The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill includes instructions that compromise credential security by printing them to the console.
It executes env | grep -E "(API_KEY|API_TOKEN|LANGSMITH|OPENAI)" to display active environment secrets.
It executes cat backend/.env | grep -E "^[A-Z_]*API" which reads sensitive configuration files and filters for API keys to be displayed in the output.
[REMOTE_CODE_EXECUTION]: The skill contains a high-risk command sequence in section 5: curl -s http://localhost:8123/ | python3 -m json.tool. Piping network-delivered content directly into a language interpreter (Python) is a significant security risk if the server's output is compromised or manipulated, and this specific pattern was flagged by automated security scanners as an untrusted RCE vector.
[DATA_EXFILTRATION]: By intentionally printing secrets to the standard output, the skill facilitates the movement of sensitive data (API keys, tokens) into the agent's conversation logs and potentially to the upstream LLM provider's infrastructure.
[COMMAND_EXECUTION]: The skill makes extensive use of powerful shell commands and subprocesses (uv run, pytest, uvicorn, kill -9, and complex grep/sed loops). While standard for CI/CD tasks, these capabilities provide a broad attack surface if the project being validated contains malicious scripts or configurations.

validate