Prompt: move a Yjs, Trystero or PeerJS app onto ird p2p, no rewrite

Have an agent take a collaborative-editing app that already runs on y-webrtc, Trystero/WebTorrent or PeerJS and point it at an ird p2p app — the same hostname speaks all three signaling dialects, so the change is one URL in the client config, plus TURN so peers behind strict NATs still connect. Origins locked per environment, tested with two browsers, cost known.

Needs: ird access (MCP server or the ird CLI, logged in); the app's source with its provider/joinRoom/Peer setup; two browsers (or one plus a private window) for the test Costs: hourly per app plus per MB relayed through TURN (credits_prices → recurring.p2p-app-hourly: per_hour, per_mb_relayed) and a tiny fee per TURN credential mint (p2p.credentials); signaling traffic is currently free Group: p2p

Paste everything below the line into a fresh conversation. Fill in the FILL IN block first.

Task

Create one ird p2p app per environment below, wire my existing collaborative-editing client to it without changing its data layer, add TURN, and prove two browsers sync through it. Nothing is created before I confirm the price.

FILL IN

NAME:      notes                          # base name → notes.ap2p.ro, notes-staging.ap2p.ro, notes-dev.ap2p.ro
STACK:     y-webrtc                       # y-webrtc | trystero | peerjs
PROD:      https://notes.example.ro       # production origin(s), comma-separated
STAGING:   *.staging.example.ro           # staging origin(s) — bare wildcard patterns allowed
DEV:       http://localhost:5173          # dev origin(s) — try it; if refused, leave blank = any origin
ENVS:      prod, staging, dev             # which apps to create now
ROOM_VAR:  roomName                       # where my code builds the room/topic id (file or variable)

Non-negotiable rules

1. Price before creating. Read credits_prices and show per_hour × 730 per app × number of ENVS, plus the per_mb_relayed rate. Wait for a yes.
2. Origins replace whole. p2p_set_origins overwrites the list; always read it from p2p_list_apps and resend it complete. Wildcards are bare (*.example.ro), never https://*.example.ro. An empty list means any origin — acceptable only on the dev app, and say so.
3. Read p2p_protocol before touching client code. It has the exact compat endpoints, the PeerJS options and the TURN response format. Quote it; do not guess.
4. Do not change my CRDT/data layer. Only the signaling URL, the ICE config and the room id derivation move.

Steps

A. Create. For each ENV: p2p_list_apps — reuse NAME[-env] if it exists. Otherwise p2p_create_app(name="NAME-env", origins=[…]) (CLI: ird p2p create NAME-env --origin … --origin …); prod is plain NAME. Keep signaling_url, turn_credentials_url, ice_urls and the hostname. Confirm each app's allowlist with p2p_list_apps; fix with p2p_set_origins(app, origins=[full list]) (CLI: ird p2p origins NAME --set a,b) — takes effect within a minute.

B. Point the client at the compat endpoint. Per p2p_protocol (also served at https://<host>/.well-known/ap2p/protocol.md):

• y-webrtc: new WebrtcProvider(room, doc, { signaling: ["wss://<host>/y-webrtc"] }) — remove the public demo signaling servers from the list.
• trystero: joinRoom({ appId, trackerUrls: ["wss://<host>/tracker"] }, room) — /tracker speaks the WebTorrent tracker protocol (P2PT and WebTorrent clients work too).
• peerjs: new Peer({ host: "<host>", port: 443, secure: true }) — server key peerjs, path / (PeerJS's default — leave it), ids allocated at GET /peerjs/id.

Read the hostname from an env variable so each build picks its own app. Rooms are namespaced per app, so ROOM_VAR needs no environment prefix.

C. Add TURN. None of these frameworks fetch TURN by themselves. Add a small function that does GET turn_credentials_url from the page (origin-checked, so only from an allowed origin) and builds iceServers from the response's urls, username, credential. Re-fetch a few minutes before ttl expires; on failure fall back to an empty iceServers list (direct connections still work). Pass it as peerOpts: { config: { iceServers } } for y-webrtc, rtcConfig for Trystero, config: { iceServers } for PeerJS — option names per p2p_protocol.

D. Verify with two browsers.

1. GET https://<host>/.well-known/ap2p → the descriptor lists the compat endpoint, max_rooms and max_room_peers.
2. p2p_turn_credentials(app=NAME, ttl_sec=600) (CLI: ird p2p creds NAME --ttl 600) → username/credential; urls include turn: entries.
3. Open the app in two browsers on the same room: an edit in one shows in the other within a second. In chrome://webrtc-internals (or about:webrtc) the selected candidate pair is host/srflx.
4. Force the relay path once: set iceTransportPolicy: "relay" in the same config and repeat — the pair must now be relay, and p2p_usage (CLI: ird p2p usage) shows relayed bytes growing. Remove the override afterwards.
5. Open the app from a non-allowed origin (or curl -H "Origin: https://evil.example.com" turn_credentials_url) — it must be refused.

E. Capacity check. All three dialects share one per-app max_rooms pool (default 500); y-webrtc topics and tracker swarms count peers against max_room_peers (default 16). If a document may have more collaborators at once: p2p_set_limits(app, max_room_peers=N) (CLI: ird p2p limits NAME --set max_room_peers=N).

F. Cost recap. p2p_usage → hourly and monthly cost plus relayed bytes; traffic that goes direct over WebRTC costs nothing.

Deliverable

Per environment: app hostname, signaling_url, turn_credentials_url, ice_urls, the origin allowlist. The client diff (signaling URL, TURN helper, env variable), test results for D.1–D.5, monthly estimate, and the teardown commands (ird p2p delete NAME-dev etc.).