GeezerVison

Summary

• clarify that read permissions provide read-only file tools and exclude shell
• require write permissions for shell usage or workspace edits
• update README and API design examples to use the top-level agent field
• align action.yml inputs with the API design: replace prompt/tools/skills/mcp with the agent and permissions inputs, and pass them to the binary as ZEROLOOP_AGENT and ZEROLOOP_PERMISSIONS

Why

README and the API design now put the model, runtime, prompt, mcp, and skills under one agent input, but action.yml still declared the old flat inputs, so a workflow written from the docs would not match the action.

Validation

• git diff --check