KanBorg 1.2.31

Updates now pause Autopilot across saved workspaces and send a durable notice to each active coordinator. Notices remain available to paired coordinators even when their workspace is closed; supported wake delivery continues while the update is pending. The update panel shows acknowledgement status. Polling-only harnesses receive notices at their next check-in.

Workers awaiting a user decision no longer block restart after their process has exited. Questions, evidence and waiting state stay intact. Running processes and checks still block restart.

After update, each workspace receives one reconciliation request and restores its prior Autopilot setting. Cancellation clears obsolete pause notices and restores the prior setting without changing controller ownership.

Validation: desktop regression suite (136 passed), coordinator tests (48 passed), desktop/web type checks, and signed/notarized macOS package verification. Existing settings and connections retain their storage identity.

KanBorg 1.2.30

KanBorg 1.2.30 makes Autopilot settings persist across relaunches and makes the Start/Apply action usable for existing assignments.

• Restores saved perspectives, effort and parallelism in the Autopilot form.

• Retains On, Off and wind-down state. When On, a normal relaunch queues one coordinator reconciliation request without blindly restarting workers.

• Start Autopilot starts or resumes using the displayed settings; Apply settings updates a running assignment without replacing workers.

• Attempts coordinator delivery immediately and shows pending, received or failed delivery. External coordinators still need a working wake adapter or check-in.

• Failed delivery remains pending and can be retried by applying the settings; it is no longer reported as received.

• Passes the saved effort and objective into delegated worker instructions and hides finished worker history from the launch controls.

• Preserves reset credits instead of redeeming them at 95% usage. Automatic redemption now only rescues credits within 30 minutes of reported expiry; unknown expiry is preserved.

Validation: 133 desktop tests, final coordinator regression tests, desktop and web TypeScript checks, signed and Apple-notarized macOS packaging, and isolated update/restart recovery.

Claude recovery: locates native installations even when GUI PATH differs, distinguishes CLI login from usage-token expiry, keeps Reconnect available, installs the official CLI automatically when genuinely missing, and allows a confirmed subscription CLI to renew on eligible work while keeping historical quota labeled stale. Browser consent remains required when renewal cannot recover the login.

KanBorg 1.2.29

KanBorg 1.2.29 fixes the macOS verification warning that prevented downloaded releases from opening.

• Signed with the existing Developer ID Application identity and hardened runtime enabled.
• Notarized by Apple, with the notarization ticket attached to the application.
• Packaging now requires notarization credentials and refuses to generate an update manifest unless ticket validation and macOS Gatekeeper assessment pass.
• Retains the existing application identity and data location so saved projects, settings, and encrypted connections remain available.

Validation: 122 desktop tests, TypeScript check, signature verification, stapled-ticket validation, Gatekeeper acceptance, disk-image integrity, and isolated update/restart recovery.

Download the Apple Silicon DMG below, open it, and replace KanBorg in Applications. Existing settings are retained.

KanBorg 1.2.28 — reliable update cancellation

Cancelling an update no longer leaves Autopilot stuck winding down when staged installer cleanup fails. Cancellation is durable; bounded cleanup retries end in a warning without blocking workflow restoration. The coordinator also reconciles stale wind-down state from cancelled, failed or completed updates before resume and during its normal checks.

An update cancelled while worker-readiness checks are pending cannot proceed to installation afterward. Existing workers are not stopped or duplicated by cancellation.

Validation: all 122 desktop tests and TypeScript checks passed, including cleanup failure, cancellation during readiness and stale wind-down recovery. The affected running 1.2.26 instance was recovered through its existing controller by repeating cancellation with a fresh request ID, then resuming Autopilot; the coordinator confirmed its workers remained active. This release includes the 1.2.27 phase-context and stable Developer ID signing changes. Signed, not notarized.

KanBorg 1.2.27 — phase checkpoints and stable signing

KanBorg now saves structured phase checkpoints at stage transitions, completed implementation, passed QA, and merge boundaries. Follow-ups consume the checkpoint once and start fresh model context in the same run and worktree. Task history, exact commit/check/review references, and unresolved constraints remain available through the recorded evidence.

Routed workers recover from a context-limit rejection using a bounded background envelope and a reference to the full saved session. Recovery occurs only before useful output or tool execution, with bounded retries. It does not restart ambiguous work or create a duplicate worker. External coordinators receive checkpoint guidance; their own chat compaction depends on their harness.

macOS builds now require the stable Developer ID signature instead of ad-hoc signing, preserving the app identity used by Keychain. Existing storage identifiers and saved connections remain unchanged. A one-time Keychain approval may still be needed after migration. This release is signed but not notarized.

