The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill instructs the agent to retrieve a secret API key by sourcing a hidden file in the user's home directory (~/.deepapi/env). Accessing credential files outside of the immediate project scope is a sensitive operation.
[DATA_EXFILTRATION]: The skill transmits research queries and the retrieved API key to the third-party domain https://deepapi.co. While this is the primary function of the skill, the domain is not on the trusted services list.
[COMMAND_EXECUTION]: The skill executes shell commands to load environment configurations and perform network requests using curl to interact with the DeepAPI REST service.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from "production data," "real traffic," and external research responses to perform live measurement suites. This creates a vulnerability surface where instructions embedded in external data could influence the agent.
Ingestion points: Production traffic logs, upstream provider data, and external research results from DeepAPI.
Boundary markers: None specified in the instructions for isolating external data.
Capability inventory: Network communication via curl, file system access for credentials, and writing evaluation reports to the project's docs/evals/ directory.
Sanitization: No sanitization or validation steps are defined for the ingested data.

risky-changes