The Agent Skills Directory

[COMMAND_EXECUTION]: The skill runs orchestration scripts like register.ts and validate-end-to-end.ts using the Bun runtime to manage the flow creation and validation processes.- [DYNAMIC_EXECUTION]: The core functionality involves generating and executing scripts (TypeScript, Go, Python) to simulate system behavior. This is protected by manual user checkpoints and strict path restrictions.- [DATA_EXPOSURE]: The skill reads project source code to identify endpoints and schemas, and accesses ~/.seeflow/config.json to resolve the Studio API endpoint.- [EXTERNAL_DOWNLOADS]: The skill communicates with a local or user-configured SeeFlow Studio API via HTTP and SSE for flow registration and monitoring.- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from prompts and the codebase. Mitigations include: (1) Ingestion points: userPrompt and project files; (2) Boundary markers: isCleanRelativePath restricts script paths; (3) Capability inventory: file writing and script execution; (4) Sanitization: htmlNode data is stripped of dangerous elements like scripts and iframes.

create-seeflow