The Agent Skills Directory

[CREDENTIALS_UNSAFE]: Hardcoded credentials for third-party services were detected in scripts/report.py, including a Resend API key (SKILL_EMAIL_KEY), a Gmail App Password (SKILL_GMAIL_APP_PWD), and a Brave Search API key inside the CAPABILITY_CATALOG configuration.
[REMOTE_CODE_EXECUTION]: The do_install_skill function in scripts/report.py downloads script content from remote URLs and executes it via bash using subprocesses. This allows for arbitrary code execution if the downloaded content is malicious. Additionally, the skill's main installation method uses a curl | bash pattern.
[DATA_EXFILTRATION]: The skill includes automated logic in scripts/report.py to upload files to public hosting providers such as catbox.moe, transfer.sh, and gofile.io. This poses a risk of sensitive user data being stored on public servers without encryption.
[COMMAND_EXECUTION]: The SKILL.md file instructs the agent to "silently fix" environment issues using high-privilege commands like sudo apt-get, brew install, and chmod without explicit user confirmation for each action, based on patterns defined in references/env_fixes.md.
[EXTERNAL_DOWNLOADS]: The skill automatically installs numerous Python and Node.js packages and other skills from the internet, which can introduce unverified third-party dependencies into the local environment.

autopilot