chore: publish new package versions

[github-actions]

This PR was opened by the Changesets release GitHub action. When you're ready to do a release, you can merge this and the packages will be published to npm automatically. If you're not ready to do a release yet, that's fine, whenever you add more changesets to main, this PR will be updated.

Releases

@electric-ax/agents@0.4.15

Patch Changes

• 8bcadb7: Preserve existing undici global dispatcher interceptors when installing the Durable Streams fetch cache so Electric Agents Desktop keeps injecting Cloud auth headers after the built-in agents runtime starts.

• 5aa2d78: Give Horton a fork tool that creates a child session inheriting this conversation's history up to the latest completed response. Takes an optional entityUrl (omit for self-fork), an optional initialMessage (server delivers to the fork in the same round-trip — no follow-up send needed; not atomic with fork creation), and optional tags. The fork is created as a CHILD of the calling entity (same parent-ownership model as spawn_worker) and wires reply delivery through the same manifest-anchored wake — when the fork's next run finishes, the parent wakes with the response in the wake message.

  Horton's system prompt grows a "When to fork (vs spawn_worker)" section framing the two tools as a pair: both create a child the parent owns and gets replies from, the difference is what the child boots with — spawn_worker starts with an empty context (you brief it from scratch), fork starts with a copy of the conversation up to the latest completed response. Includes an explicit trigger pattern ("prefer fork when generating multiple variants the user wants to compare; don't inline") to route "give me three takes" / "evaluate these N approaches" prompts to fork rather than collapsing them into one inline response, plus the workflow for the parallel-exploration loop (end-turn-first, fork-once-per-branch with a different initialMessage each, wait for all responses before synthesising).

• a1c1e30: Add built-in schedule tools to Horton and document them in the system prompt.

• 1099366: Fix leftover Docker sandbox containers (electric-sbx-*) piling up.

  Sandbox containers are meant to be short-lived, but several gaps let them
  outlive the work they were created for — opening the desktop app could leave
  15+ containers running that were never explicitly started. This closes those
  gaps so a container only exists while something is actually using it:

  • Created only when used. A container now starts the first time an agent
    actually uses its sandbox (runs a command, reads/writes a file), so trivial
    wakes (scheduled ticks, bookkeeping) no longer spin one up.
  • Cleaned up on quit. Shutdown now tears down idle containers immediately
    instead of leaving their delayed-teardown timers to die with the process.
  • Leftovers reclaimed at startup. Containers are tagged with the process
    that created them; at startup, those whose owner is gone are reclaimed
    (throwaway ones removed, reusable ones stopped so their files survive), while
    containers a live process is still using are left untouched.

  Also: a failed container setup step no longer strands an untracked container,
  and all sandboxes are grouped under one electric-sandboxes entry in Docker
  Desktop so they can be stopped/removed together.

• Updated dependencies [d15852d]

• Updated dependencies [5aa2d78]

• Updated dependencies [1099366]

• Updated dependencies [1099366]

  • @electric-ax/agents-runtime@0.3.11

@electric-ax/agents-runtime@0.3.11

Patch Changes

• d15852d: Fix runtime-originated agent send attribution by sending from_principal, from_agent, and the active wake write token, and accepting from_agent when backed by a valid agent write token.

• 5aa2d78: Add ctx.fork(opts?) to HandlerContext, with an opts shape that mirrors ctx.spawn's where the semantics map:

  ctx.fork(opts?: {
    targetEntityUrl?: string  // omit for self-fork
    initialMessage?: unknown  // server delivers to the fork in the same round-trip (not atomic with creation)
    wake?: Wake               // overrides the default runFinished + includeResponse
    tags?: Record<string, string>
    observe?: boolean         // `false` = fire-and-forget (no parent, no wake, no manifest entry)
  })

  By default (observe: true), the new fork is a CHILD of this entity (same parent-ownership model as ctx.spawn), and a runFinished + includeResponse wake is registered on it server-side. Reply delivery uses the same manifest-anchored wake mechanism ctx.spawn uses — when the fork's next run finishes, this entity wakes with the response. observe: false opts out of the parent relationship entirely: no parent URL, no wake subscription, no manifest entry on the parent's stream.

  Internally writes a kind: 'child' manifest row on the parent's stream alongside the server-side wake registration, mirroring the spawn flow's bookkeeping so the relationship persists across wakes. Wired through new fields on RuntimeServerClient.forkEntity (parent, wake, initialMessage, tags) and WiringConfig.forkEntity. A normalizeWake helper translates the user-facing Wake type into the wakeRegistry-compatible shape, same logic createOrGetChild uses for spawn.

