v1.4.1

1.4.1 - 2026-06-08

• v1.4.0 was cut but its publish step never ran; v1.4.1 ships the same onboarding rebuild as the first publicly published version of the 1.4.x line. The fullscreen Agent workspace, Agent-local behavior, isolated Agent Knowledge, connected-host operator integration, explicit side-effect boundaries, and release hardening from 1.3.x all stay in force.
• Fixed a release-evidence test that hardcoded a 1.3.0-specific phrase (compact model-visible harness pass) and prevented the v1.4.0 publish workflow from completing. The assertion now targets the stable required-theme fullscreen Agent workspace, so future release-notes rotations no longer break release CI.
• Rebuilt the Agent Workspace onboarding modal so first-time users land in a focused ONBOARDING-only category list, no longer see HOME and post-onboarding categories competing for attention, and never get kicked out of the modal when activating a slash-command row.
• Added typed commandBehavior (inline | compose | exit) on workspace actions so safe read-only commands can run inside the onboarding modal with their captured output rendered into the result pane, instead of forcing the modal closed to dispatch through the composer.
• Guarded the new inline dispatch path so a missing executeCommand no longer clobbers context.print; the modal surfaces a clear "command unavailable" result and stays open.
• Replaced release-engineering vocabulary (smoke history, receipt gaps, durable receipt, closeout policy, schema status, event cursor, publication guarantee, step history, repeated blocker, setup checkpoint summary) with a tight progress + single next-action summary in the onboarding right pane.
• Closed a leak where channel-guide context lines could expose internal userRoute strings containing model-tool-call syntax; the channels guide now shows only human-readable labels.
• Trimmed the user-facing setup checklist from fifteen items to thirteen by removing the install-smoke and browser-pwa release-engineering items; both remain available to model-facing routes but no longer appear in the onboarding flow.
• Renamed the misleading command field on AgentWorkspaceSetupChecklistItem to breadcrumb so the field's purpose (UI navigation hint) no longer collides with the kind: 'command' action type.
• Added clear verb-led row labels in the onboarding modal so users can see exactly what activating a row does: "Choose provider and model", "Edit MCP server", "Switch to ", "Run: ", "Open: ", "Finish setup", and so on, replacing generic labels like "Open option" and "Open guided form".
• Added a sticky "Finish setup" footer row visible on every ONBOARDING category, colored green when prerequisites (provider/model, connected-host auth, runtime) are ready and warn-colored with the unmet items listed otherwise; the row is wired through the real activation path so pressing Enter triggers completeOnboarding().
• Added per-category readiness glyphs in the onboarding left pane: a green checkmark for categories whose mapped checklist items are all ready, a warn glyph for categories with blockers or recommended items, and no glyph for optional-only categories.
• Relocated the diagnostic fields on AgentSetupWizard (smokeHistory, stepHistory, receiptGaps, closeout, checkpoint, repeatedBlocker) under a _diagnostic subkey so model-facing wizard mechanics are cleanly separated from user-facing wizard state, while preserving the existing model-tool surface.
• Tightened AgentWorkspaceCategory.group and the workspace _onlyGroup filter to a literal union AgentWorkspaceCategoryGroup, so category-group typos fail at compile time instead of producing silent navigation bugs.
• Added end-to-end acceptance tests for the new behavior: inline command dispatch keeps the workspace active and populates lastActionResult with captured output, first-run filter restricts the modal to ONBOARDING categories only, and the jargon allowlist guarantees no banned vocabulary leaks back into user-facing copy.