feat(daemon): sprint 4 outbound routing + CLI thin-client + ambient mode
Some checks failed
CI / Lint (push) Has been cancelled
CI / Typecheck (push) Has been cancelled
CI / Broker tests (Postgres) (push) Has been cancelled
CI / Docker build (linux/amd64) (push) Has been cancelled

Daemon outbox now stores resolved target_spec + crypto_box ciphertext
+ nonce per row. Drain worker is a forwarder; no per-row resolution at
drain time. Outbound routing is no longer a placeholder.

Schema additions (additive, NULL allowed for legacy rows): outbox.mesh,
target_spec, nonce, ciphertext, priority. v0.9.0 rows keep draining via
the broadcast fallback so existing in-flight rows finish cleanly.

IPC /v1/send resolves the user-friendly to (display name, hex prefix,
full pubkey, @group, *, #topicId) into a broker-format target_spec at
accept time. DMs encrypt via crypto_box; broadcast/topic/group base64
the plaintext. Hex prefixes (16+ chars) match against connected peers.

CLI thin-client routing extends trySendViaDaemon pattern to peer list
and skill list/get. Three new helpers in services/bridge/daemon-route.ts.

SKILL.md gains ambient mode section: after claudemesh install, raw
claude works for the daemon's attached mesh. Launch stays as the
override path.

Spec at .artifacts/specs/2026-05-04-v2-roadmap-completion.md orders
the remaining v2.0.0 work: multi-mesh daemon (1.26), CLI-to-thin-client
(1.27), mesh-to-workspace rename (1.28), HKDF identity (2.0).

Released as 1.25.0 on npm.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
Alejandro Gutiérrez
2026-05-04 01:36:16 +01:00
parent 6794aa8512
commit 0e3a5babd9
13 changed files with 482 additions and 23 deletions

View File

@@ -45,7 +45,7 @@ claudemesh send "<from_name>" "..." --mesh "<mesh_slug>"
If the parent Claude session was launched via `claudemesh launch`, an MCP push-pipe is running and holds the per-mesh WS connection. CLI invocations dial `~/.claudemesh/sockets/<mesh-slug>.sock` and reuse that warm connection (~200ms total round-trip including Node.js startup). If no push-pipe is running (cron, scripts, hooks fired outside a session), the CLI opens its own WS, which takes ~500-700ms cold. **You don't manage this** — every verb auto-detects and falls through.
### Daemon path (v0.9.0, opt-in, fastest)
### Daemon path (v1.24.0+, REQUIRED for in-Claude-Code use)
`claudemesh daemon up [--mesh <slug>]` starts a persistent per-user runtime that holds the broker WS, a durable SQLite outbox/inbox, and listens on `~/.claudemesh/daemon/daemon.sock` (UDS) plus an optional loopback TCP. When the daemon socket is present, every verb routes through it first (~1ms IPC) before falling back to bridge / cold paths. The send envelope carries a caller-stable `client_message_id`, so a `claudemesh send` that started before a daemon crash survives the restart via the on-disk outbox.
@@ -60,11 +60,15 @@ claudemesh daemon outbox requeue <id> # re-enqueue an aborted/d
claudemesh daemon down # SIGTERM + wait
```
`claudemesh install` (MCP + hooks registration) and the daemon are independent — install does not start the daemon, and the daemon does not require install. Run both for the warmest path: install gives you the in-session push-pipe, daemon gives you cross-invocation persistence and a survivable outbox.
As of 1.24.0 `claudemesh install` registers the MCP entry **and** installs/starts the daemon service for the user's primary mesh. The MCP shim hard-requires the daemon to be running — it bails at boot with actionable instructions if the socket isn't present. There is no fallback. CLI verbs (`send`, `peer list`, `inbox`, `skill list/get`, etc.) keep working without a daemon via bridge or cold paths, but for any in-Claude-Code use the daemon must be up.
### Ambient mode (1.25.0+)
Once `claudemesh install` has run (registers MCP entry + starts daemon service), **raw `claude` Just Works** for the daemon's attached mesh. No `claudemesh launch` ceremony, no manual flags, no per-session keypair. Channel push, slash commands, and resources all flow through the daemon-backed MCP shim. Use `claudemesh launch` only when you need to override defaults (different mesh, custom display name, system-prompt injection, headless modes).
## Spawning new sessions (no wizard)
`claudemesh launch` is the canonical way to start a new Claude Code session connected to claudemesh. Pass every required flag up front so no interactive prompt fires — that's what makes the verb scriptable from tmux send-keys, AppleScript/iTerm spawn helpers, hooks, cron, and the `claudemesh launch` you call from inside another session. **Always use this verb, never `claude` directly with hand-rolled flags** — it sets up the per-session ed25519 keypair, exports `CLAUDEMESH_DISPLAY_NAME`, isolates the mesh config in a tmpdir, and passes the `--dangerously-load-development-channels server:claudemesh` plumbing that the MCP push-pipe needs.
`claudemesh launch` remains useful for non-default cases: explicit mesh selection, fresh display name, headless `--quiet` runs, system-prompt injection, multi-mesh users with one daemon attached to mesh A who want to spawn into mesh B. For the common case (single joined mesh, daemon installed), prefer raw `claude`. Pass every required flag up front so no interactive prompt fires — that's what makes the verb scriptable from tmux send-keys, AppleScript/iTerm spawn helpers, hooks, cron, and the `claudemesh launch` you call from inside another session.
### Full flag surface