The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill instructs the agent to read the contents of config/secrets.md, which is a sensitive file path specifically used for storing private authentication tokens and secrets.
[DATA_EXFILTRATION]: Sensitive credentials retrieved from the local filesystem are transmitted to a non-trusted third-party domain (https://api.apicore.ai/v1/images/generations) via an Authorization header. This pattern of reading local secrets and sending them to a third-party service provider constitutes a high-risk data exfiltration vector.
[COMMAND_EXECUTION]: The skill uses the curl command to perform network operations. When combined with sensitive file access, this provides a direct mechanism for the exfiltration of credentials or other local data to external endpoints.
[PROMPT_INJECTION]: The skill demonstrates an indirect prompt injection surface. 1. Ingestion points: Untrusted user input is accepted in the form of image descriptions and URLs. 2. Boundary markers: Absent; user-supplied content is interpolated directly into a JSON payload for a shell command. 3. Capability inventory: Uses curl for network access to external APIs. 4. Sanitization: Absent; there is no validation or filtering of the input before it is used in the network request.

gemini-image