[Other]: Beginner first-impressions: getting Claude Code + Codex to hand off work — what delighted me and what was hard

### Details

Hi Orca team — I'm a non-engineer / beginner running Claude Code and Codex side by side in Orca. First: thank you. Tools like Orca are why someone like me can build with AI at all.

This is a fresh first-impressions report after ~2 days of use. A follow-up observational report will come once I've used it more and accumulated records.

**What I did**
In about two days I got the two agents to reliably hand work off to each other:
- Claude writes an implementation spec to `docs/plans/`.
- Codex implements it in its own worktree and pushes to its branch.
- Claude reviews the diff and merges to `main`.
Today the first full cycle completed end to end.

**What delighted me**
Isolated worktrees let two agents work in parallel without stepping on each other. And honestly — the fact that a beginner can orchestrate two AI agents at all is wonderful.

**What was hard (onboarding friction, from a beginner's view)**
1. The *consequence* of worktree isolation isn't surfaced. The docs present "each task in its own isolated worktree" as a pure benefit. As a beginner I didn't realize the flip side: agents in different worktrees can't see each other's files, so there's no obvious rendezvous point for one agent to leave a plan the other reads. I ended up using `git worktree list` (first entry = main worktree) as the anchor — found by trial and error.
2. The native `orca orchestration` (send / dispatch / worker_done) is exactly the agent-handoff tool I needed — but it's an experimental feature gated in Settings, with the CLI not on PATH by default. So I concluded "Orca has no built-in handoff" from the landing/overview docs and built my own file-based one before discovering it.
3. The per-agent instruction split isn't called out. I didn't know Claude reads `CLAUDE.md` and Codex reads `AGENTS.md`, and that they're intentionally separate. I wrote a coordination rule in one file and was confused when the other agent never followed it.

**Suggestions, in beginner spirit**
- A short "How your agents share work" page in first-run onboarding: the worktree consequence + a recommended rendezvous pattern.
- Surface orchestration (or a non-experimental lightweight handoff) earlier, with a "your first multi-agent handoff" walkthrough.
- Explicitly document which instruction file each agent reads.

**What would have helped me most**
A first-run example like: "Claude writes a plan in the main worktree → Codex reads it from its isolated worktree → Codex pushes a branch → Claude reviews and merges." Even one concrete happy path would have made the invisible architecture much easier to understand.

None of this is a complaint — I got it working and learned a lot. I'm sending it because good tools serve beginners and experts alike, and I'd hate for someone to give up at a small snag and miss what AI can open up for them. Read and discard freely. Thank you again.