The Agent Skills Directory

[COMMAND_EXECUTION]: The monitoring script scripts/monitor-bus.py uses os.system to clear the terminal screen during the watch loop. Evidence (scripts/monitor-bus.py): os.system('cls' if os.name == 'nt' else 'clear').
[PROMPT_INJECTION]: The design of the message bus and board protocol allows agents to ingest and process data from other agents, which presents an attack surface for indirect prompt injection.
Ingestion points: Data enters the agent context through shared files including queue.jsonl, assessments.json, and discussion.jsonl (SKILL.md, scripts/monitor-bus.py).
Boundary markers: There are no explicit delimiters or system instructions used in the message templates to distinguish between control commands and data content.
Capability inventory: Agents utilizing the bus can read and write files within the track_path, potentially influencing orchestrator decisions based on deltas in the board/ files.
Sanitization: The provided implementation relies on basic JSON parsing and does not sanitize or validate strings retrieved from the message queue before they are interpolated into director prompts.
[SAFE]: The skill's functionality is consistent with its stated purpose of providing inter-agent communication infrastructure using standard local file system operations.

message-bus