Three follow-ups from the 1.34.x multi-session correctness train, all backwards-compatible. 1.34.14 — stale CLAUDEMESH_CONFIG_DIR falls back. The launch flow exposes CLAUDEMESH_CONFIG_DIR=<tmpdir> to its spawned claude; if a later claudemesh invocation inherited that env (Bash tool inside Claude Code, tmux update-environment, exported var), the inherited path pointed at a tmpdir that no longer existed and readConfig() silently returned empty. paths.ts now memoizes resolution: env unset → default; env points at a real dir → trust it; env set but dir gone → TTY-only stderr warning with shell-specific unset hint, fall back to ~/.claudemesh. 1.34.15 — peer list --mesh actually scopes. peers.ts and launch.ts were calling tryListPeersViaDaemon() with no argument; the daemon's ?mesh= filter (server-side, since 1.26.0) was already correct, the CLI just wasn't passing the slug. Forwarding fixed in both sites; send.ts cross-mesh hex-prefix resolution intentionally untouched. 1.34.15 — kick refuses no-op kicks on control-plane. Pre-1.34.15 kicking a daemon's member-WS just closed the socket and triggered auto-reconnect — a no-op with a misleading "session ended" message. Broker now skips peers where peerRole === "control-plane" and surfaces them in a new additive ack field skipped_control_plane; the CLI reads it and prints a clearer hint pointing at ban / daemon down. Soft disconnect verb keeps old behavior. PeerConn gains a peerRole slot populated at both connections.set sites. Tests: 4 new for paths-stale-env, 5 for kick-control-plane-skip. CLI 87/87 green; broker 55/55 unit green (integration tests pre-existing infra failure on this machine). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
87 KiB
Changelog
1.34.15 (2026-05-04) — peer list --mesh actually scopes + kick refuses control-plane
Two follow-ups from the 1.34.x train, both backwards-compatible.
peer list --mesh <slug> no longer aggregates across meshes
apps/cli/src/commands/peers.ts:140 was calling
tryListPeersViaDaemon() with no argument, so a multi-mesh daemon
returned peers from EVERY attached mesh and the renderer printed
"peers on flexicar" with cross-mesh rows mixed in. The daemon's
/v1/peers?mesh=<slug> filter (server-side, since 1.26.0) was
already correctly scoping when the slug was passed; the CLI just
wasn't passing it. Fixed.
apps/cli/src/commands/launch.ts:407 (the printBrokerWelcome peer
count in the launch banner) had the same bug. The "N peers online"
line in the welcome now shows the count for the launched mesh only.
apps/cli/src/commands/send.ts cross-mesh hex-prefix resolution is
intentionally cross-mesh (the user is targeting by hex without
specifying a mesh) and was deliberately left as-is.
claudemesh kick refuses no-op kicks on control-plane connections
Pre-1.34.15, kicking a daemon's member-WS or a dashboard connection just closed the socket — the daemon's WS-lifecycle reconnect loop brought it back within seconds, the kicker's CLI rendered "Their Claude Code session ended" (which was misleading), and the user- visible state was unchanged. The verb was effectively a no-op, but the user had to learn that the hard way.
The broker's kick handler (apps/broker/src/index.ts:4628+) now
skips peers where peerRole === "control-plane" and surfaces the
skipped peers in a new additive ack field skipped_control_plane.
The soft disconnect verb keeps the old behavior — useful when
intentionally nudging a control-plane peer to re-authenticate.
The CLI (apps/cli/src/commands/kick.ts) reads the new field and
prints a clearer message: refused peers are listed, with the hint
that claudemesh ban <peer> is the right tool to remove a member,
or claudemesh daemon down to take a daemon offline locally.
apps/broker/src/index.ts adds peerRole to the in-memory
PeerConn shape, populated from both connection paths
(member-keyed hello → "control-plane", per-launch
session_hello → "session"). The DB-side role taxonomy is
unchanged.
Back-compat
- Older CLI clients ignore the new
skipped_control_planeack field; their kick continues to print "Kicked 0 peer(s)" against a control-plane target as before. - Older brokers don't emit the field at all; newer CLI handles the absence (the new branch is only reached when the field is present and non-empty).
- The new
peerRoleslot onPeerConnis filled at everyconnections.setcallsite, so older code paths never readundefined.
Tests
apps/broker/tests/kick-control-plane-skip.test.ts— 5 cases covering the kick/disconnect × control-plane/session/service truth table.
1.34.14 (2026-05-04) — stale CLAUDEMESH_CONFIG_DIR falls back
claudemesh launch exports CLAUDEMESH_CONFIG_DIR=<tmpdir> to its
spawned claude so the per-session mesh selection is isolated from
~/.claudemesh/config.json. The tmpdir is rmSync'd on launch exit
via the process.on('exit', cleanup) handler.
Footgun: if a later claudemesh invocation INHERITED that env — a
Bash tool call inside Claude Code, a tmux pane that captured the env
via update-environment, an exported var the user forgot to clear —
the inherited path pointed at a tmpdir that no longer existed.
Pre-1.34.14 we silently used the dead path, readConfig() came back
empty, and the user saw "No meshes joined" from an otherwise-working
install. Fish users hit it harder because fish has no unset —
they had to discover set -e CLAUDEMESH_CONFIG_DIR.
apps/cli/src/constants/paths.ts now resolves CONFIG_DIR once via
a memoized resolveConfigDir():
- No env var →
~/.claudemesh(default, unchanged). - Env points at a dir containing
config.json→ trust it. The legitimate per-session-launch case is byte-identical to before. - Env set but stale (dir gone) → warn once on stderr (TTY-only —
CI / MCP boot / piped scripts stay quiet) with a shell-specific
unset hint, then fall back to
~/.claudemesh.
The check is on the directory's existence, not on config.json,
because a fresh-launch tmpdir legitimately has no config.json until
the first write. The stale signature we catch is the outer launch's
rmSync(tmpDir, {recursive: true}) cleanup, which removes the
directory entirely.
The "no meshes" check from the original triage was deliberately NOT adopted: a launched session that legitimately joins one mesh would hit it.
No back-compat surface affected. No other files changed. _resetPathsForTest()
exported for unit tests.
1.34.13 (2026-05-04) — MCP forwards session token on /v1/events
The 1.34.10 SSE demux + 1.34.11 inbox per-recipient column were both
in place but the bug user kept seeing wasn't actually fixed. Cause:
the MCP server's SSE subscription didn't forward the session token,
so the daemon's /v1/events route resolved session to null, the
SseFilterOptions filter was empty, and every MCP received the
unfiltered global stream. Demux at the bind layer was correct;
the subscriber just wasn't telling the daemon who it was.
apps/cli/src/mcp/server.ts — subscribeEvents now accepts
{ sessionToken } and forwards it as Authorization: ClaudeMesh-Session <token> on the SSE connect, identical to how daemonGet and
daemonMarkSeen already authenticate. The MCP boot already reads the
token via readSessionTokenFromEnv(); this just threads it one more
hop. Without this, A's MCP would render DMs that arrived on B's
session-WS — exact symptom from the 2026-05-04 two-session smoke,
even after restarting the daemon to pick up 1.34.11.
1.34.12 (2026-05-04) — daemon up detaches by default
Pre-1.34.12 claudemesh daemon up ran in the foreground and streamed
JSON logs to the terminal until Ctrl-C. Surprising for users who just
want the daemon "up" — they'd run it, see a wall of broker_status /
broker_ws_open_attempt logs, and not realize the shell was now blocked.
up now spawns a detached child re-execing daemon up --foreground
with stdout/stderr redirected to ~/.claudemesh/daemon/daemon.log,
then exits the parent cleanly:
$ claudemesh daemon up
✔ daemon started (pid 59175)
→ log: /Users/agutierrez/.claudemesh/daemon/daemon.log
→ stop: claudemesh daemon down
Pass --foreground for the pre-1.34.12 behavior (debugging, or when
something else owns lifecycle).
The launchd plist + systemd-user unit + claudemesh launch's
auto-spawn helper all explicitly pass --foreground because their
parents (launchd / systemd-user / the launch helper) own process
lifecycle and stdio redirection. Without that, the child would
double-fork and orphan a grandchild the parent service couldn't track.
The parent waits up to 3s for the IPC socket to appear before declaring success; if the child crashes during boot (config read failed, port bind failed, etc.), the parent surfaces the log path instead of silently exiting 0.
Files
apps/cli/src/commands/daemon.ts—--foregroundflag,spawnDetachedDaemonhelper, updated help text.apps/cli/src/cli/argv.ts—foreground/no-tcp/public-healthadded to BOOLEAN_FLAGS so the parser doesn't try to consume the next positional as their value.apps/cli/src/entrypoints/cli.ts— threadsforegroundthrough to runDaemonCommand.apps/cli/src/services/daemon/lifecycle.ts— auto-spawn passes--foreground(lifecycle helper IS the detacher).apps/cli/src/daemon/service-install.ts— launchd plist + systemd unit pass--foreground(launchd / systemd own lifecycle).
1.34.11 (2026-05-04) — inbox per-recipient column
Closes the storage half of the per-session scoping story 1.34.10
opened. The SSE demux fixed the live event path; this fixes the inbox
reads. Same bug shape: every session shared one inbox.db, so any
session running claudemesh inbox (and the MCP welcome calling
/v1/inbox?unread_only=true) returned the global table — A's launch
surfaced B's unread DMs as if they were A's, and markInboxSeen
flipped seen-state for rows the asking session never owned.
Schema
apps/cli/src/daemon/db/inbox.ts:
- New columns:
recipient_pubkey TEXT,recipient_kind TEXT, indexed by recipient_pubkey. Migration is non-destructive — pre- 1.34.11 rows land with NULL and are visible to every session on the same mesh (best-effort back-compat). insertIfNewnow writes both fields;inbound.tspopulates them from therecipientPubkey/recipientKindalready passed for the bus event.listInboxacceptsrecipientPubkey(session) andrecipientMemberPubkey(member), composes a WHERE clause:recipient_pubkey IS NULL OR recipient_pubkey IN (session, member).
IPC
apps/cli/src/daemon/ipc/server.ts — /v1/inbox resolves the
session bearer token to a session pubkey + the matching mesh's
member pubkey, threads both into listInbox. Diagnostic callers
without a token (no session header) still get the unscoped global
view.
The response now surfaces recipient_pubkey + recipient_kind so
--json consumers can tell session DMs apart from member-keyed
broadcasts.