Validation: 119 desktop tests, 9 supervisor tests, and desktop TypeScript check passed. Developer ID signature verification, DMG integrity verification, and isolated legacy-installer startup/recovery checks passed; the project was restored and an external coordinator resume request was durably queued. This release does not claim to resolve the separate multi-project runtime retention or dispatch-timeout reporting issues.

KanBorg 1.2.26 — reliable capacity monitoring

Capacity monitoring now preserves valid historical readings across temporary failures and prevents unknown grant balances from activating paid fallback. Codex reads quota before models, so model discovery failures cannot consume the reporting timeout. Empty quota responses preserve the last same-account reading, and late Claude responses cannot replace a newer sign-in's snapshot.

Provider settings invalidate capacity immediately, including changes made during a refresh. The capacity panel includes Refresh capacity, honors provider cooldowns, and retains readings on refresh failure. Routine monitoring no longer launches Claude auth-status commands. Expired reset windows and invalid balances cannot authorize automatic spending.

Also includes the scoped existing-chat Codex CLI adapter previously tested in the isolated integration build. Transport acceptance remains separate from receiver acknowledgement and action completion.

Validation: 114 desktop tests, 10 Claude usage tests, desktop/web typechecks, local bundle signature, disk-image checksum, and isolated legacy-update/startup/project-restoration/coordinator-resume smoke test passed.

Live account limitations: expired Claude authorization still needs renewal; old manually entered grant balances need confirmation or organization-report recovery. The release does not claim those account connections were repaired or their live balances verified. No production credentials were changed.

KanBorg 1.2.25 — provider monitoring recovery

Provider monitoring recovery improvements

• Detect expired Claude sign-in before sending usage requests; reconnect no longer accepts an expired credential as successful authorization.
• Preserve provider rate-limit deadlines across restart and credential changes.
• Keep Codex quota snapshots for the same verified account; read usage even when its model catalogue fails, and stop requests promptly after CLI exit.
• Refresh Anthropic organization reports independently so one failed report does not suppress another. Persist retry deadlines and recover previously denied connections on bounded checks.
• Preserve grant balance provenance, retain rollover balances across cycles, and avoid subtracting the same local spend twice from provider estimates.
• Add direct recovery links from capacity details to provider settings.

Known live recovery requirements: expired Claude credentials require sign-in recovery. Legacy monthly balances without a confirmation timestamp require a current balance confirmation or working organization reporting. Unknown balances are never promoted to fresh credit automatically.

KanBorg 1.2.24

KanBorg 1.2.24 improves coordinator handoffs and prepares the KanBorg side of existing-chat integration.

• Built-in coordinator wrap-up is delivered directly, including after a busy turn. Progress and retry states are visible.
• Queued coordinator messages offer Steer now, which interrupts only the coordinator turn and prioritizes the selected direction.
• Adds durable, epoch-bound outbound events, exact chat-binding proposals with owner approval, receiver receipt/action acknowledgement, and explicit incoming handoff reconciliation.
• Adds getControlCoverage to report supported MCP controls and owner-only/UI-only boundaries.
• Host delivery uses a trusted adapter interface. No production host adapter is enabled in this build; automatic existing-chat wakeup is not claimed. A documented CLI candidate is undergoing separate same-chat acceptance testing.

Validation: desktop/web typechecks, 82 desktop tests, complete macOS bundle signing verification, and isolated legacy-update/startup recovery. Live host wakeup and iOS acceptance remain pending.

KanBorg 1.2.23

Fixes the incomplete application signature that caused macOS to report the downloaded KanBorg app as damaged.

Local macOS packaging now seals the complete bundle with an ad-hoc signature and requires strict signature verification before creating the installer. This remains a local, non-notarized build, not an Apple Developer ID release.

Verified: complete bundle signature, installer checksum, and isolated legacy update/startup test with project restoration and coordinator resume request.

KanBorg 1.2.22

KanBorg 1.2.22 makes the built-in coordinator conversation the primary surface and keeps setup and update details in Activity & settings.

• Send follow-ups while the coordinator is working; inspect, edit, or remove queued messages. Steer now stops only the current coordinator turn and prioritizes your new direction; existing workers continue.
• Questions for you keeps structured questions visible with proposed choices, Tell me more, and free-text answers. Replies return to the originating thread. Explanation requests leave questions pending; failed delivery does not mark an answer complete.
• Change coordinator lists every approved connection, plus the built-in coordinator. Existing connections can receive control through the normal handoff without repeating pairing.
• Queued messages and questions persist locally across restarts. Interrupted delivery is not automatically replayed.

Validation: desktop and web typechecks; 75 desktop tests; isolated browser preview of the coordinator layout and approved destinations. Live provider conversations and harness handoffs remain for user acceptance testing.