chore: release v0.42.0 — "KITT"

[josephfung]

KITT (Knight Rider, 1982, Glen A. Larson) — the Knight Industries Two Thousand, a car-bound AI you simply talk to: it holds a conversation by voice and deploys a deep bag of specialized functions on command. This release gives Curia the same — a voice you speak with, new channels to reach it through, and a real toolkit of skills it draws on when asked.

[0.42.0] — 2026-07-26 — "KITT"

Added

• Voice channel — talk to Curia by voice from the console: duplex WebRTC over self-hosted LiveKit, Deepgram speech-to-text, Cartesia text-to-speech, and barge-in. Join, mute, and hang up from chat; a principal hangup ends the session (1-hour call tokens, room teardown). Ships with a LiveKit Compose sample and an operator security/privacy guide (ADR-037). (Voice channel: duplex conversation via console WebRTC (Phase 1) #1414)
• True LLM streaming — Anthropic and OpenRouter now stream tokens and tool events, the primitive the voice tool-loop runs on. (Voice channel: duplex conversation via console WebRTC (Phase 1) #1414, feat(voice): duplex console WebRTC voice channel (Phase 1) (#1414) #1544, Extract shared streaming turn primitive (VoiceTurnRunner ↔ AgentRuntime) #1552, Opt AgentRuntime.handleTask into streaming turn primitive #1563)
• Console settings & system pages — Personality, Ghostwriting, Posture, Autonomy, Memory, and System pages replace the Workspace stub, backed by read-only GET /api/system and GET /api/memory/retention. (Epic: Operator config-refinement pages in the web console #1376)

Changed

• Slack channel — Socket Mode DMs, @mentions, and in-thread replies, with a slack-send skill and a channel-agnostic inbound.reaction bus event (ADR-033). (Slack channel: Socket Mode DMs + @mentions (workspace-owned app) #1477, Add slack-send skill — bring Slack to channel parity (proactive outbound + Gate C) #1526)
• SMS channel — two-way texting over a Telnyx office DID with medium-trust webhook inbound and an sms-send skill (ADR-036). (SMS channel: Telnyx two-way office DID (text messaging) #1478)
• Tools & skills system — invocation atoms are now tools (tool.json, ToolRegistry) and bundles are skills; pin a whole bundle, a single tool, or an MCP server projected as a skill, and activate a skill's tools + instructions at runtime. Unmodified Anthropic SKILL.md folders can be dropped in and activated (ADR-031, ADR-032). (Phase 1: tool/skill rename (groundwork) #1485, Phase 2: bundle (skill) model #1489, Phase 3: imported NL + asset skills #1490, Phase 2 follow-up: bundle the remaining tool clusters into skills #1494, Phase 3a: skill activation runtime (unified discovery + lazy instruction-loading) #1495)
• Contact identity editing — add, verify, retire, remove, and merge a contact's channel identities from the console or API; verifying the principal's Slack/Signal binds outbound detection. (Add a way to edit contact channel identities (bind principal's non-email channels; fixes principal-detection gap) #1514)
• Reaction approvals — 👍/👎 reactions (including skin tones) on Slack and Signal resolve pending approvals. (Wire inbound.reaction to approval decisions (Slack + Signal, channel-agnostic) #1479)
• Audit hardening (Phase 1) — structured columns, a seq-ordered hash chain, an archivable llm_call_archive (kill-switch + TTL), authorization.decision logging for Gate-1/Gate-C outcomes, and pnpm audit:verify. (feat: audit log hardening Phase 1 — structured columns, LLM provenance, hash chain #1383, feat: audit-log authorization decisions (allow/deny/escalate) before v1.0 #1379)
• Database-outage resilience — DATABASE_UNAVAILABLE errors, pool timeouts, skill retry, and a CEO alert after 5 minutes down; new errorType on ToolResult (public API). (feat: graceful degradation when database unavailable during task execution #1381)
• Operator home — the console root shows health, attention, and activity cards plus a chat call-to-action. (Operator home: replace the blank console landing page #1375)
• Channel plumbing — each channel owns its outbound request variant and recipient projection and contributes its principal carve-out rules via a registry, so the gateway has no central per-channel switch (ADR-035); shared contact resolve-or-create closes a 1:1 orphan leak. OutboundSendRequest gains slack and sms variants (public API); a new dev guide walks through building a channel end-to-end. (Extract channel contact auto-create (create→link→orphan cleanup) into ContactService #1480, Refactor: make principal-recipient resolution channel-modular (not a central per-channel switch) #1510, Refactor: channel-owned outbound recipient projection (zero gateway edits per channel) #1513)
• Smaller refinements — approve-grant-recommendation risk raised to critical (feat(skills): bundle remaining tools + polymorphic pins #1499); KG relationship tools moved to the memory skill (Move query-relationships / delete-relationship into the memory skill (core KG management) #1502); unresolved scheduled pins now log at error (Scheduled agent with unresolved pinned tools runs blind — should alert / support required tools #1501); Ant Farm renders the real office art (Centralize Ant Farm licensed art into the core repo (LimeZu redistribution approved) #1504); specs consolidated and dev docs re-synced to what shipped (docs(specs): merge resumable-tasks into tasks spec, prune stale checklists #1554).

