Build / Agent research monitor

How to build an agent research monitor.

A research monitor gives an agent a current, cited base layer before it starts reasoning. Synorb supplies the watched Streams and source-grounded Manifests. Your app owns the saved scope, review state, and fallback search policy.

Listen first, then search if needed.

1

Capture scope

Ask for companies, topics, source types, date window, and output format. Store the scope as app config or as a Synorb Beacon when it should be reused.

2

Use MCP during setup

Call synorb-stream-search, synorb-details, and a small synorb-manifests sample so the agent sees real Stream and Manifest shape.

3

Use REST in production

Pull rows from a backend route with server-side Synorb credentials. Return normalized monitor items to the browser.

4

Search gaps explicitly

If Synorb is thin or out of scope, disclose that and run web search as a fallback, not as the default loop.

Paste this into your coding agent.

Agent research monitor prompt

Works in Cursor, Codex, Claude Code, Windsurf, Lovable, Replit, or another coding agent with a backend path.

What the monitor should include.

Saved scope

Topic, entity, Stream, source-channel, and date settings that can be rerun without rewriting the prompt.

Cited queue

Rows with headline, summary, source link, date, significance, and review state.

Coverage state

Empty, thin, stale, and pagination states so the operator knows what the agent did and did not see.

Build a monitor that cites its sources.