D
DevcoreAI

Core Workflows

Plan & Act Modes

Think first, then build. DevcoreAI's dual-mode system for structured development.

DevcoreAI operates in two modes: Plan and Act. Switching between them gives you precise control over when the agent thinks versus when it executes. You can even use different AI models for each mode — a cheaper, faster model for planning and a more capable model for implementation.

Plan Mode

Explore and strategize. DevcoreAI reads files, asks clarifying questions, identifies risks, and produces a detailed implementation plan. No code is written during planning.

Act Mode

Execute and build. DevcoreAI writes files, runs commands, installs packages, and iterates on errors. Every file change is shown as a diff for your review before it's applied.

Plan Mode

Start in Plan mode when you're dealing with unfamiliar code, complex architectures, or tasks that could have far-reaching side effects. Plan mode is exploration without risk.

Best used for:

Exploring an unfamiliar codebase before making changes
Discussing architecture decisions and tradeoffs
Identifying edge cases and potential issues upfront
Creating a clear implementation strategy for large features
Reviewing and understanding complex workflows

Switch to Plan mode

Click the mode selector in the DevcoreAI panel → Plan

Act Mode

Act mode is where work gets done. DevcoreAI has access to all tools: file read/write, terminal execution, browser, and MCP tools. Every destructive action (file write, command execution) requires your approval unless auto-approval is enabled.

Act Mode · claude-sonnet-4-5

> Add user authentication to this Express app

✦ Reading src/app.ts, src/routes/...

✦ Creating src/middleware/auth.ts

+ JWT validation middleware (12 lines)

✦ Editing src/routes/users.ts

+ Added auth guard to protected routes

✦ npm install jsonwebtoken @types/jsonwebtoken

✓ Done · checkpoint saved · 3 files changed

Typical Workflow

1

Start in Plan mode

Describe what you want to build. Let DevcoreAI explore the relevant files and understand the codebase.

2

Discuss the approach

Ask questions, consider alternatives, and identify edge cases. The agent will ask clarifying questions too.

3

Switch to Act mode

When you're satisfied with the plan, switch to Act mode. DevcoreAI implements the solution based on your planning session.

4

Review and approve

Inspect each file change as a diff. Accept, modify, or reject individual edits before they're saved.

5

Iterate

If something fails, DevcoreAI reads the error and fixes it automatically. Checkpoints ensure you can always roll back.

Using Different Models for Each Mode

You can configure a different AI model for Plan mode and Act mode. A common pattern is to use a fast, cheap model for planning and a more capable model for execution.

Plan mode modelgemini-2.0-flash
Act mode modelclaude-sonnet-4-5

Settings → Model Configuration → Enable separate Plan/Act models

Deep Planning with /deep-planning

For large, complex tasks, use the /deep-planning slash command to activate an exhaustive planning mode that:

  • Explores the codebase systematically, reading all relevant files
  • Identifies all affected files and their dependencies
  • Creates a detailed, step-by-step implementation plan
  • Asks clarifying questions before generating a single line of code
Deep Planning guide

Choosing the Right Approach by Task Size

Small tasks

Act only

Fix a bug, add a utility function, write a test

Act mode only — describe the task and let the agent execute directly.

Medium tasks

Plan → Act

New feature, refactor a module, API integration

Plan → Act — spend a few exchanges planning, then switch to Act for implementation.

Large tasks

/deep-planning

Architectural change, new service, major refactor

Use /deep-planning for exhaustive codebase analysis before any code is written.