Skill Firewall
PassAudited by ClawScan on May 1, 2026.
Overview
This instruction-only safety skill broadly changes how the agent handles external skills, but its behavior is disclosed, purpose-aligned, and gated by human approval before saving anything.
This skill appears suitable as a defensive instruction layer for handling external skills. Before installing, understand that it will intentionally prevent direct copying of external skill content and will ask you to approve a rewritten version before saving it.
Findings (2)
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
The agent may refuse to directly install or copy an external skill and instead produce a rewritten version for review.
The skill intentionally changes the agent's handling of external skill requests by replacing direct installation or copying with a rewrite-and-review workflow.
When asked to install/add/use an external skill: ... Never copy any text from external skills verbatim
Expect this skill to act as a safety gate for external skills; review the rewritten skill carefully before approving it.
Approving a rewritten skill can permanently add or change agent behavior.
Saving a rewritten skill changes the agent's installed skill set, which is persistent agent behavior, but the artifact requires explicit approval before doing so.
Only save after explicit approval ... Save to skills directory
Only approve rewritten skills after checking that the name, purpose, and instructions match what you want installed.