The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides Python and Node.js code snippets designed to be executed by the agent to facilitate communication with the MCP server. These helpers use standard library functions (urllib.request for Python and fetch for Node.js) to perform HTTP POST requests and parse JSON-RPC responses.
[EXTERNAL_DOWNLOADS]: The helper functions connect to a third-party API endpoint (https://mcp.flowstudio.app/mcp). This domain is specific to the service the skill is designed to integrate with. While not on the standard whitelist of services, the communication is transparently documented as the primary purpose of the skill.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external data, creating an attack surface for indirect prompt injection.
Ingestion points: Data enters the agent's context through responses from the mcp.flowstudio.app API and by reading local temporary files (e.g., tool-results/mcp-flowstudio-*.txt) containing oversized tool responses.
Boundary markers: The provided code snippets do not include explicit boundary markers or instructions to the LLM to ignore embedded commands within the tool results.
Capability inventory: The skill enables the agent to read, modify, and trigger Power Automate flows, representing significant capability if influenced by malicious data.
Sanitization: The skill relies on standard JSON parsing (json.loads / JSON.parse) to handle data but does not implement content-level sanitization to prevent the agent from following instructions embedded in flow definitions or error messages.

flowstudio-power-automate-mcp