Hand each agent its own pile: the codebase, the docs, the web through Claude in Chrome. They all read at once and report to one lead session, which writes the brief and notes which helper found each claim.
One session goes through the sources in order, and its context fills with raw notes well before it gets to the writing. When it finally does, the early findings have become a summary of a summary.
Tell the lead what you need to find out and which sources count.
One per source, each in its own terminal and attached to the lead.
A short report with sources from each helper, one message each.
A single brief built from those reports, with the open questions at the end.
Behind it, one helper per source. With --chrome, a Claude helper gets Claude in Chrome for web sources. Read the CLI reference
$ overvibe new --attach --chrome --cwd . --name "research: web" \ --prompt "Find how the three biggest competitors price their API. Cite each page. When done, run: overvibe send \"$OVERVIBE_SESSION\" \"<findings>\"" $ overvibe new --attach --tool codex --cwd . --name "research: code" \ --prompt "List every place our API counts usage. When done, run: overvibe send \"$OVERVIBE_SESSION\" \"<findings>\""
A Claude Code helper started with --chrome can drive Claude in Chrome. Codex and agy bring their own tools.
Run overvibe resume on the helper. It comes back with its conversation, so it picks up where it stopped.
No. Point the agents at a folder of PDFs, notes or exports and each one reads what's in its folder.
Claude drafts the advertorial, Codex goes after the hook and every claim, and you choose which version goes live.
From your phoneEvery Claude session starts with Remote Control on. From the Claude app, ask your coordinator which sessions need you.
Parallel features in worktreesTwo or three helpers build separate features at once, each in its own git worktree and branch, merged when they're done.
Install Overvibe, start a lead session, and let it send helpers out to the sources.
