message-bus
Pass
Audited by Gen Agent Trust Hub on Oct 10, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The script
scripts/monitor-bus.pyusesos.systemto clear the terminal screen. This operation uses hardcoded literals ('cls' and 'clear') and does not incorporate any external input, making it a safe use of command execution for UI purposes. - [INDIRECT_PROMPT_INJECTION]: The skill enables multi-agent coordination through shared files, which creates a surface for indirect prompt injection if an agent trusts instructions embedded in messages from other agents.
- Ingestion points: Data is ingested from
queue.jsonl,board/assessments.json, andboard/discussion.jsonlin the.message-busdirectory. - Boundary markers: The protocol relies on JSON structure but lacks specific prompt-level delimiters or instructions to ignore embedded commands in message payloads.
- Capability inventory: The skill documentation outlines agents evaluating, discussing, and voting on content retrieved from the bus, while secondary scripts handle file management.
- Sanitization: There is no evidence of content sanitization for message payloads within the provided protocol scripts.
Audit Metadata