Fixed

• Voice reliability — hard or repeated TTS errors end the call cleanly (tts_error); LiveKit room deletion uses the internal address; bounded STT waits and safe session cleanup on token failure; the Cartesia voice ID is now a required, console-set credential. (Voice channel: duplex conversation via console WebRTC (Phase 1) #1414, Voice: LiveKit room DeleteRoom 404s — management client uses public signaling URL, not internal livekit:7880 #1555, Voice: TTS synthesis failure is warn-and-swallowed per turn — caller hears silence with no escalation #1556)
• Audit — empty audit_log no longer breaks migration 080, llmCallArchive config validates, and canonicalJson round-trips Dates so verification matches write-time hashes. (feat(audit): Phase 1 hardening — structured columns, hash chain, LLM archive #1540)
• Scheduler & tool output — scheduler-list bounds and trims results and returns timestamps in your timezone; tool outputs cap total object size, not just per-leaf. (scheduler-list overflows model context on large jobs tables → generic 'unable to process' failure #1487)
• Dashboard Recent Activity — the timeline card loads with a 24-hour scope. (Dashboard "Recent Activity" timeline always fails to load (scopeless /api/antfarm/timeline request) #1548)
• OpenRouter errors — last_error now names the upstream provider and reason, not the opaque wrapper.
• Skills bundling — calendar risk/docs, manifest and registry-install correctness, and test-harness expansion of pinned_skills bundles. (Phase 2: bundle (skill) model #1489, feat(skills): bundle remaining tools + polymorphic pins #1499)

Security

• Dependency hardening — pinned or upgraded @fastify/static, hono, @hono/node-server, body-parser, fast-uri, find-my-way, sharp, shell-quote, js-yaml, and brace-expansion to clear known path-traversal, XSS, and DoS CVEs. (Security: clear remaining medium/low Dependabot CVEs (body-parser, @hono/node-server, js-yaml) #1518)
• SPA file serving — console and Ant Farm serve only containment-checked paths under an explicit root.
• Supply chain — pnpm trustPolicy: no-downgrade (7-day grace) and baked npm bumped 11.17.0→11.18.0 with a refreshed node:24-slim digest. (Security: bump baked npm to clear new node-tar + brace-expansion CVEs in Docker image #1521)

---

It learns to listen—
speech becomes text becomes thought.
It answers. Out loud.

[josephfung]

@coderabbitai review

[coderabbitai]

✅ Action performed

Review finished.

Note: CodeRabbit is an incremental review system and does not re-review already reviewed commits. This command is applicable only when automatic reviews are paused.

[coderabbitai]

📝 Walkthrough

Walkthrough

The changelog now documents the 0.42.0 “KITT” release across added features, changes, fixes, and security updates. The README version badge and package.json package version are both updated from 0.41.0 to 0.42.0. A modest release-number adjustment, requiring only the intellectual capacity of a particularly tired calculator.

🚥 Pre-merge checks | ✅ 4

✅ Passed checks (4 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly and concisely identifies the 0.42.0 KITT release, which matches the changeset.
Description check	✅ Passed	The description directly summarises the release contents and aligns with the files and objectives in this PR.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

---

_{Comment @coderabbitai help to get the list of available commands.}

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@CHANGELOG.md`:
- Line 16: Update the CHANGELOG.md section under ## [Unreleased] with one
concise bullet describing the logical change introduced by this PR, keeping the
existing 0.42.0 release entry unchanged.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: d4272d0a-01cc-4c4c-9727-9549fbc09251

📥 Commits

Reviewing files that changed from the base of the PR and between b009bef and 656b6b7.

📒 Files selected for processing (3)

• CHANGELOG.md
• README.md
• package.json