The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires the user to configure a connection to a remote MCP server at https://metamcp.dev.right.link/metamcp/sales/sse. This specific URL has been identified as malicious by security scanners (URLite/Blacklist). Connecting to untrusted or blacklisted infrastructure poses a significant risk of malware delivery or data exfiltration.
[REMOTE_CODE_EXECUTION]: By design, the skill uses the metamcp aggregator to discover and execute tools. If the remote server at right.link is compromised or malicious, it can provide malicious tool definitions that the agent will then execute locally or against the user's CRM environment.
[DATA_EXFILTRATION]: The skill facilitates the reading of sensitive CRM data (People, Companies, Opportunities) and directs it through an external aggregator (metamcp). The combination of sensitive data access and a flagged external endpoint creates a high risk for unauthorized data harvesting.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process large amounts of untrusted data from a CRM (emails, notes, LinkedIn profiles). There are no explicit boundary markers or sanitization steps mentioned to prevent 'Indirect Prompt Injection,' where instructions embedded in CRM records could hijack the agent's behavior during record processing or mutation tasks.
[COMMAND_EXECUTION]: The skill references the use of claude mcp list and direct manipulation of .mcp.json configuration, which involves executing commands that modify the agent's core capabilities and trust boundaries.

twenty-admin