Security audit
Security checks for vulnerabilities and agentic risk
This GLSL reference skill discloses its local cache and official-doc lookup behavior, and its scripts are scoped to workspace notes and docs.vulkan.org pages.
Install only if you are comfortable with the skill creating a .GLSL-Encyclopedia directory in the active workspace and caching consulted GLSL documentation plus local shader notes there. Do not put secrets in those notes, and review cached content if workspace files are shared or committed.
The skill's stated purpose centers on answering GLSL questions, but it also initializes directories and manages a persistent workspace layout. That hidden operational behavior expands the attack surface by creating or modifying files in the environment, which can be abused or can violate least-surprise expectations for a supposedly knowledge-focused skill.
The skill's stated purpose centers on answering GLSL questions, but it also initializes directories and manages a persistent workspace layout. That hidden operational behavior expands the attack surface by creating or modifying files in the environment, which can be abused or can violate least-surprise expectations for a supposedly knowledge-focused skill.
The skill instructs the agent to perform network access and local file writes, but it does not declare any explicit tool scope or permission boundary. That creates an authorization gap: a caller may believe this is a read-oriented GLSL reference skill, while the skill can fetch remote content and persist data locally, increasing the risk of unintended data writes, cache poisoning, or broader-than-expected tool use.
The manifest describes a GLSL-specific skill for language/specification questions, but this script is explicitly hardcoded to fetch from docs.vulkan.org and cache pages under /glsl/latest. That behavior broadens the implementation into maintaining an external documentation cache tied to a Vulkan-hosted site, which is not stated in the manifest description.
The code performs outbound HTTPS requests and writes fetched content into a local cache directory. For a skill described as handling GLSL syntax, semantics, and compiler-error triage, persistent network-backed document acquisition is an additional capability that is not explicitly justified by the manifest text.
No suspicious patterns detected.
