables, call external endpoints, or exfiltrate data. The scope is limited to analyzing the user-provided deliverable.
Install Mechanism: okNo install specification and no code files beyond markdown examples — the skill is instruction-only. That is the lowest-risk install model (nothing written to disk or downloaded by the skill).
Credentials: okThe skill declares no required environment variables, no credentials, and no config paths. There are no requests for secrets or unrelated service tokens, which is proportionate to its stated purpose.
Persistence & Privilege: noteFlags are defaults: always:false, user-invocable:true, disable-model-invocation:false. Allowing the agent to invoke the skill autonomously is the platform default; this is not itself a red flag. If you want manual control, you can disable autonomous invocation in the agent settings.