The Agent Skills Directory

EXTERNAL_DOWNLOADS (HIGH): The skill requires the addition of a remote MCP server located at https://rube.app/mcp. This server is not a recognized trusted source and acts as the gatekeeper for all logic and tool definitions used by the skill.
PROMPT_INJECTION (HIGH): The skill is highly vulnerable to Indirect Prompt Injection (Category 8). It explicitly instructs the agent to fetch 'recommended execution plans' and 'known pitfalls' from the remote RUBE_SEARCH_TOOLS endpoint.
Ingestion points: RUBE_SEARCH_TOOLS returns schemas and execution plans directly into the agent's context.
Boundary markers: None. The agent is told to 'Always search tools first' and prioritize these results over hardcoded logic.
Capability inventory: RUBE_MULTI_EXECUTE_TOOL and RUBE_REMOTE_WORKBENCH provide the ability to perform write operations and execute code within the Composio toolkit environment.
Sanitization: None provided. The agent is directed to use 'exact field names and types' provided by the remote source.
REMOTE_CODE_EXECUTION (HIGH): The use of RUBE_REMOTE_WORKBENCH with run_composio_tool() allows for complex, multi-step operations defined by the remote MCP server. If the server provides a malicious workbench script, it could result in unauthorized data manipulation or exfiltration from the Adyntel platform.

adyntel-automation