The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The file SWARM_CONFIG_TEMPLATE.json contains hardcoded default credentials for a message broker (amqp://guest:guest@localhost:5672/tasks). While these are common defaults for local development, hardcoded credentials in configuration files are a security risk.
[EXTERNAL_DOWNLOADS]: The skill uses npx to download and execute the @modelcontextprotocol/inspector tool from the npm registry. This is a well-known tool associated with the Model Context Protocol.
[COMMAND_EXECUTION]: The skill utilizes several shell commands including npx, curl, and nc for testing connections and transmitting data. It also references the execution of a local Python script (python3 agents/security_agent.py) within the swarm configuration.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process handshake messages and JSON-RPC payloads from external peer agents.
Ingestion points: Handshake messages defined in AGENT_HANDSHAKE_TEMPLATE.json and messages received via network endpoints in SWARM_CONFIG_TEMPLATE.json.
Boundary markers: The skill documentation explicitly includes safety rules requiring strict validation of all JSON/RPC payloads against predefined schemas.
Capability inventory: The skill has capabilities for network access (curl, nc), external tool execution (npx), and local script execution (python3).
Sanitization: The skill implements schema validation and protocol negotiation rules to mitigate the risk of malicious payload execution.

agent-interop