The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The /system/config endpoint exposes sensitive configuration data, including LLM api_key values and Neo4j database connection details. Accessing this data creates a risk of credential exposure if the agent outputs the full configuration object.
[COMMAND_EXECUTION]: The /mcp/import endpoint allows for the execution of arbitrary shell commands on the host system. It specifically uses npx to download and run Node.js packages, which can be exploited if an attacker provides a malicious package name or command string.
[PROMPT_INJECTION]: The skill allows the agent to read and overwrite its own system instructions via the /system/prompt endpoint. This could be used to bypass security constraints or permanently alter the agent's behavior through persistent prompt modification.
[EXTERNAL_DOWNLOADS]: The skill instructions explicitly encourage the agent to search the internet (via online_search) for MCP tools and then install them using npx. This creates a direct path for indirect prompt injection or supply chain attacks where the agent executes untrusted code found on public repositories or registries.
[DATA_EXFILTRATION]: Since the skill can read the entire system configuration and also has access to network-capable tools (like online_search), there is a risk that sensitive system secrets could be exfiltrated to an external server if the agent is compromised.

naga-config