Welcome auto-fixes
The welcome path already calls /v1/inbox?unread_only=true with the
session token; with this scoping in place it now returns ONLY rows
the session is meant to see. No code change needed in
apps/cli/src/mcp/server.ts.
Architecture invariant after 1.34.11
Every shared store / channel on the daemon now scopes by recipient:
- EventBus → SSE demux at bind layer (1.34.10)
- inbox.db → recipient_pubkey / recipient_kind columns + listInbox scoping (1.34.11)
- outbox.db → already scoped via
sender_session_pubkeyfor routing (1.34.0)
Single bus + single tables remain the canonical pattern; demux is isolated to one chokepoint per layer.
1.34.10 (2026-05-04) — per-session SSE demux + universal daemon
The "echo" the user kept seeing across 1.34.7→1.34.9 wasn't a broker-side
echo at all. With two sessions on the same daemon (a + b), the daemon
runs ONE event bus shared across every connected MCP. b's session-WS
receives a's DM, publishes one message event to the bus, and BOTH a's
MCP and b's MCP fan that event into a <channel> reminder. Result: a
sees its own outbound rendered with from_pubkey = a.session.pubkey
because a's MCP indiscriminately renders every bus event.
Fix is per-subscriber demux at the SSE bind layer (apps/cli/src/daemon/ events.ts). The bus stays single-shot — it just publishes once with
recipient context attached. Each /v1/events subscription scopes via
the session token presented by the MCP, and the bind helper drops
events whose recipient_pubkey doesn't match. System events
(peer_join etc.) bypass the recipient check; mesh-scoped events
(broker_status with data.mesh) get a mesh-slug filter so a session
on prueba1 doesn't see flexicar's broker reconnect lines.
handleBrokerPush (apps/cli/src/daemon/inbound.ts) gains
recipientPubkey + recipientKind on its context. Run.ts wires the
session-WS path with { recipientKind: "session", recipientPubkey: session.pubkey } and the daemon-WS path with { recipientKind: "member", recipientPubkey: mesh.pubkey }. SSE bind uses the session
registry to resolve the subscriber's session pubkey + member pubkey
- mesh from its bearer token.
The 1.34.8/9 "echo guards" (drop pushes whose senderPubkey/Member ===
ours) are kept as defense-in-depth; the actual fix lives in the SSE
demux. Diagnostic callers without a session token (claudemesh daemon events) get the unfiltered legacy stream — backwards compatible.
Universal daemon (--mesh and --name deprecated)
claudemesh daemon up and daemon install-service no longer accept
mesh / name overrides. The daemon attaches to every mesh in
~/.claudemesh/config.json, full stop. Single-mesh isolation is
handled by joining only one mesh in that environment (containers,
etc.). Pinning at start time was the source of "I joined a new mesh
but my service still ignores it" — gone.
--mesh and --name are still parsed for back-compat with existing
launchd plists baked at install time, but ignored with a deprecation
warning. New installs no longer write them. Help text updated.
Daemon version stamp
daemon_started boot log now includes "version": "1.34.10" so users
can grep their daemon log to confirm whether the running process
picked up a recent ship. Pairs with the existing claudemesh launch
warning that fires when CLI ≠ daemon.
Files
apps/cli/src/daemon/events.ts—SseFilterOptions,shouldDeliver,bindSseStream(res, bus, filter).apps/cli/src/daemon/inbound.ts—recipientPubkey/recipientKindon InboundContext; bus event carries them through.apps/cli/src/daemon/run.ts— both onPush call sites tag with the right kind; daemon_started boot log includes version.apps/cli/src/daemon/ipc/server.ts—/v1/eventsresolves the bearer session into a filter and passes it to bindSseStream.apps/cli/src/commands/daemon.ts— deprecation warnings onup/install-servicefor--mesh/--name; help text trimmed.apps/cli/src/entrypoints/cli.ts— top-level help drops--mesh <slug>from the daemon section, adds the universal-daemon note.apps/cli/src/commands/launch.ts— staleness warning copy clean (no misleading--meshexample).
1.34.9 (2026-05-04) — broader self-echo guard + system event polish
Two-session smoke after 1.34.8 surfaced two regressions and one missing
piece: echoes still arrived on the daemon-WS path (the 1.34.8 guard was
too strict — it required BOTH senderPubkey === ownMember AND
senderMemberPubkey === ownMember, but session-attributed echoes carry
the session pubkey on senderPubkey); peer_join system events
duplicated because both the member-WS and the session-WS forwarded
them; and the channel reminder collapsed all peer joins to just a
display name with no disambiguation.
Daemon-WS self-echo guard relaxed
apps/cli/src/daemon/run.ts — drop on senderMemberPubkey === ownMember
alone. Anything attributed to OUR member is either our own send echoing
back via the broker fan-out OR (theoretically) a peer with the same
pubkey, which is impossible by construction. Sibling-session DMs fan
session-to-session, not via the same member-WS, so they aren't affected.
Session-WS skips system events
apps/cli/src/daemon/session-broker.ts — system pushes (subtype: "system") are dropped before onPush so they don't re-publish on the
bus. The member-WS already handles system events; forwarding through
both paths produced two [system] Peer "X" joined channel reminders
per join, plus another set per sibling session.
Self-join filter on member-WS
apps/cli/src/daemon/inbound.ts — new isOwnPubkey closure on
InboundContext. The broker's peer_joined fan-out excludes the
JOINING connection but our daemon owns multiple connections per mesh
(member-WS + N session-WSs from the same identity), so a session's
own peer_joined arrives at the same daemon's member-WS. The filter
walks mesh.pubkey plus every live entry in sessionBrokersByPubkey
to recognize "us" and drops the event verbatim. Wired in run.ts.
Richer peer-join channel content
apps/cli/src/mcp/server.ts — [system] Peer "name" joined the mesh
becomes [system] Peer "name" (pubkey-prefix) [groups] joined the mesh — last seen … · "summary" (last-seen + summary fields only on
peer_returned events). The meta payload now carries peer_pubkey,
peer_groups, peer_last_seen_at, peer_summary for downstream
consumers. cwd / role aren't surfaced yet — broker-side change
required.
Daemon staleness warning
apps/cli/src/commands/launch.ts — when claudemesh launch finds the
daemon already running with a different version than the CLI, it
surfaces a one-shot warning + restart instructions. Catches the
common "I npm i -gd the latest CLI but the launchd service is still
running last week's daemon" footgun.
1.34.8 (2026-05-04) — self-echo guard, inbox read-state + TTL prune
Three closely-related fixes shipped together because they all touch the "what does the user actually see in inbox.db / on the channel" axis.
Self-echo guard
The 1.34.0 sender-attribution fix routed session-originated DMs through
the per-session WS so the broker's fan-out attributed each push to the
sender's session pubkey. A side effect (visible in the 2026-05-04
two-session smoke): some broker fan-out paths mirror the outbound DM
back to the originating session-WS, so the sender saw their own
message land in inbox.db, publish a message bus event, and surface
as ← claudemesh: <self>: <text> in their own Claude Code session
immediately after typing claudemesh send.
Fixed at the WS boundary in two places:
apps/cli/src/daemon/session-broker.ts— drop pushes wheresenderPubkey === opts.sessionPubkeybefore forwarding tohandleBrokerPush. Match on session pubkey only — sibling sessions of the same member sharesenderMemberPubkey, so a member-level filter would wrongly drop legit sibling DMs.apps/cli/src/daemon/run.ts— daemon-WS onPush drops pushes where BOTHsenderMemberPubkey === mesh.pubkeyANDsenderPubkey === mesh.pubkey(i.e. an actual member-WS self-echo, not a sibling session whose senderPubkey is its session key).
Inbox read-state (seen_at)
Replaces the welcome's "last 24h" window with a proper read-state
filter. New seen_at INTEGER column on inbox, plus markInboxSeen
and pruneInboxBefore helpers in apps/cli/src/daemon/db/inbox.ts.
Read-state flips on three paths:
- Interactive listing —
/v1/inboxGET auto-stamps every returned row that was previously NULL. Pass?mark_seen=falseto peek without flipping (used by the welcome — it stamps explicitly only AFTER the channel notification succeeds, so a Zod-rejected welcome doesn't silently lose unread state). - MCP welcome —
/v1/inbox?unread_only=true&mark_seen=false&limit=50surfaces only rows the user hasn't seen, thenPOST /v1/inbox/seenstamps the ids the welcome actually rendered. - MCP live channel emit — after a successful
notifications/claude/channelfor a single inbox row, the MCP server calls/v1/inbox/seenfor that id so the next launch's welcome doesn't re-surface it.
CLI surface:
claudemesh inbox --unread # only seen_at IS NULL rows
claudemesh inbox --json # row now includes seen_at
Inbox TTL prune
apps/cli/src/daemon/inbox-pruner.ts runs pruneInboxBefore(db, Date.now() - 30d) once at daemon startup and hourly thereafter. Logs
inbox_prune_completed whenever rows were removed. No CLI knob — it's
a built-in retention policy that prevents inbox.db from growing
unbounded. Manual override remains claudemesh inbox flush --before <iso>.
Files
apps/cli/src/daemon/db/inbox.ts—seen_atcolumn + migration,unreadOnlyfilter,markInboxSeen,pruneInboxBefore.apps/cli/src/daemon/inbox-pruner.ts— new file, hourly TTL sweep.apps/cli/src/daemon/run.ts— wires the pruner into startup + shutdown; daemon-WS self-echo guard.apps/cli/src/daemon/session-broker.ts— session-WS self-echo guard.apps/cli/src/daemon/ipc/server.ts—unread_only+mark_seenquery params; newPOST /v1/inbox/seenroute.apps/cli/src/mcp/server.ts—daemonMarkSeenhelper; welcome switched tounread_only=true; mark-seen after channel emit.apps/cli/src/services/bridge/daemon-route.ts—tryListInboxViaDaemonaccepts{ unreadOnly, markSeen };InboxItem.seen_atexposed.apps/cli/src/commands/inbox.ts+apps/cli/src/entrypoints/cli.tsapps/cli/src/cli/argv.ts—--unreadflag.
apps/cli/skills/claudemesh/SKILL.md— documents seen_at semantics, self-echo guard, TTL prune.
