The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill involves processing complex data structures such as flow definitions, execution logs, and connector schemas returned from the FlowStudio API. This creates a surface for indirect prompt injection if the remote data contains malicious instructions designed to manipulate the agent's logic.\n
Ingestion points: The core mcp helper functions in SKILL.md retrieve JSON data from https://mcp.flowstudio.app/mcp and read oversized results from local temporary files.\n
Boundary markers: There are no specific boundary markers or 'ignore' instructions defined for the data returned by the API before it is processed by the agent.\n
Capability inventory: The skill enables significant capabilities including listing tenant environments, updating flow definitions via update_live_flow, and triggering flow runs via trigger_live_flow (as detailed in references/tool-reference.md).\n
Sanitization: The skill provides guidance on 'Extract, don't echo' and 'Summarize tool output' in SKILL.md, which helps reduce the data footprint but does not fully sanitize potential injection content.\n- [COMMAND_EXECUTION]: The skill provides functional Python and Node.js code snippets for the agent to execute. These snippets use standard libraries (urllib.request in Python and fetch in Node.js) to perform network operations against the FlowStudio endpoint.\n
Evidence: Code blocks titled 'Core MCP Helper (Python)' and 'Core MCP Helper (Node.js)' in SKILL.md define functions for making POST requests with JSON-RPC payloads.

power-automate-mcp