  The send tool's payload description now documents the canonical { text: "..." } shape for chat-rendered targets (Horton sessions, agent forks) so messages emitted by send render as chat bubbles instead of blank bars.

• 1099366: Docker sandbox creation now pulls the image only when it isn't already present
  locally, honoring the documented pullIfMissing semantics. Previously every
  container create called docker pull, which round-trips to the registry even
  for a fully cached digest-pinned image — making creation needlessly slow and
  prone to failing whenever the registry was briefly unreachable.

• 1099366: Fix leftover Docker sandbox containers (electric-sbx-*) piling up.

  Sandbox containers are meant to be short-lived, but several gaps let them
  outlive the work they were created for — opening the desktop app could leave
  15+ containers running that were never explicitly started. This closes those
  gaps so a container only exists while something is actually using it:

  • Created only when used. A container now starts the first time an agent
    actually uses its sandbox (runs a command, reads/writes a file), so trivial
    wakes (scheduled ticks, bookkeeping) no longer spin one up.
  • Cleaned up on quit. Shutdown now tears down idle containers immediately
    instead of leaving their delayed-teardown timers to die with the process.
  • Leftovers reclaimed at startup. Containers are tagged with the process
    that created them; at startup, those whose owner is gone are reclaimed
    (throwaway ones removed, reusable ones stopped so their files survive), while
    containers a live process is still using are left untouched.

  Also: a failed container setup step no longer strands an untracked container,
  and all sandboxes are grouped under one electric-sandboxes entry in Docker
  Desktop so they can be stopped/removed together.

@electric-ax/agents-server@0.4.18

Patch Changes

• d15852d: Fix runtime-originated agent send attribution by sending from_principal, from_agent, and the active wake write token, and accepting from_agent when backed by a valid agent write token.

• 5aa2d78: Add server-resolved fork anchor + spawn-parity body fields to POST /_electric/entities/<type>/<id>/fork.

  • anchor: 'latest_completed_run' is an alternative to fork_pointer: the server scans the source root's main history, finds the most recent runs row with status === 'completed', derives the matching { offset, sub_offset } pointer, and runs the existing pointer-fork path with it. Mutually exclusive with fork_pointer (400 if both); 400 if no completed run exists. Lets callers without access to the source's per-row pointer side-table (e.g. an agent forking via a tool) fork at the same anchor the per-row "Fork from here" UI uses.
  • parent overrides the new root fork's parent field, making it a CHILD of that URL (rather than inheriting the source's parent).
  • wake registers a subscription on the new root fork at fork time (same shape as spawn's wake).
  • initialMessage is delivered to the new root fork via entityManager.send after linkEntityDispatchSubscription runs — same ordering spawn uses, so the dispatcher is subscribed before the inbox row lands and the fork actually wakes on the message instead of sitting idle.
  • tags are stamped on the new root fork in addition to those copied from the source.

  Together these let an agent fork itself as a child and receive replies via the same manifest-anchored wake mechanism spawn uses, with a single round-trip fork-and-dispatch.

  Chat UI: readInboxText falls back to message and content keys when text isn't present, so messages sent by agents (which sometimes emit those shapes) render as a chat bubble body instead of a blank bar.

• Updated dependencies [d15852d]

• Updated dependencies [5aa2d78]

• Updated dependencies [1099366]

• Updated dependencies [1099366]

  • @electric-ax/agents-runtime@0.3.11

electric-ax@0.2.15

Patch Changes

• Updated dependencies [d15852d]
• Updated dependencies [8bcadb7]
• Updated dependencies [5aa2d78]
• Updated dependencies [5aa2d78]
• Updated dependencies [a1c1e30]
• Updated dependencies [1099366]
• Updated dependencies [1099366]
  • @electric-ax/agents-runtime@0.3.11
  • @electric-ax/agents@0.4.15

@electric-ax/example-agents-chat-starter@0.1.6

Patch Changes

• Updated dependencies [d15852d]
• Updated dependencies [5aa2d78]
• Updated dependencies [1099366]
• Updated dependencies [1099366]
  • @electric-ax/agents-runtime@0.3.11

@electric-ax/example-agents-walkthrough@0.1.3

Patch Changes

• Updated dependencies [d15852d]
• Updated dependencies [5aa2d78]
• Updated dependencies [1099366]
• Updated dependencies [1099366]
  • @electric-ax/agents-runtime@0.3.11

@electric-ax/example-deep-survey@0.1.22

Patch Changes

• Updated dependencies [d15852d]
• Updated dependencies [5aa2d78]
• Updated dependencies [1099366]
• Updated dependencies [1099366]
  • @electric-ax/agents-runtime@0.3.11

@electric-ax/agents-desktop@0.1.15

Patch Changes

• 24d2c34: Show a clear sign-in prompt when connecting an Electric Cloud server while signed out instead of surfacing pull-wake runner registration errors.