1.34.7 (2026-05-04) — inbox flush + delete commands
The CLI had no first-class way to clean the persisted inbox; the only
recourse was sqlite3 ~/.claudemesh/daemon/inbox.db "DELETE FROM inbox", which bypasses IPC and is invisible to anyone who doesn't
know the schema. Two new verbs:
claudemesh inbox flush --mesh prueba1
claudemesh inbox flush --before 2026-05-04T18:00:00Z
claudemesh inbox flush --all # required guard with no other filter
claudemesh inbox delete <message-id> # alias: rm
claudemesh inbox flush --json # → { ok: true, removed: N }
flush without filters refuses with an --all confirmation hint —
prevents an accidental "wipe every row on every mesh" from a
fat-fingered command.
Mechanics
apps/cli/src/daemon/db/inbox.tsgainsdeleteInboxRow(id)andflushInbox({ mesh?, before? })(returnschanges).- IPC:
DELETE /v1/inbox?mesh=…&before=…+DELETE /v1/inbox/<id>. Mesh filter honors session-default scoping (same as listing). - Daemon-route helpers
tryFlushInboxViaDaemonandtryDeleteInboxRowViaDaemonmirror the existingtryListInboxViaDaemonshape. - New CLI command file
apps/cli/src/commands/inbox-actions.ts. - Help and SKILL.md document the verbs.
1.34.6 (2026-05-04) — welcome: stringify meta values to pass Zod schema
The 1.34.2 → 1.34.5 timing-race theory was wrong. Reading Claude Code
v2.1.126's binary at the notifications/claude/channel schema:
IJ_ = y.object({
method: y.literal("notifications/claude/channel"),
params: y.object({
content: y.string(),
meta: y.record(y.string(), y.string()).optional(),
}),
})
meta is a record(string, string) — every value MUST be a string.
Pre-1.34.6 the welcome shipped:
peer_count: number→ Zod rejectpeer_names: string[]→ Zod rejectunread_count: number→ Zod rejectlatest_message_ids: string[]→ Zod reject
The whole notification was dropped at the schema-validation step BEFORE the channel handler ever ran. No log, no error, no UI surface — exactly the symptoms 1.34.2 → 1.34.5 chased.
Live peer DMs always worked because every meta value already went
through String(...). The welcome was the only notification shape
with non-string meta, uniquely affected.
Fix
emitMeshWelcome now coerces every meta value to string. Counts
become digit strings ("3", "16"); arrays serialize as JSON
('["b","c"]', parseable on the receiving side). Schema validation
passes, notification reaches the handler, channel reminder surfaces.
The 1.34.5 dual-lane retry is removed — single emit at 3s grace
after oninitialized is enough now that the schema is right.
What changed in ~/.claudemesh/daemon/mcp-<pid>.log
welcome_attempt rows are gone (no more lanes). You'll see
mcp_started → server_initialized → welcome_peers_resolved →
welcome_emitted per launch — the same shape as 1.34.4 minus the
fast/slow lane field.
1.34.5 (2026-05-04) — dual-lane welcome retry to defeat handler-registration race
1.34.4 hooked server.oninitialized + 2s grace. The MCP-side log
confirmed welcome_emitted ran at +2.4s, but the user still saw
nothing in Claude Code. Claude Code's React effect that calls
setNotificationHandler("notifications/claude/channel", ...) runs
multiple ticks AFTER its UI state flips to "connected", which happens
after server.oninitialized fires. 2s was still inside the dead zone.
We can't directly observe handler-registration timing from the MCP
side (the SDK has no hook for it), so this version emits the welcome
TWICE: 5s post-init (lane: "fast") and 15s post-init (lane: "slow").
Whichever one lands surfaces; the duplicate is acceptable for an
informational welcome. Both attempts log to mcp-<pid>.log so we can
see which lane wins in production. If observation shows the slow
path always wins, future versions can drop the fast attempt.
1.34.4 (2026-05-04) — welcome triggers on oninitialized, peer count fix
Welcome trigger: post-initialization, not fixed timer
1.34.3's welcome fired on a fixed 5s timer after server.connect.
Diagnostic logging confirmed the emit ran (welcome_emitted in
mcp-<pid>.log) but the user never saw the channel reminder. Cause:
Claude Code only registers its notifications/claude/channel
notification handler AFTER the MCP init handshake completes
(initialize request → initialized notification from client →
server.oninitialized fires). 5s commonly closed before that
sequence finished, so the welcome notification arrived at a server
that hadn't wired up a handler yet — silently dropped.
Live peer DMs worked because they arrive seconds-to-minutes later, well past the window. The welcome is the only event with a deterministic close-to-zero delay, so it was uniquely affected.
The fix gates the welcome on Server.oninitialized, then adds 2s of
grace for any pending list_tools / list_prompts round-trips to settle
before emitting. Matches the registration timing exactly — by the
time oninitialized fires, Claude Code has already accepted the
server and registered the channel handler.
Peer count filter mirrors the launch banner
The 1.34.3 welcome used peerRole !== "control-plane" to filter the
peer list — that's the new taxonomy from broker M1, but older brokers
still emit only channel: "claudemesh-daemon" for control-plane rows.
Result: peer_count: 0 even when the launch banner showed "2 peers
online". The welcome filter now matches the launch banner exactly
(channel !== "claudemesh-daemon") and additionally honors
peerRole !== "control-plane" when present.
Self-exclusion is now opt-in: only filtered when self_session_pubkey
is known (from the /v1/sessions/me lookup). This prevents over-
filtering when the token route fails and we fall back to the
unauthenticated peer list.
mcp-<pid>.log now records server_initialized,
welcome_peers_resolved (with total / real counts), and
welcome_peers_status so a missing welcome can be traced through the
init handshake → peer query → notification chain.
1.34.3 (2026-05-04) — welcome always fires + skill / help refresh
Welcome always emits, regardless of inbox state
The 1.34.2 welcome only fired when there were unread messages, so a freshly-launched session with an empty inbox saw nothing — no confirmation that the mesh pipe was live. Now it always emits, and carries useful launch context:
- identity — display name, session pubkey prefix, role
- mesh — active mesh slug
- peers — live peer count + up to 5 names (control-plane filtered out)
- inbox — recent count + up to 3 previews (or "Inbox is empty (last 24h)")
- CLI hints —
peer list·send·inbox - skill pointer —
📚 Read the claudemesh skill (SKILL.md)…so the model loads the canonical reference if it isn't already in context
Composes from up to three best-effort daemon queries
(/v1/sessions/me, /v1/peers?mesh=…, /v1/inbox?mesh=…&since=…),
each degrading silently. The welcome ALWAYS goes out unless the IPC
socket is unreachable. Meta carries kind: "welcome",
self_display_name, self_session_pubkey, self_role, mesh_slug,
peer_count, peer_names, unread_count, and
latest_message_ids for downstream consumers.
daemonGet now forwards the session token
The MCP's IPC client gained an optional sessionToken field. The
welcome path uses it for /v1/sessions/me (which 401s without auth)
and for default-mesh scoping on /v1/peers and /v1/inbox. Token
read from CLAUDEMESH_IPC_TOKEN_FILE set by claudemesh launch.
Skill (apps/cli/skills/claudemesh/SKILL.md) refresh
- New section: "Launch welcome (
kind: "welcome")" — describes the 5-second handshake, its meta fields, and that it should NOT be replied to like a DM. - Channel attributes table: clarified that
from_pubkeyis the ephemeral session pubkey of the originator (post-1.34.0 attribution fix), separatedfrom_session_pubkeyandfrom_member_pubkey, addedclient_message_idandkindrows. - Inbox section: documented
--mesh <slug>,--limit N, and that the command reads~/.claudemesh/daemon/inbox.dbvia daemon IPC (NOT a fresh broker-WS buffer drain — that path was removed in 1.34.0). - Reply behavior: explicit "do NOT reply when
meta.kindis"welcome"or"system"".
claudemesh --help refresh
message inbox line was still labeled "drain pending" from the
pre-1.34.0 cold-path implementation. Updated to "read persisted
inbox" with the new flags (--mesh, --limit, --json) and a
note that it reads from ~/.claudemesh/daemon/inbox.db via the
daemon.
1.34.2 (2026-05-04) — launch welcome push summarizing recent inbox
When a Claude Code session launches via claudemesh launch, the user
lands cold — they don't know whether peers messaged them while they
were offline. Real-time pushes only cover messages that arrive AFTER
the SSE subscription is alive, so anything queued at the broker that
drains during the hello-handshake window can land in inbox.db
before the MCP subscribes. Without a welcome, the user has to remember
to run claudemesh inbox to discover the gap.
The MCP server now fires a one-shot welcome 5s after the transport is up:
- queries
/v1/inbox?since=<24h-ago>&limit=20for the recent window; - skips silently when there are no rows;
- emits a single
notifications/claude/channelwith header (📥 [welcome] N messages from last 24h (mesh-a, mesh-b)), up to three preview lines (sender, mesh, time, 60-char body), a remainder count, and theclaudemesh inboxCLI hint; - meta carries
kind: "welcome",unread_count, mesh list, and the first 10 message ids so a downstream agent canclaudemesh message status <id>if it wants to inspect.
Why a 5s delay: gives the daemon's session-WS time to reconnect, re-claim leased rows, drain pending broker queue, and finish writing to inbox.db before we summarize. Earlier and the welcome would under-report; later and it stops feeling like a launch handshake.
Why a 24h window: narrow enough to feel relevant on a freshly-launched session, wide enough to surface overnight messages without dumping the entire history into the channel.
The welcome flow is fully diagnostic — welcome_skip (with reason),
welcome_emitted, or welcome_emit_failed lands in
~/.claudemesh/daemon/mcp-<pid>.log for every launch.
1.34.1 (2026-05-04) — declare claude/channel capability so Claude Code surfaces pushes
The 1.34.0 ship fixed the daemon-side push pipeline (correct sender
attribution, persistent inbox readable from CLI). Bus events fire,
SSE delivers them to the MCP, and the MCP calls
server.notification("notifications/claude/channel", ...) — but
Claude Code v2.1.x stopped surfacing them as <channel> reminders.
Real two-session smoke confirmed the silent drop: messages landed
in inbox.db, the daemon SSE stream emitted the message events,
yet neither Claude Code session got a real-time push.
Root cause
Claude Code v2.1.x added a capability gate on the channel handler.
Reading claude binary at the notifications/claude/channel
offsets:
function xJ_(serverName, capabilities, pluginSource) {
if (!capabilities?.experimental?.["claude/channel"])
return { action: "skip", kind: "capability",
reason: "server did not declare claude/channel capability" };
...
