The Agent Skills Directory

[SAFE]: The skill functions primarily as an API wrapper and documentation library for the Jev decision-making service. All network communication in the bundled Python script scripts/jev.py targets official, well-known API endpoints (openrouter.ai and typesafe.ai) and is used to fetch model judgments.
[SAFE]: The bundled script scripts/jev.py adheres to security best practices. It checks for the presence of API keys (OPENROUTER_API_KEY, TYPESAFE_API_KEY) in the environment without printing their values, uses standard libraries only, and avoids any dynamic code execution (e.g., eval/exec).
[PROMPT_INJECTION]: A deterministic detector flagged potential injection patterns in documentation files such as references/community.md and references/simulation.md. A manual review confirms these are false positives. The documentation discusses 'jailbreaks' and 'injections' in the context of using Jev as a security filter (e.g., recipe A11) to detect malicious instructions in untrusted data. Furthermore, instructions in references/simulation.md are defensive, advising the agent to treat data as untrusted and ignore embedded instructions.
[COMMAND_EXECUTION]: The skill instructions allow the agent to run the bundled Python script for valid tasks. The script's logic is restricted to validating input JSON, performing HTTP requests, and reporting API responses without arbitrary shell execution capabilities.

jev