v1.4.0

Everyone gets the Tasks form changes, session export and the fix for runs left behind on quit. The two headline features are opt-in: the Credentials Checker is off until you turn it on, and task delivery does nothing until you add a destination to a task.

Added

• A scheduled task can send its output somewhere after every run. Add Slack channels, Telegram chats or email recipients in the task's new Delivery section, and choose whether to send on success, on failure or always. Slack gets a summary message with the output in its thread, and Slack channels are picked from a searchable list. Each run's delivery results show up in Jobs, and a failed delivery never marks the run itself as failed. (#626, #634–#641)
• Reply to a task's Slack delivery thread to carry on its chat. An @Agento reply under a delivered run resumes that run's session, the same way an inbound Slack mention does. (#642)
• The Credentials Checker finds API keys and secrets that ended up in your Claude Code sessions. It is a new Security → Credentials Checker page that groups findings by session, shows only a masked snippet, and lets you mark a finding as a false positive or whitelist a value or a whole rule. It is off by default; turn it on in Settings → Security, where the whitelist is also managed. It scans in the background, and the raw secret is never stored. (#597, #599–#606)
• Export a session as Markdown, JSONL or plain text from the session list's right-click menu or the session page's Export button. You choose what to include, and the file is saved wherever you pick. (#591)
• The Tasks inspector previews an unsaved task: its next three fire times, a plain-language schedule, and the model and working directory it will actually use. (#633)

Changed

• A task no longer needs an agent. No agent is now the default in the picker, and the form says exactly what such a run gets. You can create a task on a fresh install with no agents at all, and the Model field names the model that will really be used. (#627, #628, #629)
• The Tasks form is easier to scan. Its sections now run Task → Schedule → Execution → Limits → Delivery → Instruction. Execution and Limits collapse to a one-line summary, and the Instruction box fills the remaining height of the pane. (#630, #631, #632)

Fixed

• Quitting Agento now stops its in-flight agent runs instead of leaving Claude Code processes running with nobody listening. A run from a previous session that was left stuck as Running in Jobs is now marked failed at startup, with the reason given, and a process it provably left behind is stopped (on Windows too). (#593, #594, #595, #596, #613)
• Security fix in the update signing toolchain. The Tauri CLI used to build this release now binds each updater signature to the version it was released as, so a signed update cannot be passed off as a different version. Tauri itself and its plugins also move to their latest patch releases. (#659, #660)

Installing

Builds are neither Apple notarised nor Windows code signed, so the first launch of a downloaded build needs a Gatekeeper or SmartScreen bypass — see Installation. In-app updates are unaffected.