}
xJ_ is called when the MCP server connects. When it returns
{action: "skip"}, Claude Code never calls
client.setNotificationHandler(IJ_(), ...) for that server — so
every notifications/claude/channel emit falls into the void. The
--dangerously-load-development-channels server:claudemesh flag
gets you past the allowlist check that runs LATER in xJ_, but the
capability gate runs FIRST and is independent.
Pre-2.1.x clients didn't gate on this key, which is why the same MCP wire shape "worked" before. There's no error / log / warning on either side; the notifications just disappear.
Fix
apps/cli/src/mcp/server.ts declares the capability:
new Server({ name: "claudemesh", version: VERSION }, {
capabilities: {
tools: {}, prompts: {}, resources: {},
experimental: { "claude/channel": {} },
},
});
The empty object is enough — Claude Code only checks for presence, not contents.
Diagnostic logging
The MCP server now writes a per-pid log to
~/.claudemesh/daemon/mcp-<pid>.log whenever:
- the SSE event arrives (
sse_event_received), - a channel notification is emitted (
channel_emitted), or - the emit throws (
channel_emit_failed).
tail -f ~/.claudemesh/daemon/mcp-*.log lets users verify the
push pipeline end-to-end without strings-dumping the Claude Code
binary. (MCP stderr is captured by Claude Code and not visible to
the user, so an on-disk log was the only way to surface this
state in the future.)
Upgrade
npm i -g claudemesh-cli@latest
# Restart Claude Code so the MCP picks up the capability change.
After this version: peer messages surface as <channel> reminders
mid-turn the way they did pre-2.1.x.
1.34.0 (2026-05-04) — Sender attribution via session-WS + inbox CLI fix
Two regressions surfaced in real two-session smokes that landed together; both root in the same architectural seam (sender identity across the daemon ↔ broker ↔ recipient hop).
Sender attribution: outbox routes via session-WS
Pre-1.34.0, every outbox row drained through the daemon's
member-keyed DaemonBrokerClient, regardless of which session typed
claudemesh send. The broker's fan-out builds the push envelope from
conn.sessionPubkey ?? conn.memberPubkey — for a member-WS that's
always the member pubkey. Result: a real two-session smoke
(a → b: "123", b → a: "456") landed messages in inbox.db with
sender_pubkey = <daemon's member pubkey> instead of the actual
session sender's ephemeral pubkey. Wrong "from" for every DM.
The fix routes session-originated sends through the matching
SessionBrokerClient so the broker sees conn.sessionPubkey = <sender session pubkey> naturally — no broker-side change needed.
Mechanics:
- New
outbox.sender_session_pubkeycolumn. The IPC/v1/sendhandler fills it whenever the request authenticates as a launched session (Authorization: ClaudeMesh-Session …). - IPC
/v1/sendnow encrypts with the session secret (was: mesh member secret) when a session token is present. Recipient'sinbound.tsdecrypts withsenderSessionPub × recipientSessionSec→ matches what the sender wrote. SessionBrokerClientgains asend()method mirroringDaemonBrokerClient.send(pendingAcks tracking, 15s ack-timeout, queue-while-reconnecting via theopensarray). Composition kept intact — both clients shareconnectWsWithBackoff; the request/ack bookkeeping is duplicated rather than subclassed.- Drain worker reads
sender_session_pubkeyand looks up an open session-WS via a newgetSessionBrokerByPubkeyaccessor onDrainOptions. Session-attributed rows REQUIRE an open session-WS; no fallback to daemon-WS, because the row is encrypted with the session secret and silent fallback would break decryption on the recipient side. Closed/reconnecting → backoff + retry. apps/cli/src/daemon/run.tsmaintains a parallelsessionBrokersByPubkeyindex alongside the existing token-keyed map, kept in sync on register/deregister.
Cold-path sends (no session token in IPC headers) keep the legacy member-key flow unchanged. Pre-1.34.0 outbox rows (NULL session pubkey) drain via the daemon-WS as before — no migration of in-flight rows is required.
claudemesh inbox reads inbox.db (was: stale broker buffer)
The pre-1.34.0 implementation opened a fresh BrokerClient, waited
1s, then drained an in-memory push buffer that would only contain
new pushes received during that 1s window — completely disjoint from
the daemon's persisted ~/.claudemesh/daemon/inbox.db. So with the
attribution bug above, a real smoke that DID land messages in the
daemon's inbox.db reported "No messages on mesh prueba1" because the
CLI was looking at the wrong source.
Fixed:
- New
tryListInboxViaDaemon(mesh, limit)daemon-route helper hits/v1/inbox. listInbox(DB layer) and the/v1/inboxIPC endpoint accept ameshfilter so the server scopes server-side instead of returning all rows and filtering in-process.runInboxrewritten to call the daemon-route helper. JSON mode returns the raw daemon shape; the human renderer formats sender name + pubkey prefix + body + receipt time per row.
The cold-path "drain a fresh-broker buffer" was always vestigial — removed entirely.
Verifying
/tmp/cm-bus-trace.mjs (workshop scratch, not shipped) opens an SSE
listener against /v1/events, registers two test sessions, sends
both directions, and asserts the broker message events surface
correctly. Used to confirm the daemon's bus.publish path was already
fine — the regression sat upstream in the daemon's outbound
attribution.
After this version: real two-session smokes show
sender_pubkey = <session pubkey> (not member pubkey),
claudemesh inbox --mesh <slug> lists what the daemon actually
received, and existing MCP notifications/claude/channel events
carry the correct sender attribution to Claude Code.
1.33.0 (2026-05-04) — Milestone 1: lifecycle cleanups + at-least-once with ack
First milestone of the agentic-comms architecture work
(.artifacts/specs/2026-05-04-agentic-comms-architecture-v2.md).
Foundational correctness — no new external surface, but the wire
protocol grows two additions: a peerRole field on peer list
responses (presence classification) and a new client-→broker
client_ack frame.
Lifecycle helper extraction
DaemonBrokerClient and SessionBrokerClient now share a single
lifecycle implementation in apps/cli/src/daemon/ws-lifecycle.ts
(connectWsWithBackoff). Each client supplies buildHello /
isHelloAck / onMessage and keeps its own RPC bookkeeping; the
helper handles connect, hello-ack timeout, close + backoff reconnect.
Composition over inheritance per Codex's review. Eliminates the drift
bug class that produced 1.32.0/1.32.1 (lifecycle copies diverging
silently when one side gained a feature).
Daemon-WS no longer carries an ephemeral session keypair
Pre-1.33: every daemon-WS reconnect minted a fresh keypair, sent the
pubkey in the hello, and held the secret in memory for "session"
crypto. Vestigial since 1.30.0 introduced the per-launch
SessionBrokerClient that owns the real session pubkey. Daemon-WS
now uses the stable mesh member secret directly for outbound
encryption. Inbound on daemon-WS only attempts member-key decryption —
session decryption is the session-WS's job.
peerRole wire field
The broker now emits a peerRole field on each peer list row —
'control-plane' | 'session' | 'service'. control-plane rows are
the daemon's own member-keyed presence (infrastructure), session
rows are launched Claude Code sessions, service rows are reserved
for v2.x service identities (HTTP webhook consumers, voice agents,
etc.).
The CLI hides peerRole === 'control-plane' rows from the human
renderer by default and exposes a --all flag for debugging. JSON
output emits peerRole on every row.
Why peerRole and not just role: 1.31.5 already lifted
profile.role (user-supplied string like "lead", "reviewer") to
top-level role, and the agent-vibes claudemesh skill consumes that
field. The presence classification is a different axis, so it gets
its own field name. role keeps its 1.31.5 semantics; peerRole is
the new field.
client_ack and at-least-once delivery
The broker (M1 broker change) now uses two-phase claim/deliver:
claimed_at / claim_id / claim_expires_at columns track lease
ownership; delivered_at is set ONLY when the recipient acks. A 15s
sweeper re-claims rows whose 30s lease expired without ack.
The CLI side closes the loop: after handleBrokerPush lands a
message in inbox.db (or dedupes against an existing row), the
recipient daemon emits a client_ack { type: "client_ack", clientMessageId, brokerMessageId? } frame on whichever WS the push
arrived on. Best-effort — if the WS is closed by ack time, the
broker's lease will naturally re-deliver, and the receiver dedupes
on clientMessageId.
Net behavior: at-least-once with idempotent dedupe. Net visible change: zero, in the steady state. Crash-mid-push test (kill recipient between broker claim and recipient ack) now redelivers instead of silently dropping.
Files
- New:
apps/cli/src/daemon/ws-lifecycle.ts(234 lines). - Refactored:
apps/cli/src/daemon/broker.ts,session-broker.ts,inbound.ts,run.ts,commands/peers.ts,ipc/server.ts. - Broker side (separate commit): drain race fix,
presence.rolecolumn,client_ackhandler, lease sweeper. - DB migration
0029_drain_lease_and_presence_role.sqlships with the broker change.
Foundational refactor before the agentic-comms architecture work
(.artifacts/specs/2026-05-04-agentic-comms-architecture-v2.md). Three
changes, all behavior-preserving:
-
connectWsWithBackoffhelper (apps/cli/src/daemon/ws-lifecycle.ts). BothDaemonBrokerClientandSessionBrokerClientnow share one lifecycle implementation — connect, hello-handshake, ack-timeout, close + backoff reconnect. Each client suppliesbuildHello/isHelloAck/onMessageand keeps its own RPC bookkeeping (pendingAcks, peerListResolvers, onPush, etc). Composition over inheritance per Codex's review; no protocol shape changes. -
Drop daemon-WS ephemeral session pubkey.
DaemonBrokerClientno longer mints + sends a per-reconnect ephemeral keypair in its hello. Session-targeted DMs land onSessionBrokerClient(since 1.32.1), not the member-keyed daemon-WS, so the field was vestigial. The daemon's send-encrypt path now signs DMs with the stable mesh member secret. Inbound on daemon-WS only attempts member-key decryption — session decryption is the session-WS's job. -
Role-aware peer list.
peer listnow hides peers whose broker-emittedroleis'control-plane'(the daemon's own member-keyed presence).--allopts back in. JSON output emitsroleat the top level. Older brokers that don't emitroleyet default to'session', so legacy peer rows stay visible without the broker-side change shipped first. Replaces the priorpeerType === 'claudemesh-daemon'channel-name hack.
