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@0.6.2

Patch Changes

• 9d063e4: Add desktop-configurable skill directories and common Claude/Codex skill locations so Markdown command files load recursively as slash-command skills without LLM metadata extraction.
• 16c8339: Use Anthropic's adaptive thinking API for models that require it (Opus 4.6/4.7,
  Sonnet 4.6). The builtin model catalog previously injected budget-based
  thinking: { type: "enabled", budget_tokens } for every reasoning-capable
  Anthropic model, which Opus 4.7 rejects with a 400 — it needs
  thinking: { type: "adaptive" } + output_config.effort. Those models now emit
  the adaptive shape (reasoningEffort mapped to the effort level); older models
  keep budget-based thinking.
• ac2391d: Capture OpenAI Codex reasoning summaries by default when built-in reasoning is enabled while preserving explicit reasoning summary settings from caller payloads.
• 8f4368d: Delegate built-in reasoning configuration to pi-ai instead of rewriting provider request payloads in the model catalog. Built-in OpenAI, OpenAI Codex, and Anthropic reasoning models now pass reasoning through the runtime adapter, preserving existing auto defaults and Anthropic thinking budgets while letting pi-ai own provider-specific payload shapes.
• 6dbb0bb: Require meaningful slugs when spawning workers and append a random suffix so worker URLs are easier to identify while remaining generally unique.
• ee0da19: Fix Electric Agents quickstart startup by authenticating the built-in pull-wake runner with the same principal it registers as, registering built-in agent types with the local runner as their default dispatch target, and aligning the CLI's default principal with the local quickstart user. Pin the CLI-launched agents-server Docker image to the matching released agents-server version, improve registration fetch errors so startup failures include the endpoint and underlying cause, avoid the CLI observe live-query Collection boundary, and clarify browser-only credentials settings copy.
• Updated dependencies [9d063e4]
• Updated dependencies [0418d90]
• Updated dependencies [ac2391d]
• Updated dependencies [8f4368d]
• Updated dependencies [354b01b]
• Updated dependencies [37b9922]
• Updated dependencies [ee0da19]
• Updated dependencies [c14a886]
  • @electric-ax/agents-runtime@0.6.2

@electric-ax/agents-runtime@0.6.2

Patch Changes

• 9d063e4: Add desktop-configurable skill directories and common Claude/Codex skill locations so Markdown command files load recursively as slash-command skills without LLM metadata extraction.
• 0418d90: Batch queued child completion wakes into a single wake payload so parent agents receive every child result without extra handler runs.
• ac2391d: Capture OpenAI Codex reasoning summaries by default when built-in reasoning is enabled while preserving explicit reasoning summary settings from caller payloads.
• 8f4368d: Delegate built-in reasoning configuration to pi-ai instead of rewriting provider request payloads in the model catalog. Built-in OpenAI, OpenAI Codex, and Anthropic reasoning models now pass reasoning through the runtime adapter, preserving existing auto defaults and Anthropic thinking budgets while letting pi-ai own provider-specific payload shapes.
• 354b01b: Enable provider retries for agent model calls by default so transient LLM errors are retried. Set modelMaxRetries: 0 to preserve the previous no-retry behavior.
• 37b9922: Fix agent wake handling so concurrent sessions do not invalidate each other's claim write tokens, and retry same-stream wakes after the active wake drains instead of dropping pending work.
• ee0da19: Fix Electric Agents quickstart startup by authenticating the built-in pull-wake runner with the same principal it registers as, registering built-in agent types with the local runner as their default dispatch target, and aligning the CLI's default principal with the local quickstart user. Pin the CLI-launched agents-server Docker image to the matching released agents-server version, improve registration fetch errors so startup failures include the endpoint and underlying cause, avoid the CLI observe live-query Collection boundary, and clarify browser-only credentials settings copy.
• c14a886: Remove the skill content max cap so use_skill loads complete skill instructions instead of truncating them.

