Vorno 0.15.0

Features

• Automations can now act on sessions from any event. set-status, set-labels, and send-message used to be accepted only on WebhookReceived, which made most lifecycle workflows impossible to express. They now work on any automation event — add a label and have the status follow, react to a status change by relabelling, message a session when something happens to it. Target a session with $CRAFT_SESSION_ID. Loop protection is built in, so an action that would re-trigger its own rule stops instead of running away, and every action records who caused it in the automation's history.
• Automations that can never run are now reported instead of failing silently. A rule with an invented action type, a mistyped filter key, a malformed action, or a misspelled event name used to validate clean, look healthy, and do nothing forever — and a mis-keyed filter was worse than useless, because the rule would fire on every event of its type rather than none. Those cases now surface as errors and warnings in config validation (with "did you mean" suggestions) and as entries in the Automations history view, on both first load and hot reload after a hand-edit. Disabled rules are exempt.
• Claude Opus 4.6 is selectable again — it reappears in the model pickers for direct Anthropic connections, Amazon Bedrock, and Pi-backed catalogs. Connections that were force-migrated from Opus 4.6 to Opus 4.8 get 4.6 re-added automatically (one-shot: your default model is untouched, and removing it again sticks). Opus 4.8 remains the default for new installs (from upstream)
• New archive_session session tool — agents can archive (or unarchive) another session in the workspace by ID to tidy up finished or superseded sessions. Archiving only hides the session from the active list and unread counts, never deletes it; a session cannot archive itself, and archiving is refused while the target is mid-turn (from upstream)

Improvements

• Claude Agent SDK uplifted to 0.3.220 (from 0.3.197) — Claude Code v2.1.220 parity. Fixes an abort-listener leak in streaming queries, adds CLI stderr to process-exit errors, reports correct HTTP status for rate-limit errors, and supports session resume with dash-leading IDs (from upstream)
• Subagent nesting depth is unchanged at 5. Upstream tightened its cap this release, but Vorno pins the previous limit so agents can keep spawning nested agents as before; set CLAUDE_CODE_MAX_SUBAGENT_SPAWN_DEPTH to override
• Moving a card into a closed column (Done, Cancelled) on the board still works with no extra confirmation — dragging it there is the decision. What changed is that automated paths no longer get that authority by accident: agents can never close a session, and an automation can only do so if its rule was registered with allowClosed.
• If installing a downloaded update fails at the final handoff, the app now restarts cleanly instead of continuing in a half-shut-down state where later work could be lost (from upstream)

Bug Fixes

• Running tasks show as "In Progress" on the board again. The built-in status set was missing in-progress where new workspaces are seeded, so a task the runner had actually started read back as "todo" — a board where nothing ever appeared to be running. There is now a single source of truth for the built-in statuses, and existing workspaces are repaired on load without touching statuses you authored yourself.
• macOS auto-update installs again — updates no longer end up downloaded-but-never-installed because quit cleanup interrupted the installer handoff (from upstream)
• The chat input no longer force-capitalizes the first letter, which broke pinyin and other CJK IME composition (from upstream)
• Local file links whose paths contain spaces (%20) open correctly again from chat messages (from upstream)
• New sessions no longer inherit an excluded filter — a lone "exclude Done" filter used to create new sessions as Done; only include-mode filters are inherited now (from upstream)
• Custom OpenAI-compatible endpoints validate again: relays that send an empty tool_calls: [] field no longer break stream completion (from upstream)
• Re-authenticating a ChatGPT or GitHub Copilot connection now refreshes its model list, so entitlement changes show up without a restart (from upstream)
• MCP tools with dots in their names (e.g. pat.batch_plan) work on OpenAI/Codex-backed connections (from upstream)
• Windows: image previews no longer fail by doubling the workspace path in front of absolute C:\ paths (from upstream)
• Windows: a gitBashPath configured in config.json is honored by the agent's Bash tool, fixing "No bash shell found" for per-user Git installs (from upstream)
• Explore mode no longer goes silent when a tool call is blocked — since the SDK 0.3.220 uplift the first blocked tool ended the turn before the agent could respond. The agent sees the block reason again and can suggest switching modes or submit a plan instead of stopping without a word (from upstream)
• A stale server lock whose process ID was recycled by the OS no longer prevents the app from starting — the lock now records which executable wrote it, and startup verifies the live process actually matches (from upstream)