The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a bundled Python script (scripts/watch.py) to execute local command-line tools, specifically git and herdr, to monitor workspace state and agent activity. This behavior is consistent with the skill's stated purpose as a monitoring tool.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from other agents and repository metadata, creating a surface for indirect prompt injection.
Ingestion points: Terminal output and status information are read from other agents via herdr agent list and herdr agent read. Repository state (branch names, commit messages) is read via git commands in scripts/watch.py.
Boundary markers: The instructions in SKILL.md provide behavioral boundaries, such as 'Stay read-only' and 'Never delete anything yourself.'
Capability inventory: The skill can execute shell commands (git, herdr), read agent outputs, and interact with other sessions via herdr agent send-keys as described in the instructions.
Sanitization: The Python script performs standard JSON parsing of tool outputs but does not implement specific sanitization or filtering to prevent adversarial instructions embedded in the monitored data.

overlord