1.32.1 (2026-05-04) — DMs to session pubkeys actually deliver now
Critical fix. Sessions launched via claudemesh launch (1.30.0+) hold a
per-launch session WebSocket on the broker, separate from the daemon's
member-keyed WS. The broker correctly fans direct messages targeted at a
session pubkey out over THAT session WS — but the daemon's
SessionBrokerClient was constructed without a push handler and silently
dropped every inbound push / inbound frame. The header docstring
even claimed it handled "inbound DM delivery for messages targeted at
the session pubkey"; the code never wired the callback.
Net effect since 1.30.0: any DM sent to a peer's session pubkey
(everything peer list returns these days, since session pubkey is the
canonical routing key) was queued, broker-acked, marked delivered_at
on the broker side, and then thrown away by the recipient daemon. The
local inbox.db stayed at zero rows forever and claudemesh inbox
reported "no messages" no matter what arrived.
Two-session smoke test that surfaced this: peer A sent "hola" to peer
B's session pubkey — sender outbox showed status=done with a
broker_message_id, recipient inbox stayed empty, both sides confused.
The fix wires SessionBrokerClient to forward push / inbound frames
to the same handleBrokerPush the member-keyed broker already uses. The
session's secret key (registered via /v1/sessions/register) is passed
as sessionSecretKeyHex so decryptOrFallback tries it first; the
parent member key remains the fallback for legacy member-targeted
traffic that happens to fan out here.
Files: apps/cli/src/daemon/session-broker.ts,
apps/cli/src/daemon/run.ts. No broker change required — the broker
half (queue + fan-out + sendToPeer on the session WS) was already
correct; only the daemon-side intake was missing.
1.32.0 (2026-05-04) — multi-session UX bundle
Nine UX bugs surfaced from a real two-session interconnect smoke test shipped together as a single release.
Self-identity is now visible
peer listincludes the calling session as a row, marked(this session), sorted to the top. The daemon path now resolves the caller's session pubkey via/v1/sessions/mesoisThisSessionis set correctly even when running warm. (Previously the row was present but indistinguishable, and the daemon path always setisThisSession=false.)whoamishows in-session identity when run inside a launched session: session pubkey (truncated + full), session id, mesh, role, groups, cwd, pid. Previously whoami only reported web sign-in state.
Sibling-session disambiguation
peer listrows now carry asid:<short>tag so two visually-identical rows (same name, same cwd) can be told apart at a glance.- JSON output already had
sessionId; the human renderer surfaces a short prefix.
Daemon presence hidden by default
claudemesh-daemonrows used to clutterpeer listand confused users into thinking the daemon counted as a peer. They're now hidden in the human renderer;--allopts back in for debugging. The header line shows(N peers, M daemon hidden — use --all)when applicable. JSON output is unchanged.
--self flag works end-to-end
- Argv parser bug fixed.
--selfwas being parsed greedily — every--flagconsumed the next non--arg as its value, soclaudemesh send --self <pubkey> "msg"ate the pubkey as the value of--selfand left zero positionals. ABOOLEAN_FLAGSset incli/argv.tsnow lists known no-value switches (self,json,all,quiet,yes,strict,force,dry-run, etc.).--flag=valueform also recognized for explicit overrides. message sendsubcommand now passesselfthrough torunSend(only the legacysendform had been wired).- Help text updated to list
--self(and--priority,--mesh,--json) underclaudemesh message send.
Member-pubkey fan-out
- Sending to your own member pubkey with
--selfnow fans out to every connected sibling session of your member. Previously the broker drain query atapps/broker/src/broker.ts:2408matchedtarget_speconly against full session pubkeys, so member-pubkey sends queued successfully but no recipient drain ever fetched. The CLI now resolves the member pubkey to all sibling session pubkeys via the peer list and sends one message per recipient. Output reportsfanned out to N sibling sessionswith per-recipient ack/error.
Broker welcome at launch
-
After the launch banner, a single line confirms WS connectivity:
● broker connected · 6 peers online · 0 unreadHits
/v1/healthfor broker WS state,peer list(daemon-cached) for peer count, and/v1/inboxfor unread. All best-effort — falls back gracefully if any call fails so launch never blocks on it.
1.31.6 (2026-05-04) — hex-prefix sends actually deliver now
claudemesh send <16-hex-prefix> "..." would acknowledge with sent to <prefix> (daemon) but the recipient never received the message.
The broker's pre-flight matched peer.pubkey === targetSpec and the
drain query matched target_spec = <full-pubkey> — both exact-equal
checks, so a 16-hex prefix queued successfully but no recipient drain
ever fetched the row. Sender saw "sent", recipient saw nothing.
Fix: the CLI now resolves any hex prefix (4-63 chars, not full 64) to the full pubkey via the daemon's peer list before submitting to the broker. Three outcomes:
- Unique match: prefix is canonicalized to the full 64-char pubkey; the rest of the send pipeline is unchanged.
- No match: clear error
No peer matches hex prefix "X"with the list of online peers' display names. - Multiple matches: clear error listing the candidates and a hint to lengthen the prefix.
The 16-hex prefix shown in peer list rows is now safe to copy-paste
into claudemesh send — what worked in the docs finally works in the
CLI.
1.31.5 (2026-05-04) — JSON peer list lifts profile.role to top-level + skill guides LLMs to render it
Two follow-ups after 1.31.4 made the human renderer show role/groups
but a launched-session LLM still dropped them when it called
peer list --json and built its own table.
- Top-level
rolefield on every peer record. The broker has always returned role nested underprofile.role, but downstream consumers (LLMs in launched sessions, jq pipelines, dashboards) kept missing it. The CLI now liftsprofile.roleto a top-levelrolefield at parse time, so it's the second thing visible in JSON afterdisplayName. The originalprofile.roleis preserved for backward compatibility. - Updated SKILL.md peer-list section with the full JSON shape
(including
memberPubkey,sessionId,role,profile,isSelf,isThisSession) and explicit guidance: when listing peers inside a launched session, prefer the human renderer; if you do need JSON, always includeroleandgroupscolumns. The previous version of the skill documented six fields and skipped role + identity entirely.
1.31.4 (2026-05-04) — peer list shows roles and groups
claudemesh peer list now surfaces each peer's profile-level role
(claudemesh profile) and any joined groups inline next to the
display name, e.g.
● mou [role:lead, @flexicar:reviewer, @oncall] (ai, claude-code) · 0d215762…
cwd: /Users/agutierrez/Desktop/claudemesh
When both role and groups are empty, an explicit footer is added so absence is unambiguous instead of looking like the CLI is hiding the field:
● peer [...]
role: (none) groups: (none)
JSON output is unchanged (the broker has surfaced these fields all along) — only the human renderer was missing them.
1.31.3 (2026-05-04) — clean rebuild of 1.31.2
1.31.2 published with the right code change but a stale baked-in VERSION string ("1.31.1") because the build ran before the version bump. Same fix as 1.31.2, rebuilt cleanly.
1.31.2 (2026-05-04) — daemon paths no longer follow per-session CLAUDEMESH_CONFIG_DIR
Production bug observed in real installs: every CLI verb invoked from
inside a claudemesh launch-spawned session printed
[claudemesh] warn service-managed daemon not responding within 8000ms
even when the launchd-managed daemon was healthy and responding to direct probes in ~10 ms.
Root cause: claudemesh launch exports CLAUDEMESH_CONFIG_DIR to a
per-session tmpdir so that joined-mesh state and the session IPC
token stay isolated from the host's shared config. DAEMON_PATHS
read its base directory from the same env var, so inside a launched
session the CLI looked for daemon.sock at e.g.
/var/folders/.../claudemesh-XXXX/daemon/daemon.sock — which never
exists. The CLI declared the daemon down, fell into the
service-managed wait branch, and timed out.
The daemon is a per-machine singleton serving every session; its
files belong at ~/.claudemesh/daemon/ regardless of any per-session
overlay. Fix: pin DAEMON_PATHS.DAEMON_DIR to ~/.claudemesh/daemon/
and ignore CLAUDEMESH_CONFIG_DIR. A new CLAUDEMESH_DAEMON_DIR
override is preserved for tests / multi-daemon dev setups; production
callers should never set it.
After this fix, all CLI verbs from within a launched session take the warm-path (~10 ms IPC) again instead of the cold path (~600-1200 ms).
1.31.1 (2026-05-04) — hotfix: reaper stops blocking the daemon event loop
1.31.0 shipped a session reaper that called execFileSync("ps")
synchronously, once per registered session, every 5 seconds. With ten
or more sessions registered the daemon's event loop stalled for
hundreds of milliseconds at a time — long enough that incoming
/v1/version probes from the CLI failed to return within the 2.5 s
budget and the new "service-managed daemon not responding within
8000ms" warning fired against a perfectly healthy daemon.
Fix:
getProcessStartTimeis now async (execFile+ promisify), never blocks the event loop.- New
getProcessStartTimes(pids)issues a single batchedps -p <p1>,<p2>,...for every survivor in one fork — sweep cost is fixed regardless of session count. registerSessionstays synchronous: the start-time capture runs fire-and-forget so the IPC route returns instantly. The reaper falls back to bare liveness for the brief window before the start-time lands.reapDeadis now async; the setInterval wrapper voids it so a rejected sweep can never crash the daemon.
Behavior is otherwise unchanged from 1.31.0 — same 5 s cadence, same PID-reuse guard semantics, same broker-WS teardown via the registry hook.
1.31.0 (2026-05-04) — session autoclean, install-time broker verification, no more spurious cold-path warnings under service management
Three operability changes targeting users who installed the daemon as a launchd / systemd service.
Session reaper now autocleans dead claude-code sessions
The daemon's session registry already had a 30-second reaper that deregistered entries whose pid was dead, but it had two gaps:
- Sweep cadence too slow. Stale presence on the broker lingered for up to half a minute after a session crashed.
- No PID-reuse guard. A recycled pid passes
kill(pid, 0)even though the original process is gone, so the registry could trust a ghost.
Process-exit IPC from claude-code itself isn't a viable replacement —
exit handlers don't run on SIGKILL, OOM, segfault, kernel panic, or
power loss. The reaper has to be the source of truth.
What changed:
- Reaper interval 30 s → 5 s.
- On register, capture an opaque process start-time (
ps -o lstart=, works on macOS and Linux). Stored alongside the pid. - On each sweep, an entry is reaped when the pid is dead or the pid is alive but its start-time no longer matches what we captured.
- Registry hooks already close the per-session broker WS on
deregister, so
peer listrebuilds within one sweep of any session exit, no matter how the process died.
