The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires the execution of multiple external shell scripts located in ~/agent-tools/, including ralph-init.sh, ralph-run.sh, ralph-status.sh, and ralph-stop.sh. These scripts are used to initialize project plans and run the autonomous coding loop as described in SKILL.md and references/README-ralph.md.
[REMOTE_CODE_EXECUTION]: The primary purpose of the 'Ralph loop' is 'Autonomous Execution' (SKILL.md, Workflow 1), where the agent generates and runs code through fresh instances of the AI model. The skill explicitly promotes 'unattended execution' and 'running overnight', which allows the agent to modify the local environment and execute generated code without immediate human supervision.
[EXTERNAL_DOWNLOADS]: The references/README-ralph.md file states that the necessary scripts are 'installed' in the user's home directory, but the skill does not provide the script contents or a verifiable source for their installation, creating a dependency on unverified third-party code.
[DATA_EXFILTRATION]: Given that the skill facilitates autonomous file system access and code execution, it creates a significant risk surface where a malicious implementation plan could be used to access sensitive system files mentioned in the documentation (such as ~/.zshrc) and transmit them if network tools are available.

ralph-loop