The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill implements a secure credential management system using a dedicated configuration file (~/.pexo/config). It explicitly instructs users to set restricted permissions (chmod 600) and the _common.sh script ensures that authenticated requests are only sent to the authorized vendor domain https://pexo.ai to prevent credential leaking.\n- [COMMAND_EXECUTION]: The skill utilizes standard, pre-defined system utilities (curl, jq, file) to facilitate API communication and file handling. The internal script _common.sh contains specific security logic to sanitize configuration inputs, preventing shell command injection during the parsing of configuration values.\n- [DATA_EXFILTRATION]: Outbound network communication is restricted to the vendor's API endpoint. The skill includes clear documentation and logic to ensure that only user-approved assets and project briefs are transmitted, with a mandatory consent step before the first transmission in a session to protect user privacy.\n- [INDIRECT_PROMPT_INJECTION]: While the skill processes user-supplied text and media, it employs strict boundary markers (tags like ) and utilizes jq for JSON construction to ensure that user input is treated as data rather than instructions. The agent is specifically directed to act as a delivery worker, relaying exact user messages without creative modification, which minimizes the risk of the agent being manipulated by malicious embedded instructions in user data.

pexo-agent