The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to capture and process terminal output from reproduction sessions through commands like pane read and agent read. This creates a vulnerability surface where potentially untrusted code or processes being tested could generate malicious output intended to influence the agent's behavior.
Ingestion points: Terminal output is captured via pane read, agent read, and pane wait-output as described in SKILL.md.
Capability inventory: The skill utilizes the herdr CLI for process execution, file system management (in /var/tmp), and agent control.
Boundary markers: While the instructions advise the agent to "distinguish observed facts from proposed causes," there are no formal delimiters or escaping mechanisms defined for the ingested data.
Sanitization: No explicit sanitization or filtering of terminal output is implemented before processing.
[SAFE]: The skill implements a security-positive pattern for environment isolation by using env -u to unset sensitive Herdr environment variables (e.g., HERDR_SOCKET_PATH, HERDR_SESSION) before launching disposable sessions. This prevents unintended cross-session communication and leaks between the parent and reproduction environments.

herdr-throwaway-repro