@electric-ax/agents-server@0.6.2

Patch Changes

• 37b9922: Fix agent wake handling so concurrent sessions do not invalidate each other's claim write tokens, and retry same-stream wakes after the active wake drains instead of dropping pending work.
• Updated dependencies [9d063e4]
• Updated dependencies [0418d90]
• Updated dependencies [ac2391d]
• Updated dependencies [8f4368d]
• Updated dependencies [354b01b]
• Updated dependencies [37b9922]
• Updated dependencies [ee0da19]
• Updated dependencies [c14a886]
  • @electric-ax/agents-runtime@0.6.2

electric-ax@0.2.22

Patch Changes

• ee0da19: Fix Electric Agents quickstart startup by authenticating the built-in pull-wake runner with the same principal it registers as, registering built-in agent types with the local runner as their default dispatch target, and aligning the CLI's default principal with the local quickstart user. Pin the CLI-launched agents-server Docker image to the matching released agents-server version, improve registration fetch errors so startup failures include the endpoint and underlying cause, avoid the CLI observe live-query Collection boundary, and clarify browser-only credentials settings copy.
• Updated dependencies [9d063e4]
• Updated dependencies [16c8339]
• Updated dependencies [0418d90]
• Updated dependencies [ac2391d]
• Updated dependencies [8f4368d]
• Updated dependencies [354b01b]
• Updated dependencies [37b9922]
• Updated dependencies [6dbb0bb]
• Updated dependencies [ee0da19]
• Updated dependencies [c14a886]
  • @electric-ax/agents-runtime@0.6.2
  • @electric-ax/agents@0.6.2

@electric-ax/example-agents-chat-starter@0.1.13

Patch Changes

• Updated dependencies [9d063e4]
• Updated dependencies [0418d90]
• Updated dependencies [ac2391d]
• Updated dependencies [8f4368d]
• Updated dependencies [354b01b]
• Updated dependencies [37b9922]
• Updated dependencies [ee0da19]
• Updated dependencies [c14a886]
  • @electric-ax/agents-runtime@0.6.2

@electric-ax/example-agents-walkthrough@0.1.10

Patch Changes

• Updated dependencies [9d063e4]
• Updated dependencies [0418d90]
• Updated dependencies [ac2391d]
• Updated dependencies [8f4368d]
• Updated dependencies [354b01b]
• Updated dependencies [37b9922]
• Updated dependencies [ee0da19]
• Updated dependencies [c14a886]
  • @electric-ax/agents-runtime@0.6.2

@electric-ax/example-deep-survey@0.1.29

Patch Changes

• Updated dependencies [9d063e4]
• Updated dependencies [0418d90]
• Updated dependencies [ac2391d]
• Updated dependencies [8f4368d]
• Updated dependencies [354b01b]
• Updated dependencies [37b9922]
• Updated dependencies [ee0da19]
• Updated dependencies [c14a886]
  • @electric-ax/agents-runtime@0.6.2

@electric-ax/agents-desktop@0.6.1

Patch Changes

• 9d063e4: Add desktop-configurable skill directories and common Claude/Codex skill locations so Markdown command files load recursively as slash-command skills without LLM metadata extraction.

@electric-ax/agents-mobile@0.6.2

Patch Changes

• aa6ca85: Fix mobile chat timeline flashing when sending messages, resizing the composer (multiline/attachments) and queuing messages, and fix inconsistent auto-scroll as the composer grows. The timeline WebView embed now receives dynamic updates (bottom inset, inline queued messages, scroll) imperatively instead of via props that re-render and flash it. Also fix the composer occasionally not clearing after send when typing quickly.
• Updated dependencies [9d063e4]
• Updated dependencies [0418d90]
• Updated dependencies [ac2391d]
• Updated dependencies [8f4368d]
• Updated dependencies [354b01b]
• Updated dependencies [37b9922]
• Updated dependencies [06a60fa]
• Updated dependencies [aa6ca85]
• Updated dependencies [ee0da19]
• Updated dependencies [c14a886]
  • @electric-ax/agents-server-ui@0.6.2
  • @electric-ax/agents-runtime@0.6.2

