The Agent Skills Directory

[COMMAND_EXECUTION]: The file references/recording.md provides documentation instructing users to execute system-level commands requiring administrative privileges using sudo. These instructions cover trusting local SSL certificates (sudo caddy trust, sudo security remove-trusted-cert) and modifying the system-wide /etc/hosts file. These operations alter security configurations and require elevated access.- [DATA_EXFILTRATION]: The script scripts/audio_events.py contains a hardcoded absolute directory path (/Users/mars/Downloads/ideal-production-demo/tools). This reveals a local username and potential directory structure from the developer's environment, constituting a data exposure issue. It also results in the script attempting to write to a non-existent or unintended path on other users' systems.- [PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection as it is designed to ingest and analyze untrusted frontend code (mockups, staging builds) provided by the user. * Ingestion points: The INTAKE phase in SKILL.md and the Finders logic in references/demo-driving.md describe mechanisms for the agent to scan and interact with external DOM elements and source code. * Boundary markers: There are no explicit instructions for using delimiters or boundary markers to prevent the agent from following malicious instructions potentially hidden within the user-supplied frontend data. * Capability inventory: The agent can execute bundled Python scripts which perform file system operations and data processing. * Sanitization: The skill lacks explicit sanitization or safety checks for the content extracted from the user's frontend before it is processed by the agent.

elite-product-demos