The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The file references/TROUBLESHOOTING_CF1033.md contains a hardcoded string y^48ZTz3ZJ8J presented as a Cloudflare API Token which does not follow standard placeholder patterns.
[REMOTE_CODE_EXECUTION]: Multiple files including references/BUNDLED_SCRIPTS.md and assets/enhanced-quick-start.md instruct users to pipe remote scripts from external URLs (e.g., raw.githubusercontent.com/YOUR-ORG/... and cdn.coollabs.io) directly into bash for execution.
[COMMAND_EXECUTION]: The skill documentation, particularly in references/INSTALLATION.md, relies on sudo to perform sensitive system modifications such as altering /etc/ssh/sshd_config and overwriting or appending to /root/.ssh/authorized_keys.
[DATA_EXFILTRATION]: references/INSTALLATION.md demonstrates patterns for reading and manipulating sensitive SSH keys, including extracting keys from within a Docker container and installing them into the host's root SSH configuration.
[EXTERNAL_DOWNLOADS]: The skill fetches installation scripts and binaries from external sources including github.com/cloudflare, download.docker.com, and cdn.coollabs.io for product setup and configuration.
[PROMPT_INJECTION]: Instructions in SKILL.md use strong imperative language such as 'STOP' and 'MANDATORY' to enforce user input collection but do not attempt to bypass agent safety constraints.

coolify