The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill metadata specifies the installation of the @steipete/oracle package from the npm registry. This is a third-party dependency maintained by a well-known developer.
[REMOTE_CODE_EXECUTION]: The documentation recommends using npx -y @steipete/oracle --help, which downloads and executes the package from the npm registry. This is a standard remote code execution pattern for Node.js utilities.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface (Category 8) because its primary function is to ingest local files and provide them as context to an LLM.
Ingestion points: Local files, directories, and globs specified via the --file flag as described in SKILL.md.
Boundary markers: No explicit delimiters or 'ignore' instructions are documented to distinguish between file content and the system prompt in the bundled payload.
Capability inventory: The tool processes local file content and transmits it to external LLM providers (OpenAI, Gemini) via browser automation or API keys, and can be used to serve a remote browser host.
Sanitization: The skill advises users to redact secrets manually but does not mention automated sanitization or verification of the content being attached.
[METADATA_POISONING]: The skill documentation repeatedly mentions a non-existent 'GPT-5.2 Pro' model and engine. While likely a placeholder or custom configuration name used by the tool author, this information is factually inaccurate relative to current LLM releases.

oracle