@electric-ax/agents-mobile@0.0.13

Patch Changes

• 3dbd075: Add session pinning to the mobile app: long-press a root session row (or any search result) to open a context sheet with the entity info (title, session id, type/status, subagents, runner, sandbox, spawned, last active) and a Pin/Unpin action; the in-session kebab menu also gets a Pin/Unpin item, mirroring the desktop tile menu. Pinned sessions surface in a Pinned section above the groups, persisted per-device in AsyncStorage — the mobile mirror of the web sidebar's pinning. Runner-param types in agents-server-ui's entityRuntime helpers are loosened to structural subsets so the mobile app can reuse them.
• 7892079: Per-runner recent working directories in the spawn UI, derived from the synced sessions list so the same recents appear on every device. The desktop picker becomes per-runner (replacing the localStorage list), and mobile gains sandbox-profile and working-directory selection — including sending the sandbox profile on spawn, without which the runtime ignores the chosen directory.
• Updated dependencies [3dbd075]
• Updated dependencies [d15852d]
• Updated dependencies [5aa2d78]
• Updated dependencies [5aa2d78]
• Updated dependencies [146f238]
• Updated dependencies [7892079]
• Updated dependencies [1099366]
• Updated dependencies [1099366]
  • @electric-ax/agents-server-ui@0.4.18
  • @electric-ax/agents-runtime@0.3.11

@electric-ax/agents-server-ui@0.4.18

Patch Changes

• 3dbd075: Add session pinning to the mobile app: long-press a root session row (or any search result) to open a context sheet with the entity info (title, session id, type/status, subagents, runner, sandbox, spawned, last active) and a Pin/Unpin action; the in-session kebab menu also gets a Pin/Unpin item, mirroring the desktop tile menu. Pinned sessions surface in a Pinned section above the groups, persisted per-device in AsyncStorage — the mobile mirror of the web sidebar's pinning. Runner-param types in agents-server-ui's entityRuntime helpers are loosened to structural subsets so the mobile app can reuse them.

• 5aa2d78: Add server-resolved fork anchor + spawn-parity body fields to POST /_electric/entities/<type>/<id>/fork.

  • anchor: 'latest_completed_run' is an alternative to fork_pointer: the server scans the source root's main history, finds the most recent runs row with status === 'completed', derives the matching { offset, sub_offset } pointer, and runs the existing pointer-fork path with it. Mutually exclusive with fork_pointer (400 if both); 400 if no completed run exists. Lets callers without access to the source's per-row pointer side-table (e.g. an agent forking via a tool) fork at the same anchor the per-row "Fork from here" UI uses.
  • parent overrides the new root fork's parent field, making it a CHILD of that URL (rather than inheriting the source's parent).
  • wake registers a subscription on the new root fork at fork time (same shape as spawn's wake).
  • initialMessage is delivered to the new root fork via entityManager.send after linkEntityDispatchSubscription runs — same ordering spawn uses, so the dispatcher is subscribed before the inbox row lands and the fork actually wakes on the message instead of sitting idle.
  • tags are stamped on the new root fork in addition to those copied from the source.

  Together these let an agent fork itself as a child and receive replies via the same manifest-anchored wake mechanism spawn uses, with a single round-trip fork-and-dispatch.

  Chat UI: readInboxText falls back to message and content keys when text isn't present, so messages sent by agents (which sometimes emit those shapes) render as a chat bubble body instead of a blank bar.

• 146f238: Polish the agents UI with improved spawn-form model controls, tooltips,
  macOS sidebar vibrancy styling, select sizing fixes, and a response-footer
  fork action in the timeline.

• 7892079: Per-runner recent working directories in the spawn UI, derived from the synced sessions list so the same recents appear on every device. The desktop picker becomes per-runner (replacing the localStorage list), and mobile gains sandbox-profile and working-directory selection — including sending the sandbox profile on spawn, without which the runtime ignores the chosen directory.

