The Agent Skills Directory

[DATA_EXFILTRATION]: The skill includes a file command in cli.sh and a send_file method in client.py that implement a chunked file transfer protocol. This functionality allows the agent to read any local file path and transmit it as Base64-encoded data to a remote NATS server. While intended for file sharing within the agent mesh, this provides a powerful mechanism for data exfiltration if the agent is directed to access sensitive local files.
[PROMPT_INJECTION]: The serve functionality allows an agent to register skills and process incoming requests from the NATS mesh. This creates a surface for indirect prompt injection, where malicious actors on the shared network can send payloads designed to manipulate the agent's behavior through the ingested request data.
[COMMAND_EXECUTION]: The cli.sh script executes the nats CLI tool for mesh operations. It also uses a pipe from curl to python3 -c within the health command to parse monitoring data; this is a safe pattern as it executes a hardcoded parsing script rather than the remote content.
[EXTERNAL_DOWNLOADS]: The skill depends on the nats-py library for Python connectivity and utilizes official NATS infrastructure. It also references Synadia Cloud endpoints (connect.ngs.global), which is an established service provider for NATS services.

synapse-client