The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes local Python scripts (brigade.py, land.py) to perform administrative tasks, state management, and conflict resolution within the project store.
[PERSISTENCE]: Utilizes the schedule_task host tool to establish recurring services for 'intake', 'morning', and 'evening' updates, maintaining long-term operation within the agent platform.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external sources and other agents, creating a potential vector for indirect attacks.
Ingestion points: Consumes content from menu.md, restaurant.json, and log.tsv files associated with other coordinators, and processes messages received from their respective threads.
Boundary markers: While it uses some quoting for user input, it lacks robust delimiters or sanitization for content relayed from coordinators to the user or other threads.
Capability inventory: Has the ability to execute shell commands, manage thread messages via t3_thread_send, and schedule platform tasks.
Sanitization: The instructions focus on relaying and summarizing coordinator actions, which may inadvertently propagate malicious instructions embedded in coordinator logs or messages.

brigade-admin