IronPad-Docker/docs/ai-workflow/tools.md

Tools Used

Every tool in the Ironpad development workflow and how it fits in.

IDE

Cursor IDE

The primary development environment. Cursor is a fork of VS Code with built-in AI integration.

• Why: Direct integration with Claude models, inline code editing, multi-file context
• How used: All coding, file editing, terminal commands, and AI conversations happen in Cursor
• Alternative: VS Code + Copilot, Windsurf, or any editor with AI integration

AI Models

Claude Opus 4.5 (Anthropic)

Used for the majority of Ironpad's development (Phases 1-3 of the implementation).

• Context window: 200K tokens
• Strengths: Excellent at Rust code, understands Axum patterns well, good at architecture
• Limitation: Can only hold ~5 files at once; required task splitting and handover documents
• How used: Feature implementation, debugging, code review

Claude Opus 4.6 (Anthropic)

Used for later phases and the full codebase audit.

• Context window: 1M tokens (5x increase)
• Strengths: Can hold the entire codebase at once; finds cross-file bugs; handles complex refactors in one session
• How used: Codebase audit (found 16 issues), cross-cutting refactors, feature implementation without task splitting

Perplexity AI

Used for research before coding.

• Why: Has internet access, provides current information with citations
• How used: Checking library versions, finding known issues with crates, researching approaches
• Example: Verified that serde_yaml was deprecated before we chose to use it (accepted risk for v1)

Google Gemini

Used as a second opinion on architecture.

• Why: Different training data and perspective from Claude
• How used: PRD review, architecture review, catching blind spots
• Example: Flagged the need for file locking between Task View and Editor (race condition that Claude's initial design missed)

MCP Tools

Task Master

A Model Context Protocol (MCP) tool for structured task management.

• What it does: Parses a PRD and generates ordered task lists with dependencies
• How used: Fed the PRD to Task Master to generate the implementation plan; tasks tracked through completion
• Why it matters: Turns a document into actionable, sequenced work items

Context7

A Model Context Protocol (MCP) tool for pulling current library documentation.

• What it does: Fetches up-to-date documentation for any library and loads it into the AI's context
• How used: Pulled current Axum 0.8 docs, Vue 3 Composition API docs, Milkdown editor API docs
• Why it matters: Eliminates bugs caused by the AI using outdated API knowledge

Build Tools

Rust / Cargo

• Backend language and build system
• cargo check for fast compilation checks
• cargo build --release for production binaries
• Strict compiler catches entire categories of bugs before runtime

Node.js / Vite

• Frontend build tooling
• npm run dev for development with hot reload
• npm run build for production static files
• Vue SFC compilation via @vitejs/plugin-vue

Git

• Version control for all data files
• git2 crate for programmatic access from Rust
• Automatic 60-second commit batching
• Full diff viewer in the UI

The Tool Stack in Practice

Idea
  |
  v
[Perplexity] -- Research libraries, check feasibility
[Gemini]     -- Second opinion on approach
  |
  v
[Claude]     -- Draft PRD
[Gemini]     -- Review PRD
  |
  v
[Task Master] -- Generate ordered task list
  |
  v
[Cursor + Claude] -- Implement each task
[Context7]        -- Current docs when needed
  |
  v
[Manual Testing]  -- Verify in browser
[cargo check]     -- Compiler verification

No single tool does everything. The value is in how they compose.