v3.5.0
AI Shelf 3.5.0
AI Flow advances to a full in-app workflow with scheduler, DAG inspection, and MCP-backed runner execution.
Added
- AI Flow end-to-end workflow — Added flow authoring, runner settings, source/output dialogs, run history/detail, and dedicated Flow tab UX for staged execution.
- Flow scheduling and execution engine — Added shared flow parsing/runtime modules, Electron flow service/scheduler handlers, and CLI flow commands for generate/run/status workflows.
- DAG and command visibility — Added DAG visualization with node detail and next-run visibility, including agent CLI command inspection from flow nodes.
- Flow system skills and runner integrations — Added always-output system skill injection, MCP runner resolution/config handling, and safer agent tool argument shaping.
- Usage insights for Cursor — Added billing-cycle Cursor spending provider and remaining allowance display in usage views.
Changed
- Flow defaults and settings UX — Unified per-flow settings editing in dialog, moved scheduled-run controls into flow settings, and defaulted Claude agent model toward haiku for generated flows.
- Documentation visuals — Refreshed README/docs screenshots and terminal demo GIF for current UI.
Fixed
- Flow scheduler correctness — Fixed cron minute matching and in-app schedule matching so due runs trigger at the expected times.
- Flow theme/readability — Aligned Flow UI colors with global theme tokens for consistent contrast.
- Clipboard reliability and notifications — Improved clipboard write verification/retry and reduced pane-agent ready notification spam.
Full changelog: CHANGELOG.md