AI-Assisted Workflow
Relevant source files
The following files were used as context for generating this wiki page:
Augenblick is a specialized note-taking application designed with an "AI-assisted" development philosophy, often referred to as "vibe coding." This approach prioritizes rapid iteration and functional parity through direct communication with AI agents rather than traditional manual software development cycles. The project itself was built 100% using AI tools like Codex CLI, Claude Code, and Cursor.
The workflow centers on a robust "Agent Protocol" that defines how AI agents interact with the codebase, manage dependencies, and execute complex migrations, such as the transition from Svelte to React. This system ensures that the application remains maintainable and secure while being evolved almost entirely by autonomous or semi-autonomous agents.
Sources: README.md:52-56, AGENTS.md:1-10
Agent Protocol and Guardrails
The development environment is governed by a strict protocol defined in AGENTS.md. This protocol minimizes token usage through "telegraphic" communication and prioritizes the use of "Agent Skills" (like React or Convex best practices) over external web searches.
Key Operational Guardrails
AI agents must follow specific safety and structural rules to maintain project integrity:
- File Management: Files are kept under 400 lines of code to facilitate readability and refactoring. Deletion requires explicit consent, and the
trashcommand must be used instead of permanent removal. - Dependency Management: New dependencies require a "health check" focusing on recent activity and adoption rates.
- Error Handling: The system prefers explicit result types over
try/catchblocks for internal logic, though exceptions are allowed for external systems like Git or GitHub.
Sources: AGENTS.md:20-33, AGENTS.md:46-51
Git and CI Integration
Agents utilize the GitHub CLI (gh) for PRs, issues, and CI monitoring. Commits follow the Conventional Commits standard, and agents are instructed to verify tasks through CI runs (gh run list/view) until they reach a green status.
This diagram illustrates the standard cycle an AI agent follows when implementing features or fixes.
Sources: AGENTS.md:35-44, AGENTS.md:53-57
Migration and Feature Parity Workflow
The AI-assisted workflow is heavily utilized for large-scale architectural migrations. A primary example is the "Augenblick → React migration plan," which utilizes the AI to ensure feature parity between the legacy Svelte implementation and the new React/Tauri 2 stack.
Migration Phases
The workflow divides complex migrations into distinct, AI-executable phases:
| Phase | Focus | Key AI Tasks |
|---|---|---|
| 0: Baseline | Capture | Record screenshots and behavior specs of the existing app. |
| 1: Backend | Rust Parity | Port Tauri commands, capabilities, and database logic. |
| 2: Scaffold | Frontend Setup | Initialize Tailwind v4 and shadcn/ui v4 components. |
| 3: Data Layer | Stores | Implement Zustand stores and API wrappers for Tauri invoke. |
| 4: UI Build | Components | Port Sidebar, Editor (Tiptap), and Command Palette. |
| 5: Integration | Tauri Events | Wire menu events and window-close prompts. |
Sources: plans/migration.md:113-200
Technical Implementation logic
The AI is instructed to maintain specific data models and persistence logic during transitions. For instance, the transition of the editor from Milkdown to Tiptap involves removing custom ProseMirror plugins in favor of standard Tiptap extensions to reduce technical debt and complexity.
Editor Logic Flow
This sequence shows how the AI agent structures the data flow between the editor and the state management layer.
Sources: plans/migration.editor.tiptap.md:38-60, plans/migration.inventory.md:100-110
Decision-Making and Autonomy
The workflow includes a "Completion and Autonomy Gate." Agents are programmed to assume a "continue" state unless explicitly stopped. To minimize friction, agents are empowered to choose "opinionated defaults" when faced with ambiguity, documenting these choices in a dedicated /docs/choices.md file rather than blocking progress with questions.
Autonomy Checklist
Before ending a turn, an agent must privately verify:
- Was the initial task completed?
- Were all definition-of-done items verified?
- Is the pending question truly blocking, or can a default be chosen?
Sources: AGENTS.md:65-80
The AI-assisted workflow in Augenblick transforms the development process into a high-level orchestration of specialized agents. By codifying protocols for communication, migration, and autonomy, the project maintains a rapid development pace while ensuring that the resulting code follows modern best practices for React, Rust, and Tauri.