The Agent Skills Directory

EXTERNAL_DOWNLOADS (HIGH): The skill relies on npx -y @steipete/oracle, which fetches and runs code from the npm registry without user confirmation. The author @steipete is not a pre-approved trusted source, posing a risk of supply chain attacks.
REMOTE_CODE_EXECUTION (HIGH): The skill documents a 'Remote browser host' feature (oracle serve --host 0.0.0.0) that opens a listening port (9473) on the host machine. This allows remote control of the system via a token, which is a high-risk capability if exposed.
DATA_EXFILTRATION (MEDIUM): The skill is designed to read local files via the --file argument and transmit their contents to external AI model providers (OpenAI, Gemini) or browser sessions. This creates a risk of exfiltrating proprietary code or configuration secrets.
COMMAND_EXECUTION (MEDIUM): The skill facilitates the execution of shell commands through the npx utility, providing the AI agent with a direct method to run arbitrary code on the local machine.
PROMPT_INJECTION (LOW): The skill is vulnerable to indirect prompt injection because it processes untrusted local files without explicit sanitization or boundary enforcement.
Ingestion points: Local files and directories provided via the --file flag.
Boundary markers: Absent; the skill bundles raw file content directly into the model prompt.
Capability inventory: Shell command execution (npx), file system read access, and network communication (to AI APIs).
Sanitization: Absent; the instructions do not specify any validation or filtering of the file contents before they are processed by the AI.

oracle