The Agent Skills Directory

[DATA_EXFILTRATION]: The skill manages authentication credentials by storing them in a local configuration file and transmitting them to a remote API for legitimate service interaction.
Evidence: scripts/referencer.mjs implements credential persistence in ~/.referencer.toml and network transmission to https://api.referencer.ai/graphql.
Security Mitigations: The runner script explicitly redacts API keys from console output and error messages using a dedicated redactKey function. Additionally, the configuration file is created with 0600 permissions (owner read/write only), ensuring it is protected from other users on the system.
[PROMPT_INJECTION]: The skill processes structured data from the remote Referencer GraphQL service, which constitutes an attack surface for indirect prompt injection if the service returns malicious content.
Ingestion points: Data is received as structured JSON payloads from the Referencer API in scripts/referencer.mjs.
Capability inventory: The skill can read and write its own configuration file and perform network requests to the Referencer API.
Boundary markers: The skill does not use specific boundary markers or instructions to the agent to ignore embedded content when processing remote data.
Sanitization: Data is handled as structured JSON, reducing the risk of accidental execution compared to raw text processing.

referencer