• Updated dependencies [d15852d]

• Updated dependencies [5aa2d78]

• Updated dependencies [1099366]

• Updated dependencies [1099366]

  • @electric-ax/agents-runtime@0.3.11

@core/electric-telemetry@0.2.8

Patch Changes

• 81f9d4b: Expose key stack-level metrics on the Prometheus /metrics endpoint: number of
  defined shapes, number of active shapes, replication lag (byte-based slot lag
  and time-based receive-lag histogram), retained WAL size, and per-status-code
  counts of shape-endpoint HTTP responses
  (electric_plug_serve_shape_requests_count{status="200"}, 409, 503, ...),
  and admission-control concurrency/rejection counts
  (electric_admission_control_acquire_current{kind=...},
  electric_admission_control_reject_count{kind=...}).
  These metrics were already collected and exported to OTel/StatsD/Call-Home but
  not to Prometheus, which previously only served system-level (CPU/RAM/BEAM)
  metrics. A new additional_prometheus_metrics telemetry option routes them
  through the single shared Prometheus aggregator without double-reporting via the
  other reporters. Assumes a single stack per instance.
• c025d2e: Export a per-request electric.plug.serve_shape.requests.count metric tagged by status, known_error and live.

@core/sync-service@1.6.10

Patch Changes

• 55c063a: Speed up relation change routing by collecting shape ids directly from the filter's shape table instead of walking where-condition indexes.

• 9dfa0eb: Fix a KeyError: key :consider_flushed? not found in: nil crash in Electric.Shapes.Consumer. consumer_can_suspend?/1 now refuses to suspend while a transaction is pending, so the consumer hibernates instead and suspends only once the transaction completes.

• 52549d3: Cap the overload-protection (OLP) mailboxes of the default console, OpenTelemetry, and Sentry logger handlers so error/log bursts shed messages instead of blocking Logger callers or growing unbounded.

  This is leading-edge protection only: it shields against the early phase of a redeployment/error burst but is not sufficient under deep scheduler starvation — the real fix for that is upstream (request-proxy admission control and snapshot-pool sizing).

  Note: sync_mode_qlen is intentionally not set on the OpenTelemetry log handler — its module forces sync_mode_qlen == drop_mode_qlen, so the option would be a no-op there.

• 81f9d4b: Expose key stack-level metrics on the Prometheus /metrics endpoint: number of
  defined shapes, number of active shapes, replication lag (byte-based slot lag
  and time-based receive-lag histogram), retained WAL size, and per-status-code
  counts of shape-endpoint HTTP responses
  (electric_plug_serve_shape_requests_count{status="200"}, 409, 503, ...),
  and admission-control concurrency/rejection counts
  (electric_admission_control_acquire_current{kind=...},
  electric_admission_control_reject_count{kind=...}).
  These metrics were already collected and exported to OTel/StatsD/Call-Home but
  not to Prometheus, which previously only served system-level (CPU/RAM/BEAM)
  metrics. A new additional_prometheus_metrics telemetry option routes them
  through the single shared Prometheus aggregator without double-reporting via the
  other reporters. Assumes a single stack per instance.

• fcf9749: Add queryable_columns as a means to restrict access to sensitive columns, a server-side shape allow-list for columns that may be synced or queried by subset snapshots.

• c025d2e: Export a per-request electric.plug.serve_shape.requests.count metric tagged by status, known_error and live.

• d3e1a64: Automatically recover sources whose replication slot creation is stuck waiting on a pending transaction, by periodically calling pg_log_standby_snapshot(). Previously such sources required a manual restart. When the function is unavailable (PostgreSQL < 14 or missing EXECUTE privilege), Electric falls back to the previous behavior and emits a one-time notice.

• c015163: Add a total_processing_time span attribute that tracks the wall-clock time
  taken to process all fragments of a single transaction.

[netlify]

✅ Deploy Preview for electric-next ready!

Name	Link
🔨 Latest commit	667329e
🔍 Latest deploy log	https://app.netlify.com/projects/electric-next/deploys/6a2815ee065e200008ce926a
😎 Deploy Preview	https://deploy-preview-4525--electric-next.netlify.app
📱 Preview on mobile	Toggle QR Code... Use your smartphone camera to open QR code link.

---

To edit notification comments on pull requests, go to your Netlify project configuration.