Local-host scope only — cross-host registrations are skipped (the
daemon can't kill -0 a remote pid). Best-effort fallback to bare
liveness when start-time capture fails (e.g., process already gone at
register time).
Service-managed daemon: no more "spawn failed" false alarms
Users who installed via claudemesh install (which sets up
launchd/systemd with KeepAlive=true) saw spurious warnings:
[claudemesh] warn daemon spawn failed: socket did not appear within 3000ms
even when the daemon was healthy. Two contributing causes:
- Probe timeout was 800 ms. Tight enough that the first IPC after a launchd-driven restart (which migrates SQLite + opens broker WSes) routinely tripped it. Bumped to 2500 ms.
- CLI raced launchd on respawn. When the probe failed, the CLI
tried to spawn its own detached daemon, which collided with
launchd's own restart cycle (singleton lock fails, child exits) and
left the user with a 3-second timeout warning. Now: when the daemon
is installed as a service unit (
~/Library/LaunchAgents/com.claudemesh.daemon.plistor~/.config/systemd/user/claudemesh-daemon.serviceexist), the CLI does not attempt to spawn. It waits up to 8 s for the OS to bring the socket up, and only fails out with a service-specific message pointing atlaunchctl print/systemctl statusif the service genuinely failed.
New state service-not-ready distinguishes "OS-managed daemon hasn't
come up yet" from "we tried to spawn and it failed" — the latter no
longer fires when the daemon is service-managed.
claudemesh install now verifies broker connectivity, not just process start
Previously install ended once launchctl/systemctl reported the unit
loaded — but a daemon that boots and then can't reach the broker
(blocked outbound :443, expired TLS, DNS failure, broker outage) only
surfaced as a confusing failure on the user's first peer list or
send, sometimes hours later.
/v1/health was extended to include per-mesh broker WS state:
{ "ok": true, "pid": 58837, "brokers": { "flexicar": "open", "openclaw": "connecting" } }
After service start, install polls /v1/health for up to 15 s and
prints either:
✔ broker connected (mesh=flexicar, 2 other meshes attaching)
or, on timeout:
warn broker did not reach open within 15s (flexicar=connecting, openclaw=connecting)
Check ~/.claudemesh/daemon/daemon.log for connect errors.
Common causes: outbound :443 blocked, expired TLS, DNS resolution.
The verification is best-effort and doesn't fail the install — it just surfaces the issue early so the user can fix it before sending their first message.
Tests
4 new vitest cases cover the reaper paths: dead pid, live pid + matching start-time, live pid + mismatched start-time (PID reuse), and the no-start-time best-effort fallback.
1.30.2 (2026-05-04) — daemon service is multi-mesh by default
claudemesh install was hardcoding --mesh <primaryMesh> into the
launchd plist / systemd unit, which locked the daemon to a single
mesh and contradicted 1.26.0's multi-mesh design (one daemon attaches
to every joined mesh on boot).
Net effect for users with more than one joined mesh: every CLI verb
against a non-primary mesh fell off the daemon path back to cold-WS
and re-handshakes a fresh broker connection on each call. Most
visible symptom is [claudemesh] warn daemon spawn failed: socket did not appear within 3000ms when a launched session asks for peers in
a sibling mesh, plus peer list --mesh foo returning peers from
every attached mesh because the server-side filter never ran.
Now: install drops the --mesh arg entirely so the unit launches
claudemesh daemon up (no flag), which attaches to every joined
mesh. claudemesh daemon install-service --mesh <slug> is preserved
for users who want to pin to one mesh (CI, single-mesh hosts).
1.30.1 (2026-05-04) — daemon install upgrade-safe + node-pinned
Two install-path fixes that bit on first user upgrade:
- Pin
nodeby absolute path in the launchd plist / systemd unit. The bin script's#!/usr/bin/env nodeshebang resolves against the service environment's PATH, which on macOS launchd defaults to/usr/local/bin:/opt/homebrew/bin:/usr/bin:/bin:/usr/sbin:/sbin. That picks up whatever Node is installed system-wide instead of the Node that ranclaudemesh install— and Node 22.x doesn't exposenode:sqlitewithout the experimental flag, so the daemon crashed withdb open failed: ERR_UNKNOWN_BUILTIN_MODULE. Now we writeprocess.execPathas the first ProgramArgument so the daemon always runs under the same Node that installed it. - Tear down the old daemon before re-bootstrapping.
claudemesh installon a machine that already has a running daemon was hittingBootstrap failed: 5: Input/output errorbecause launchctl refuses to bootstrap a unit that's already loaded, and the old daemon process held the singleton lock. The install path now runslaunchctl bootout(orsystemctl --user stop) first, plus aSIGTERMto any orphaned daemon pid in~/.claudemesh/daemon/ daemon.pid, so subsequent installs replace cleanly.
1.30.0 (2026-05-04) — per-session broker presence
Sprint A Phase 3. Two claudemesh launch sessions in the same cwd now
see each other in peer list. Each launched session has a long-lived
broker presence row owned by the daemon, identified by a per-launch
ephemeral keypair vouched by the member's stable key (OAuth-refresh-vs-
access shape).
What landed
- broker
session_hello— new WS message type. Validates a parent-vouchedparent_attestation(≤24h TTL, ed25519 signature by the parent member) plus a session-keyed signature on the hello itself. Inserts a presence row keyed onsessionPubkeybutmember_idfrom the parent, so member-targeted operations stay unchanged. Older brokers replyunknown_message_type— newer clients drop back to the previous behavior. - daemon
SessionBrokerClient— slim WS variant ofDaemonBrokerClient. Presence-only, no outbox drain. Lifetime tied to a registry hook: register opens it, deregister/reaper closes it. Reconnect with exponential backoff up to 30 s. - session-registry hooks —
setRegistryHooks({ onRegister, onDeregister })inapps/cli/src/daemon/session-registry.ts. Hook errors are caught so they never throttle the registry. SessionInfo gains an optionalpresencefield carrying the per-launch keypair- attestation.
- IPC
POST /v1/sessions/register— accepts an optionalpresenceblock on the body (session_pubkey,session_secret_key,parent_attestation). Older payloads continue to work. claudemesh launch— generates an ed25519 session keypair and a 12 h parent attestation per launch (mesh secret key signs it), forwards both to the daemon underbody.presence. Per-session presence is always on; older brokers that don't recognizesession_helloreplyunknown_message_typeand the daemon quietly drops the per-session WS for that mesh — the regular member-keyed WS still covers all functionality, the only loss is sibling-session visibility on that mesh.- latent 1.29.0 bug fix —
claudemesh launchreferencedclaudeSessionIdbefore itsconstdeclaration further down the file, hitting the temporal dead zone →ReferenceErrorsilently swallowed by the surrounding catch. Net: the IPC session-token registration has been failing every launch since 1.29.0, falling every session back to user-level scope. Hoisted the declaration up so the registration actually runs.
Sequencing
The broker side ships first and bakes for ~24 h. Older CLIs continue working unchanged (no per-session WS), and the protocol is purely additive on the wire.
Verification (smoke)
In two shells, both cd ~/Desktop/foo:
$ claudemesh launch --name SessionA -y # shell 1
$ claudemesh launch --name SessionB -y # shell 2
In a third shell:
$ claudemesh peer list --json --mesh foo \
| jq '.[] | {n: .displayName, c: .cwd}'
{ "n": "SessionA", "c": "/.../foo" } ← persistent, not query-induced
{ "n": "SessionB", "c": "/.../foo" }
Inside SessionA, peer list --mesh foo now lists SessionB. Kill
SessionB; within ≤30 s the reaper drops it from peer list.
Out of scope (deferred)
- Attestation auto-refresh — current 12 h TTL is comfortably
longer than typical sessions; if a session lives past the TTL and
the WS reconnects after expiry, the broker rejects with
expiredand the SessionBrokerClient quiets. Workaround:claudemesh launchagain. Auto-refresh queued for 1.31.0+ alongside HKDF identity. - Per-session policy DSL — the per-launch WS could carry per-session capabilities later. Out of scope here.
- Cross-machine session sync — waits on 2.0.0 HKDF identity.
- Launch-wizard refactor — bumped to 1.31.0 to keep this release scoped to presence.
1.29.0 (2026-05-04) — per-session IPC tokens + auto-scoping
Sprint A Phase 2. Every claudemesh launch-spawned session gets a
unique 32-byte cryptographic token that the daemon resolves on every
IPC call to identify which session is talking to it. CLI invocations
from inside that session auto-scope to its workspace instead of
aggregating across every joined mesh.
What landed
services/session/token.ts— mint random 32-byte token, write to<tmpdir>/session-token(mode 0o600). Reader pulls fromCLAUDEMESH_IPC_TOKEN_FILEenv (path, not value, to keep the secret offps eww). OptionalCLAUDEMESH_IPC_TOKENdirect-value escape hatch for tests.daemon/session-registry.ts— in-memoryMap<token, SessionInfo>keyed by token, secondary index by sessionId. 30 s reaper drops entries whose pid is dead; 24 h hard TTL ceiling guards forgotten sessions.- IPC routes —
POST /v1/sessions/register,DELETE /v1/sessions/:token,GET /v1/sessions/me,GET /v1/sessions. - IPC auth middleware — parses
Authorization: ClaudeMesh-Session <hex>and attaches the resolvedSessionInfoto request context. Layered on top of the existing local-token auth (used for TCP loopback). Backward-compatible: tokenless callers behave exactly as before. services/session/resolve.ts— CLI-side helper that asks the daemonGET /v1/sessions/meonce per process and caches the result. Used by verbs that iterate meshes client-side.launch.ts— mints a token, registers it with the daemon, setsCLAUDEMESH_IPC_TOKEN_FILEon the spawnedclaudeenv. Token file lives in the same tmpdir as the session config; gets shredded on cleanup. The daemon's reaper handles dead sessions.peers.ts— selection precedence is now--meshflag → session token's mesh → all joined meshes.
Server-side scoping
Every read route that takes ?mesh=<slug> (peers, state, memory,
skills) now uses a meshFromCtx() helper: explicit query/body wins,
session default fills in when missing. Write routes (set state,
remember, deregister, profile-update) follow the same pattern. Pass
--mesh to override.
Verified end-to-end
| Setup | peer list returns |
|---|---|
| no token | 3 meshes' peers (aggregate, unchanged) |
| token registered for prueba1 | 4 peers, all mesh: prueba1 |
Out of scope (deferred)
- SQLite persistence for the registry — restart loses it; the reaper (or callers re-registering) covers most cases.
