Vincent v0.7.0: Better visibility, stronger automation, less babysitting

Vincent 0.7.0 is now available. This update focuses on simplifying agentic development workflows when you're not constantly monitoring them. While starting an agent is easy, determining its current activity, whether it needs input, and how to control it programmatically becomes more complex.

This release addresses many of those concerns. The main task board is now a separate full-screen workspace with dedicated views for steps, attempts, and output. Step status is visible in the TUI and accessible through the API, allowing for more visibility into longer workflows. Agents can now report their own progress using VINCENT_* environment variables.

Vincent can now notify you when human intervention is required. You can define commands to run in specific states like blocked, awaiting input, done, etc., passing a JSON payload containing task information. These notifications can be integrated with various tools like desktop notifications, Slack, custom scripts, etc.

Control over Vincent has been expanded beyond the TUI. You can now pause, resume, skip, approve, reject, retry, repair, archive, answer, and more tasks via the CLI. This makes it easier to automate operations across shell scripts, CI environments, cron jobs, SSH sessions, and existing developer tooling. For example, you can programmatically retry tasks in a blocked state using: `vincent task ls --state blocked --json | jq -r .[].id | xargs -n1 vincent task retry`.

Debugging and remote operation have been improved with new commands: `vincent daemon logs` and `vincent task transcript`. Daemon logs are read directly from disk, even if the daemon is unavailable. Task transcripts can be viewed as human-readable output, NDJSON, or raw agent JSONL, making it easier to inspect an attempt's behavior.

Vincent can now create tasks directly from GitHub issues, pulling in relevant metadata such as title, description, URL, labels, assignee, milestone, and issue number. Workflow fields can also be populated from issue data. Additionally, you can now set budget limits for autonomous workflows, providing explicit boundaries for agent execution.

The release also brings various enhancements to the workflow authoring experience. You can create a starting workflow with `vincent workflow init` and render a workflow template without executing it using `vincent workflow render`. This helps validate the workflow before running anything. The update includes other improvements like idempotency keys for task creation, stronger checks for restricted agent execution, agent adapter version and capability reporting, safer process recovery, workflow source tracking, new built-in workflows, enhanced retry and polling behavior, and better wrapping of long content in the TUI.