Docs

Quickstart

Prove the cross-agent handoff in five minutes.

The whole point of uac is that one agent’s knowledge reaches the other. Here’s how to see it, end to end.

Set up a project (once):
```
cd your-project
uac init
```
Then trust the workspace in each agent — see Install.
Let Claude learn something. Run claude, give it a real task, and let it finish. On the way out, uac’s hook nudges it to save anything durable:
```
uac memory list      # did Claude write something itself?
```
Ask Codex about it — without mentioning memory. Run codex and ask a natural question that the saved fact answers. It should surface what Claude learned:

“Avoid npm run lint — ESLint isn’t configured and it hangs a non-interactive session. Use npx tsc --noEmit.”

If Codex repeats back something Claude discovered — and it’s in no file — the handoff works. That’s the tool doing its one job.

Seed a memory by hand

You don’t have to wait for an agent to learn something:

uac memory add "Staging deploys need VPN — the runner has no direct DB access" \
  --kind gotcha --tags deploy,staging

kind is one of decision, fact, gotcha, preference.

Everyday commands

uac memory list                 # what both agents have learned
uac memory search "deploy"      # this project + global + linked projects
uac skills list                 # shared skills
uac sync                        # regenerate CLAUDE.md + .claude/skills/

Run uac <command> --help for the flags.