The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches content from a remote GitHub repository (https://raw.githubusercontent.com/herdrdev/herdr/refs/heads/master/skills/herdr/SKILL.md) to perform skill updates.
[REMOTE_CODE_EXECUTION]: The agent is specifically instructed to update its local skill definition with new features or corrections downloaded from a remote URL. This mechanism allows remote content to modify the agent's behavior and operational logic.
[COMMAND_EXECUTION]: The skill uses the herdr command-line tool to execute various terminal operations, including creating workspaces, splitting panes, and running arbitrary commands (herdr pane run).
[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted data by reading the output of terminal panes and agent logs, which could contain malicious instructions from other running processes.
Ingestion points: Terminal output and agent transcripts are ingested via herdr pane read and herdr agent read in SKILL.md.
Boundary markers: None identified. The instructions do not specify the use of delimiters or sanitization when processing the output of these reads.
Capability inventory: The skill has significant local capabilities, including starting new agents (herdr agent start), running shell commands (herdr pane run), and sending keyboard input to other processes (herdr agent send-keys).
Sanitization: None identified. The skill does not mention validation or escaping of the content read from terminal panes before it is acted upon.

herdr