The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external Markdown documentation to answer user queries, which represents an indirect prompt injection surface.
Ingestion points: Documentation index and pages are fetched from https://docs.nvidia.com/nemoclaw/.
Boundary markers: The skill instructs the agent to cite sources with direct links, which helps distinguish documentation content from agent instructions.
Capability inventory: The skill uses the searchDocs tool and standard HTTP retrieval to fetch content.
Sanitization: Content ingestion is restricted to official vendor-controlled documentation, reducing the risk of malicious payloads.
[COMMAND_EXECUTION]: The skill provides commands for configuring the agent's environment, specifically for adding a documentation search server.
Evidence: Instructions in references/docs-access.md provide the command claude mcp add --transport http fern-docs https://docs.nvidia.com/nemoclaw/_mcp/server.
Context: The command configures an MCP server hosted on the official NVIDIA domain, which is the vendor for this skill.
[DATA_EXFILTRATION]: The skill includes proactive security instructions to prevent the mishandling of sensitive user data.
Evidence: SKILL.md contains explicit directives to "Stop before requesting credentials, API keys, bot tokens, or private URLs" and "Never ask the user to paste secrets into chat," mandating the use of redacted placeholders in examples.

nemoclaw-user-guide