The Agent Skills Directory

[DATA_EXFILTRATION]: The skill directs the agent to read highly sensitive Google Play service account keys from the /secrets/ directory. Accessing these files exposes private credentials to the agent's context. Evidence: Example commands in SKILL.md such as gps accounts add phosum --credentials /secrets/phosum-sa.json.
[PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection by ingesting data from a local configuration file. 1. Ingestion points: The skill reads the ~/.config/gps/config.json file to list applications. 2. Boundary markers: None present. There are no instructions to delimit or ignore potentially malicious content within the configuration file. 3. Capability inventory: The agent is empowered to execute various gps CLI tools and manage developer accounts. 4. Sanitization: No sanitization, validation, or schema enforcement is described for the configuration data.

gps-accounts