I am testing a small local-first pattern for people running Cursor, Claude Code, Codex, or local scripts across multiple projects.
The problem: an agent often stops being useful because it needs the owner to do something external: approve a deploy, export data, grant account access, make a product decision, or check a result after a time window. That action gets buried in terminal output or chat history.
Prototype:
It is intentionally small right now:
- agent writes one JSON status file
- local renderer turns it into a static HTML board
- board shows blocker, exact owner action, recent commits, and next review window
I am not building a Cursor-specific connector until I know the workflow matters. For people using Cursor agents seriously: would you want this as a repo file, a Cursor rule/hook, an MCP server, or something else?