The Agent Skills Directory

[COMMAND_EXECUTION]: The skill exposes the herdr pane run command, which allows the agent to execute arbitrary shell commands in any managed terminal pane. This provides a mechanism for significant system modification and can be abused to perform unauthorized actions if the agent is influenced by malicious input.
[DATA_EXFILTRATION]: Through the herdr pane read command, the agent can access the contents of other terminal panes, including visible text and scrollback history. This allows the agent to potentially harvest sensitive information (secrets, logs, or private data) from concurrent terminal sessions that are not part of its primary task.
[PROMPT_INJECTION]: The skill creates a surface for Indirect Prompt Injection by reading and processing terminal output from other panes using herdr pane read.
Ingestion points: Untrusted data enters the agent context via terminal text read from panes (SKILL.md).
Boundary markers: The instructions do not specify any delimiters or warnings to treat pane output as untrusted content.
Capability inventory: The agent can execute commands (herdr pane run), send keyboard input (herdr pane send-text), and spawn additional agent instances (SKILL.md).
Sanitization: There is no evidence of sanitization or filtering of the terminal output before it is processed by the agent.
[COMMAND_EXECUTION]: The skill uses python3 -c to dynamically parse JSON output from CLI commands at runtime, which involves executing code snippets to extract specific values like pane IDs.

herdr