The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/monitor-bus.py contains a call to os.system() used to clear the terminal screen (cls or clear) when in watch mode. While os.system is a sensitive function, its usage here is restricted to hardcoded strings for UI purposes and does not present a security risk.
[INDIRECT_PROMPT_INJECTION]: The skill implements a coordination protocol where multiple agents read and write to shared JSON files. This creates an attack surface where a compromised agent could inject instructions into the message bus to influence other agents.
Ingestion points: queue.jsonl and discussion.jsonl files (read via read_messages and get_messages_for_director functions).
Boundary markers: Data is structured as JSON, providing field separation, but payload content is intended to be processed as natural language by participating agents.
Capability inventory: The skill manages local file I/O; dangerous capabilities like arbitrary shell command execution or network exfiltration are not present in the scripts.
Sanitization: No specific sanitization or validation of message content is performed before storage or retrieval, which is standard for an internal communication protocol but creates an injection surface.

message-bus