@electric-ax/agents-server-ui@0.6.2

Patch Changes

• 9d063e4: Add desktop-configurable skill directories and common Claude/Codex skill locations so Markdown command files load recursively as slash-command skills without LLM metadata extraction.
• 06a60fa: Fix reasoning section elapsed timers so each thinking block measures from when that block starts, rather than from the beginning of the run.
• aa6ca85: Fix mobile chat timeline flashing when sending messages, resizing the composer (multiline/attachments) and queuing messages, and fix inconsistent auto-scroll as the composer grows. The timeline WebView embed now receives dynamic updates (bottom inset, inline queued messages, scroll) imperatively instead of via props that re-render and flash it. Also fix the composer occasionally not clearing after send when typing quickly.
• ee0da19: Fix Electric Agents quickstart startup by authenticating the built-in pull-wake runner with the same principal it registers as, registering built-in agent types with the local runner as their default dispatch target, and aligning the CLI's default principal with the local quickstart user. Pin the CLI-launched agents-server Docker image to the matching released agents-server version, improve registration fetch errors so startup failures include the endpoint and underlying cause, avoid the CLI observe live-query Collection boundary, and clarify browser-only credentials settings copy.
• Updated dependencies [9d063e4]
• Updated dependencies [0418d90]
• Updated dependencies [ac2391d]
• Updated dependencies [8f4368d]
• Updated dependencies [354b01b]
• Updated dependencies [37b9922]
• Updated dependencies [ee0da19]
• Updated dependencies [c14a886]
  • @electric-ax/agents-runtime@0.6.2

@core/sync-service@1.7.3

Patch Changes

• 214e644: Store parsed UUID values as 16-byte binaries instead of canonical text inside the sync-service filter evaluator. This reduces memory use in where-clause filter indexes, especially equality and subquery index keys, while converting UUIDs back to canonical strings at API and SQL output boundaries. The shape metadata version is bumped so persisted shape comparable terms created with the old UUID string representation are ignored and rebuilt.

• 0632b1c: Stop the ShapeLogCollector from crashing when table introspection fails on the hot replication path.

  Previously the replication pipeline only handled {:error, :connection_not_available} from the inspector; every other legal inspector return crashed the ShapeLogCollector and took the whole shapes supervision tree down with it, putting the replication client into a 10-minute noproc retry loop:

  • :table_not_found (table dropped or renamed between the WAL record and introspection) crashed Partitions.handle_relation/2 and the pk-column lookup with a CaseClauseError. Relation messages for dropped tables are now ignored and changes for dropped tables are keyed on the full record, the same fallback used for tables without a primary key.
  • In-band database errors (e.g. out-of-memory, statement_timeout cancelling the catalog query) crashed Partitions.handle_relation/2 with a CaseClauseError and made Partitions.add_shape/3 raise. They are now propagated as {:error, reason}; the collector logs a warning and pauses replication via the existing retry path until introspection succeeds.
  • Connection-class errors returned in-band by Postgrex (e.g. "ssl recv: closed") are now classified as :connection_not_available instead of leaking through as strings.
  • Shape restore no longer crashes with a MatchError when the connection pool isn't ready yet; it retries the introspection in place and only gives up (with a descriptive error) after ~10 seconds.

• 0cf38e8: Honor an upstream head-sampling rate hint received via the W3C tracestate header (electric=rate:<N>) on shape GET requests:

  When the remote parent trace is sampled, the Plug_shape_get root span (and the shape_get.plug.stream_chunk child spans) are stamped with the SampleRate attribute — N for responses with status < 500 and 1 for 5xx responses — so tracing backends that understand sampling weights scale aggregates over Electric's spans by the upstream sampling rate instead of under-reporting traffic ~N-fold.

  Requests without a remote trace context, with a not-sampled remote parent, or with a missing/invalid rate hint, behave exactly as before.