The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits a vulnerability to indirect prompt injection because it retrieves and processes unstructured data from external websites to research runtime documentation.
Ingestion points: Processes data from multiple documentation URLs using WebFetch and WebSearch as defined in the SKILL.md process.
Boundary markers: Lacks delimiters or instructions to ignore potential commands embedded in the fetched documentation content.
Capability inventory: Accesses tools including Bash, Read, Glob, and Grep, creating an execution surface if malicious instructions are ingested.
Sanitization: Fetched documentation is parsed and summarized without validation or sanitization of its content.
[EXTERNAL_DOWNLOADS]: The skill fetches documentation from several well-known technology domains including Anthropic's Claude documentation, OpenCode, Gemini CLI, and OpenAI's Codex documentation to maintain format compatibility.
[COMMAND_EXECUTION]: Employs Bash to execute grep searches within the project's bin/install.js file. These commands are restricted to auditing specific function signatures within the local repository.

gsp-runtime-compat