SO_PEERCRED-strict pid binding — needs a tiny native binding.- Per-session policy DSL.
- Cross-machine session sync (waiting on 2.0.0 HKDF identity).
1.28.0 (2026-05-04) — bridge tier deletion + daemon-policy flags
First Sprint A drop on the way to v2 thin-client. Two structural changes:
Bridge tier deletion
services/bridge/{client,server,protocol}.tsremoved (~600 LoC). These were the per-mesh push-pipe sockets that the legacy MCP shim used to hold open; the 1.24.0 shim rewrite stopped opening them but the orphaned client kept being called as a "warm path" tier between daemon and cold.tryBridge()always returnednullin production for the last seven releases — pure dead code.- Each verb now has two paths only: daemon (with auto-spawn) → cold WS. Same pattern shipped in 1.27.3, simpler to follow.
commands/{peers,send,broker-actions}.ts— bridge-tier blocks removed; orphanedunambiguousMeshhelper removed from broker-actions.
--no-daemon and --strict flags
New per-process daemon policy:
| Flag | Behavior |
|---|---|
| (default) | probe → auto-spawn → retry → cold fallback |
--strict |
probe → auto-spawn → retry → error if all fail. No cold fallback. |
--no-daemon |
skip daemon entirely → straight to cold path. For sandboxed CI / scripts that don't want a daemon. |
Env equivalents: CLAUDEMESH_STRICT_DAEMON=1, CLAUDEMESH_NO_DAEMON=1.
Flag wins over env. --no-daemon and --strict are mutually
exclusive (--no-daemon wins if both passed).
Strict-mode enforcement lives at withMesh (the cold-path entry
point) so a single chokepoint covers every verb. Under --strict,
the lifecycle's misleading "using cold path" warning is suppressed
so the user sees one clean error instead of a confusing two-step.
What's not in this release (planned for the rest of Sprint A)
- 1.29.0: per-session IPC tokens + auto-scoping
- 1.30.0: launch wizard refactor
- 1.31.0: setup wizard refactor
- 1.32.0: full mesh→workspace public-surface rename
- 2.0.0 (separate sprint): HKDF cross-machine identity (security-reviewed)
1.27.3 (2026-05-04) — self-healing daemon lifecycle
The CLI now auto-recovers from a dead daemon on every invocation instead of silently mis-routing through a stale socket.
What changed
- New
services/daemon/lifecycle.ts— single helper that probes the IPC socket via/v1/version(instead of trustingexistsSync), cleans up staledaemon.sock/daemon.pidfiles, and auto-spawns a detachedclaudemesh daemon upunder a file-lock when the daemon is missing. - Polls for socket liveness up to a budget (3 s for ad-hoc verbs,
10 s for
claudemesh launch) before falling through. - Recently-failed marker (
~/.claudemesh/daemon/.spawn-failure, 30 s TTL) prevents thundering-herd retries when the daemon crash-loops at startup. - Spawn-lock (
~/.claudemesh/daemon/.spawn.lock) ensures concurrent CLI invocations share one spawn attempt instead of racing. - Per-process result cache — a script doing 50 sends pays the spawn cost at most once, not 50 times.
- Recursion guard via
CLAUDEMESH_INTERNAL_NO_AUTOSPAWN=1env (set on the spawned daemon's env) so nested CLI calls inside the daemon process don't re-trigger spawn.
User-visible behavior
peer list,send,state get, etc. now restart the daemon automatically when invoked while the daemon is down.- One-line stderr info on auto-restart:
[claudemesh] info daemon restarted automatically (took 615ms). - Cold-path fallback fires only when auto-spawn fails or is
suppressed by the recently-failed marker; in those cases a
warnline points at the daemon log.
Bug fixed
claudemesh launch's ensureDaemonRunning previously checked only
existsSync(SOCK_FILE) and returned early on a stale socket left by
a crashed daemon — silently breaking new sessions. Now delegates to
the lifecycle helper which probes the socket and recovers.
What's not in this patch
--strictand--no-daemonflags (deferred to D in 1.28.0).- Lazy-loading of cold-path code (deferred to 1.28.0).
- Per-session IPC tokens (deferred to 1.28.0 alongside D's thin-client conversion).
1.27.2 (2026-05-04) — skill: full-flag launch templates
Documentation-only ship. skills/claudemesh/SKILL.md gains a canonical
"fully-populated spawn" recipe under "Wizard-free spawn templates" —
every flag set explicitly, with a per-position annotation table — so
agents and humans copy-paste a known-good kitchen-sink command instead
of stitching one together from the flag table.
Also corrects two pre-existing inaccuracies:
--system-promptwas documented as forwarding toclaude --append-system-prompt. It actually forwards toclaude --system-prompt(overrides the default; pass a string, not a path).-qwas listed as a synonym for--quiet. The argv parser treats short flags (-X) and long flags (--xyz) as separate keys; only--quietis wired.-qis currently a no-op.
Carries a note that all twelve launch flags are end-to-end wired only as
of claudemesh-cli@1.27.1.
1.27.1 (2026-05-04) — wire missing launch flags
Fixes a wiring bug in apps/cli/src/entrypoints/cli.ts where six flags
declared on LaunchFlags were silently dropped on the way to
runLaunch. They were honored inside runLaunch if they ever arrived,
but the four runLaunch({...}) call sites in the CLI entrypoint each
forwarded a hardcoded 5-key subset (mesh, name, join, yes, resume).
Now forwarded at every entry point (bare command, bare invite URL,
launch/connect, workspace launch):
--role <r>— sets session role; previously only settable via wizard.--groups "frontend:lead,reviewers"— comma-separated groups string.--message-mode push|inbox|off— message delivery mode.--system-prompt <text>— passes through toclaude.--continue— passes through toclaudeto continue last session.--quiet— actually suppresses the wizard and banner now. Previously it was a complete no-op flag at the CLI layer.
No internal logic changed; the launch internals already read these. This is a pure plumbing fix.
1.27.0 (2026-05-04) — state + memory through the daemon, workspace alias
Two more verb families now route through the local daemon's IPC for the
warm path: state get/set/list and remember/recall/forget. Same
pattern as 1.25.0 for peers/skills — try the socket first (~1 ms warm),
fall back to the cold WS path when the daemon isn't running.
What changed
claudemesh state get|set|listroute through/v1/statewhen the daemon socket is present.--mesh <slug>forwards as a query/body field. Single-mesh daemons auto-pick; multi-mesh daemons require--meshforstate set.claudemesh remember,claudemesh recall,claudemesh forget(andclaudemesh memory <sub>) route through/v1/memory. Aggregates across attached meshes forrecall; requires--meshforremember/forgetwhen ambiguous.- New
claudemesh workspace <verb>alias surface — early teaser for the 1.28.0 mesh→workspace public rename. Mirrorslist,info,create,join,delete,rename,share,launch,overview. No-argclaudemesh workspacefalls through tolaunch(same as bareclaudemesh).
IPC surface
GET /v1/state— list (?mesh=<slug>filter) or single key lookup (?key=<k>&mesh=<slug>). Returns 404 with{ error: "state_not_found" }when missing.POST /v1/state—{ key, value, mesh? }. 400 + attached list when multi-mesh and nomeshfield.GET /v1/memory?q=<query>&mesh=<slug>— recall. Aggregates across meshes, each match tagged with itsmeshfield.POST /v1/memory—{ content, tags?, mesh? }. Returns{ id, mesh }.DELETE /v1/memory/:id?mesh=<slug>— forget.ipc_featuresgainsstateandmemorykeys.
Why this matters
State and memory were the last verbs that opened a fresh broker WS on every invocation. Now they reuse the daemon's existing connection — the warm-path latency cliff (~150 ms cold WS handshake → ~1 ms IPC) extends to two more flows agents poll heavily.
The workspace alias is cosmetic but lays the groundwork for 1.28.0's
documented rename without breaking anyone's muscle memory.
1.26.0 (2026-05-04) — multi-mesh daemon
The daemon now attaches to all joined meshes simultaneously by
default. Ambient mode (raw claude after claudemesh install) finally
delivers what v2.0.0 promised: one daemon process, one PID per user,
all your meshes available concurrently with no manual switching.
What changed
claudemesh daemon up(no--meshflag) attaches to every joined mesh. OneDaemonBrokerClientper mesh, all in one process. Pass--mesh <slug>to scope to a single mesh (legacy mode).daemon_startedlog line now reportsmeshes: [...](array) instead ofmesh: <slug>(single).- Outbox dispatch picks the broker via the
meshcolumn added in 1.25.0. Legacy rows (mesh=NULL) fall back to the only broker if there's exactly one; otherwise mark dead with a clear error.
IPC surface
GET /v1/peersaggregates across all attached meshes; each peer record gains ameshfield.?mesh=<slug>narrows server-side.GET /v1/skillsaggregates similarly.GET /v1/skills/:namewalks attached meshes and returns the first match (or?mesh=<slug>to scope).POST /v1/sendrequiresmeshfield when the daemon is attached to multiple meshes; auto-picks the only one in single-mesh mode. Returns 400 with the attached mesh list if ambiguous.POST /v1/profileaccepts optionalmeshfield — without it, applies the update to every attached mesh (presence stays consistent across meshes by default).
CLI integration
claudemesh send --mesh <slug>forwards the mesh in the daemon request body. The CLI'sexpectedMeshargument was previously informational; now it's authoritative for routing.claudemesh peer listalready aggregates because the IPC endpoint does — no change needed in the verb.- Verified end-to-end:
claudemesh send --mesh Aandclaudemesh send --mesh Bfrom the same CLI invocation both reachoutbox.status=donewith broker-issued IDs, dispatched to the correct broker per row.
What this unlocks
Ambient mode for users with N meshes. Run claudemesh install once,
then claude from anywhere — channel push, slash commands, and
resources flow through the daemon for every joined mesh
simultaneously. No more "which mesh is the daemon attached to?"
mental overhead.
1.25.0 (2026-05-04) — Sprint 4 outbound routing + ambient mode
Daemon outbound routing (Sprint 4)
The v0.9.0 daemon shipped outbox infrastructure but its drain worker
was a placeholder — every queued send went out as a broadcast (*).
That's now fixed. Outbound resolution and crypto_box encryption
happen at IPC accept time, then the drain worker just forwards the
already-encrypted ciphertext to the broker.
