The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill's installation process and documentation facilitate downloading and executing code from remote URLs. Specifically, SKILL.md and README.md recommend executing curl -fsSL https://psh.now/install.sh | sh, and the internal install.sh script downloads multiple executable scripts (ask.sh, setup.sh, status.sh, psh.sh) from psh.now and runs them locally.\n- [PRIVILEGE_ESCALATION]: The skill attempts to modify global system state by installing binaries to directories outside the immediate workspace. The install.sh and setup.sh scripts attempt to write the psh CLI to /usr/local/bin, /opt/homebrew/bin, and ~/.local/bin, and use chmod +x to grant execution permissions to these files.\n- [DATA_EXFILTRATION]: The skill is designed to send task context externally. The ask.sh script transmits task titles and detailed messages to the psh.now API. While this is the primary function of the skill (human-in-the-loop approval), it constitutes a transfer of agent context to a third-party service.\n- [INDIRECT_PROMPT_INJECTION]: The skill provides an attack surface for remote instructions to influence agent behavior. \n
Ingestion points: The answer field is parsed from the psh.now API response in ask.sh.\n
Boundary markers: There are no delimiters or safety instructions when handling the remote answer string.\n
Capability inventory: The skill allows shell command execution via the psh CLI and associated scripts.\n
Sanitization: No validation or sanitization is performed on the remote input; the agent is explicitly instructed in SKILL.md to "read answer, do what it says".\n- [COMMAND_EXECUTION]: The skill relies on multiple shell scripts that perform extensive file system operations, networking via curl, and local script execution to manage its lifecycle.

psh-now