The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill relies on several well-known and trusted external libraries for its core functionality.
Python dependencies: nats-py, anthropic, openai, jsonschema, and OpenTelemetry SDKs.
Node.js dependencies: nats, @anthropic-ai/sdk, openai, and uuid.
[REMOTE_CODE_EXECUTION]: Automated scans detected a pattern involving piping network output to a Python interpreter. In several example scripts (e.g., examples/cross-org/test.sh and examples/cross-org/README.md), the skill fetches JSON data from a local NATS monitoring port and pipes it to python3 -c or python3 -m json.tool for parsing or formatting. While this involves piping data to an interpreter, the source is localhost and the execution is intended for data processing in a developer environment.
[COMMAND_EXECUTION]: Multiple examples and setup guides (e.g., examples/cli/complete-cli-mesh.sh, setup.sh) execute shell commands to manage NATS servers, generate cryptographic keys, and run agent processes. These are consistent with the skill's purpose as a developer implementation guide.
[PROMPT_INJECTION]: The skill contains multiple 'LLM Agent' examples that are susceptible to indirect prompt injection.
Ingestion points: Untrusted data enters the agent context through NATS message payloads in claude-agent.ts, claude_agent.py, and orchestrator-agent.ts.
Boundary markers: Absent; the skill does not use delimiters to wrap the payload.input.message before passing it to the model.
Capability inventory: The agents have the capability to make API calls to Anthropic or OpenAI models (e.g., claude.messages.create) based on these inputs.
Sanitization: Absent; the inputs are interpolated directly into the message content.

synapse