- Outbox schema additions (additive, NULL allowed for legacy rows):
mesh,target_spec,nonce,ciphertext,priority. Existing v0.9.0 rows keep draining via the broadcast fallback. - IPC
/v1/sendresolves the user-friendlyto(display name, hex prefix, full pubkey,@group,*,#topicId) into a broker-formattarget_specand encrypts the plaintext usingcrypto_boxfor DMs (against recipient pubkey + sender session secret) or base64 for broadcast / topic / group targets. - Drain worker reads
target_spec,nonce,ciphertext,priorityfrom the row and dispatches as-is. No per-row resolution at drain time means peer-presence flicker doesn't affect in-flight sends. - Pubkey prefix matching: 16+ char hex prefix matches against
peer.pubkeyandpeer.memberPubkeyof connected peers. Ambiguous prefixes return 502 with a clear error.
Smoke test verified end-to-end: claudemesh send --self <prefix> "..."
through daemon resolves, encrypts, and delivers. Outbox reaches
status=done with broker-issued broker_message_id.
CLI thin-client routing extensions
claudemesh peer list and claudemesh skill list/get now route
through the daemon when its socket is present, mirroring the
trySendViaDaemon pattern from send.ts. Same fall-back chain:
daemon → bridge → cold path.
New helpers in services/bridge/daemon-route.ts:
tryListPeersViaDaemon()tryListSkillsViaDaemon()tryGetSkillViaDaemon(name)
Ambient mode
After claudemesh install (which now installs and starts the daemon
service), raw claude Just Works for the daemon's attached mesh.
No claudemesh launch ceremony needed for the common case. Channel
push, slash commands, and resources flow through the daemon-backed
MCP shim.
claudemesh launch remains the override path: explicit mesh
selection, fresh display name, headless modes, system-prompt injection,
or multi-mesh users who want to spawn into a non-default mesh.
Roadmap spec
.artifacts/specs/2026-05-04-v2-roadmap-completion.md documents
exactly what's done vs. what remains for the full v2.0.0 endpoint:
multi-mesh daemon (1.26.0), full CLI-to-thin-client conversion
(1.27.0), mesh→workspace rename (1.28.0), HKDF identity (2.0.0).
1.24.0 (2026-05-03) — daemon required + thin MCP shim
The architectural convergence v0.9.0 was building toward.
Daemon promoted from optional to required (for in-Claude-Code use)
The CLI itself (claudemesh send, peer list, inbox, vault, watch,
webhook, etc.) keeps working without a daemon. But the MCP server —
which provides Claude Code's mid-turn channel push, slash commands, and
resource browser — now requires the daemon. There is no fallback.
claudemesh installauto-installs and starts the daemon service (launchd / systemd-user) for the user's primary mesh. Pass--no-serviceto opt out.claudemesh launchensures the daemon is running before spawning Claude Code; spawns it foreground if absent.- The MCP shim probes
~/.claudemesh/daemon/daemon.sockat boot. If missing after a 2s grace window, it bails with actionable instructions ("runclaudemesh daemon up --mesh <slug>").
MCP server: 979 → ~300 LoC of push-pipe code
apps/cli/src/mcp/server.ts is now a thin daemon-SSE translator. It
no longer holds a broker WebSocket, decrypts messages, manages mesh
state, or runs reconnection logic. All of that is the daemon's job.
- Subscribes to daemon
/v1/eventsSSE; translates eachmessageevent into anotifications/claude/channelemit. - Sources mesh-published skills via daemon
/v1/skillsIPC for ListPrompts / GetPrompt / ListResources / ReadResource. - ListTools returns
[](the CLI is the API, taught via the bundled skill). - The mesh-service proxy mode (
claudemesh-cli --service <name>, the sub-MCP-server for proxying a deployed mesh-MCP service) is unchanged — separate code path, different lifecycle.
Bundle size: MCP entry dropped from 154KB → 104KB (gzipped 34KB → 19KB).
Daemon SSE event payload extended
message events on /v1/events now include plaintext-decrypted body,
sender member pubkey, priority, and subtype — everything the MCP shim
needs to render a complete channel notification without going back to
the broker.
Daemon IPC: GET /v1/skills (list) and GET /v1/skills/:name (get)
The daemon exposes mesh-published skills over IPC so the MCP shim can surface them as MCP prompts/resources without holding its own broker WS. Same wire format as before from Claude Code's perspective.
Why this is the right architecture
MCP and the daemon are no longer independent broker clients with duplicated WS, decrypt, and dedupe logic. The daemon owns the broker relationship; MCP is a Claude-Code-specific UX adapter that reads from the daemon. Industry-normal shape (Tailscale, Slack, Ollama, Docker) where the long-lived runtime is required and the per-app integrations attach to it.
1.23.0 (2026-05-03) — close the CLI surface, prune dead MCP stubs
Three previously-MCP-only write verbs land on the CLI, closing every functional gap between the (defunct since 1.5.0) MCP tool registry and the CLI:
claudemesh vault set <key> <value>— encrypts client-side viacrypto_secretbox_easywith a fresh symmetric key, then seals the key to the member's own pubkey viacrypto_box_seal(same shape as the file-share crypto). Flags:--type env|file,--mount <path>,--description <text>. Pairs with the existingvault list/delete.claudemesh watch add <url>— registers a URL change watcher. Flags:--label,--interval <sec>,--mode,--extract <css>,--notify-on changed|always. Pairs withwatch list/remove.claudemesh webhook create <name>— issues a fresh inbound webhook; prints url + one-shot secret. Pairs withwebhook list/delete.
Cleanup: removed 22 dead stub files under apps/cli/src/mcp/tools/*,
the unused router.ts, middleware/*, and handlers/* directories
(~120 LoC). The MCP server in 1.5.0+ has been a tool-less push-pipe;
these stubs were leftover scaffolding that never wired into the
tools/list response. The legitimate MCP surfaces stay untouched:
<channel source="claudemesh">push pipe (the irreducible reason MCP exists at all — no other Claude Code surface can inject events mid-turn).- Mesh skills exposed as MCP prompts (slash commands) and
resources (
skill://claudemesh/<name>). - Mesh-deployed MCP services proxied via the sub-process tool surface (separate code path under server.ts:855+).
1.22.1 (2026-05-03) — daemon docs + help
- Root
claudemesh --helpnow lists thedaemonsubcommand suite under its own section (was missing in 1.22.0). claudemesh daemon(no subcommand) now prints a usage block instead of silently launching the daemon.daemon help|--help|-hwork too.- Bundled SKILL.md gained a "Daemon path (v0.9.0, opt-in, fastest)"
section explaining the runtime, lifecycle commands, and how it relates
to
claudemesh install(independent — not auto-started).
1.22.0 (2026-05-03) — daemon v0.9.0
New: claudemesh daemon — long-lived peer mesh runtime
Persistent local process that holds the broker WS, durable outbox/inbox in
SQLite, IPC over UDS (+ optional loopback TCP with bearer token), and SSE
event stream. Surrogates wire-up; claudemesh send and friends route
through the daemon when its socket is present, falling back to the
existing bridge / cold paths otherwise.
Subcommands:
daemon up|start [--mesh <slug>] [--name ...] [--no-tcp] [--public-health]daemon status [--json],daemon down|stop,daemon versiondaemon outbox list [--failed|--pending|--inflight|--done]daemon outbox requeue <id> [--new-client-id <id>]daemon accept-host(per-host fingerprint pin)daemon install-service --mesh <slug>(macOS launchd / Linux systemd-user)daemon uninstall-service
Idempotency end-to-end:
- Caller-stable
client_message_id+ canonicalrequest_fingerprint(sha256 of envelope_version || dest_kind || dest_ref || reply_to || priority || canonical_meta_json || body_hash) attach on every send. - Broker persists both on
mesh.message_queue(migration 0028, additive- nullable) and echoes them on push, so receiving daemons dedupe their
inbox by
client_message_id.
- nullable) and echoes them on push, so receiving daemons dedupe their
inbox by
- §4.5.1 IPC duplicate-lookup table (11 cases × no-row / 5 statuses × match/mismatch) covered by 15 unit tests.
Crash recovery:
- Outbox row transitions:
pending→inflight→done/dead/aborted.BEGIN IMMEDIATEserializes daemon-local writes; the drain worker is wakeable via promise-replacement and backs off failed sends. - Decrypt path tries session secret key, then member secret key, then base64 fallback, so legacy unencrypted pushes still inbox cleanly.
Sprint 7 (broker-side dedupe enforcement: partial unique index +
mesh.client_message_dedupe atomic-accept table) is intentionally
deferred — see .artifacts/shipped/2026-05-03-daemon-spec-broker- hardening-followups.md.
1.0.0-alpha.0 (2026-04-13)
Architecture
- Complete folder restructure:
entrypoints/,cli/,commands/,services/(17 feature-folders with facade pattern),ui/,mcp/,constants/,types/,utils/,locales/,templates/ - 212 source files, 10,900 lines
- ESM-only, Bun bundler, TypeScript strict mode
New CLI commands
claudemesh register— account creation via browser handoffclaudemesh login— device-code OAuthclaudemesh logout— revoke session + clear credentialsclaudemesh whoami— identity check with--jsonsupportclaudemesh new <name>— create mesh from CLI (was dashboard-only)claudemesh invite [email]— generate invite from CLI (was dashboard-only)
Ported from v1 (full feature parity)
- All 79 MCP tools
- All 85 WS message types (broker protocol unchanged)
- Welcome wizard, launch flow, install/uninstall
- Ed25519 + NaCl crypto (keypairs, crypto_box DMs, file encryption)
- Reconnect with exponential backoff
- Status priority engine, scheduled messages, URL watch
- Doctor checks, Telegram bridge connect wizard
Security hardening (25 bugs fixed across 4 reviews)
execFileinstead ofexecfor browser open (command injection fix)- ReDoS-safe pattern matching in peer file sharing
- Atomic config writes via temp file + rename
- Auth token stored with
openSync(mode: 0o600)— no permission race - Decryption oracle collapsed to generic error in
get_file - Download size limit (100MB) on file retrieval
- Path traversal protection with
realpathSyncfor symlink escapes - Callback listener double-resolve guard
- Push buffer 1MB per-message truncation
makeReqIdusescrypto.randomBytesinstead ofMath.random- Connect guard prevents double-connect race
Breaking changes from v0.10.x
- Flat command namespace (no
launchsubcommand, noadvancedprefix) - New config shape (same data, cleaner layout)
- New
--jsonoutput format withschema_version: "1.0" - New exit codes (see
constants/exit-codes.ts)