The Agent Skills Directory

[Broad Command Execution Capabilities]: The skill utilizes system tools such as Bash and Write. These provide the agent with a versatile interface for performing local environment diagnostics and managing data during the troubleshooting process.
[Automated Flow Modification]: The skill is authorized to use mcp__flowagent__edit_flow and mcp__flowagent__update_flow. This enables the agent to autonomously suggest and apply corrections to flow definitions based on its analysis. This is a significant capability that allows for direct changes to logic within the flow service.
[Indirect Prompt Injection Surface]: The diagnostic workflow involves reading execution traces and action details from external flow runs. If these runs contain untrusted data, there is a possibility this content could be interpreted as instructions by the agent.
Ingestion points: Instructions in SKILL.md direct the agent to retrieve data using tools like mcp__flowagent__get_run_actions and mcp__flowagent__get_run_details.
Boundary markers: The current instructions do not implement specific delimiters to isolate potentially untrusted run data from the agent's core logic.
Capability inventory: The skill possesses powerful capabilities, including Bash for command execution and tools for modifying cloud-based flow definitions.
Sanitization: There are no explicit steps defined for sanitizing or validating the data fetched from flow runs before it is processed by the AI model.

diagnose-flow