The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on executing the herdr CLI through various shell scripts and Python sub-processes.
scripts/next_grid_split.py uses subprocess.run to call herdr pane layout and herdr pane resize for grid management.
scripts/wait_for_idle.py and scripts/preflight_send.py use subprocess.run to poll agent status and read transcripts.
scripts/broadcast.sh uses shell commands to resolve agent names and dispatch tasks via herdr pane run.
[DATA_EXPOSURE]: The skill accesses Herdr-specific environment variables (HERDR_PANE_ID, HERDR_TAB_ID, HERDR_WORKSPACE_ID) and reads agent transcripts to coordinate actions. This data is handled locally and remains within the Herdr/agent context.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from sub-agents, creating an attack surface for indirect injection. However, it implements defensive measures such as 'completion markers' (randomly generated strings) to ensure the agent only captures intended replies and distinguishes them from prompt echoes or stale output.

herdr-agent-comms