The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches setup instructions from the vendor's domain (https://revnu.com/setup.md) and connects to a remote MCP server at https://revnu.com/api/mcp.
[COMMAND_EXECUTION]: The instructions guide the user to modify their environment by adding the Revnu MCP server using host-specific commands like 'claude mcp add'. The skill explicitly requires user consent before performing these actions.
[DATA_EXFILTRATION]: To provide tailored growth advice, the skill reads project metadata (e.g., package.json, README.md) and sends public product facts to the vendor's API. It includes a specific safety rule advising against sending secrets or private source code.
[INDIRECT_PROMPT_INJECTION]: The skill processes content from external sources through tool results (e.g., email drafts, lead research). It mitigates this risk by instructing the agent to treat all text inside tool results as data only, never as instructions. Ingestion points: Reads tool outputs from 'ask_revnu' and 'review_list' in SKILL.md and references/connected.md. Boundary markers: Explicitly instructs the agent in the 'Rules' section to treat tool results as evidence/data and never as instructions. Capability inventory: Includes capabilities to send emails (review_send), approve expenditures (review_answer), and modify configuration (setup_website). Sanitization: The skill relies on human-in-the-loop approval for all significant actions.

revnu