Automatic 0.8.0: AI-Generated Instructions and Recommendations

Automatic 0.8.0 is out today. The headline is AI-powered instruction and recommendation generation — you can now ask Automatic to analyse your project and write its own configuration. But there's a lot more in this release across almost every part of the app.

AI-generated instructions and recommendations

The most significant change in 0.8.0 is that Automatic can now write its own configuration for you.

For instructions, that means pointing Automatic at a project and clicking Generate. It will inspect the project, reason about the codebase, and produce a tailored instruction file — the same kind of guidance you'd write by hand, but done automatically. It also watches for changes made outside the app, so you won't accidentally overwrite edits you made directly in your editor.

For recommendations, Automatic now suggests relevant Skills and MCP servers based on what it finds in your project. If you're in a Terraform repo, it might recommend the terraform-engineer Skill and a GitHub MCP server. Each suggestion includes a description you can expand inline, and installing is a single action. Recommendations recalculate every time you save your project, so they stay relevant as your work evolves.

Both features require an Anthropic API key. The first-run wizard now includes a dedicated step to set one up, with keychain storage so it's secure and an obfuscated hint so you know it's saved.

Project context via MCP

Automatic now exposes project context through a dedicated MCP tool: get_project_context. Your agents can call this tool during a task to retrieve structured context about the current project — its rules, instructions, and any metadata Automatic knows about it.

Context generation is AI-powered. Automatic analyses your project and produces a summary that's useful to agents without being bloated. The context panel in the app shows a live task log of what the generation process is doing, so it's not a black box.

Under the hood, context storage has moved from TOML to JSON. This is a breaking change for any tooling that reads Automatic's internal files directly, but the app handles the migration automatically.

Rules get their own tab

Rules have been promoted from a secondary concept to a first-class project feature. They now live on a dedicated Rules tab in the project view, alongside a custom rule editor and a dropdown for picking from global rules.

Inline custom rules per project are now supported. All write paths correctly route rules to .claude/rules/ as intended. The selected rule indicator has been updated — it now uses a subtle highlight instead of a solid brand fill, which is a lot easier to read when you have multiple rules active.

The recommendations copy has also been updated to point users to the Rules tab rather than the old location.

Dashboard and project overview improvements

The dashboard now shows a project health bar above the use cases section. It gives you a quick read on overall project health without having to dig into individual projects.

Project overview pages have been reworked with a two-column layout. A rules widget sits alongside an elevated setup callout so the most important actions are easy to find. Projects are now grouped by folder with compact cards, which makes the overview page much more manageable when you have a lot of projects open.

Task log

Task log entries are now persisted to disk, so they survive app restarts. A header toggle lets you show or hide the log, and copy actions make it easy to grab specific entries.

AI Playground

The built-in AI Playground now supports API key management and pulls a live model list so you're always working with current options. Two new tools have been added to the built-in agent: library read and marketplace search. Generate buttons are now gated on API key presence — the environment variable fallback has been removed.

---

Changelog

[0.8.0] — 2026-03-08

Onboarding

• Anthropic API key step added to the first-run wizard with keychain storage and an obfuscated hint display

Dashboard

• Projects health bar added above the use cases section

Projects

• Two-column summary layout with a rules widget and elevated setup callout
• Folder grouping with compact cards on the overview page

Context

• AI-powered project context generation with an integrated task log panel
• Project context exposed via a dedicated MCP tool (get_project_context)
• Context storage migrated from TOML to JSON; Tauri commands registered

Instructions

• AI generation for project instruction files with reactive recommendations
• Externally-modified instruction files are detected to prevent accidental overwrites
• Conflict modal simplified to a summary view with aligned DriftReport types

Rules

• Rules moved to a dedicated project tab with project-level scope
• Custom rule editor on the Rules tab with a dropdown global rule picker
• Inline custom project rules support
• Rules correctly routed to .claude/rules/ in all write paths
• Selected rule indicators replaced solid brand fill with subtle highlights

Recommendations

• AI skill and MCP server suggestions with a proper install flow
• Compact single-line rows with a collapsible description toggle
• Recommendations recompute on project save
• Rules recommendation copy updated to reflect the Rules tab

Task Log

• Task log entries persisted to disk
• Header toggle button and copy actions added

Agent / AI Playground

• API key management and live model list for AI Playground
• Library read and marketplace search tools added to the built-in agent
• Generate buttons gated on API key presence; environment variable fallback removed