Agent-Setup Diagnostic · Foresttasks

One at a time — I drive a single chat/session.A couple, but I switch between them manually.Several, and a few genuinely overlap.A fleet — many run concurrently as a matter of course.

They don’t — work queues behind whichever one I’m watching.I open more terminals/tabs and keep them apart in my head.I hand-split the work and hope they don’t collide.They pull from a shared queue and claim independent work themselves.

I don’t — I read back through the chat to reconstruct it.Notes, a scratch doc, or my own memory.A general task tracker (Jira/Linear/Notion) I update by hand.A live view that shows every agent’s current task without me logging it.

In chat scrollback — it’s gone when I close the window.Scattered across commits, files, and pasted snippets.A tracker entry I wrote up afterwards from memory.A durable task record the agent itself kept current as it worked.

The agent says it’s done. I mostly take that at face value.I skim the output and eyeball it before accepting.I run tests/checks manually when I remember to.Closure is gated — acceptance criteria must pass before it can close.

No — there’s no record of how/whether it was checked.Maybe, if the chat or PR thread still exists.Roughly — from commit history and my own recollection.Yes — an append-only trail records who verified what, and when.

They borrow my session/credentials — actions look like mine.A shared key or token reused across agents.Per-agent keys, but no real scoping or per-agent audit.Each agent is its own principal — scoped credentials, its own audit trail.

One at a time — I drive a single chat/session.A couple, but I switch between them manually.Several, and a few genuinely overlap.A fleet — many run concurrently as a matter of course.

How well is your agent setup actually working?

How well is your agent setup actually working?