D
DevcoreAI

Core Workflows

Auto-Approve & YOLO Mode

Grant DevcoreAI permission to act on specific operation types without stopping to ask. Start conservative and expand permissions only as needed.

How It Works

By default, DevcoreAI stops before every file write, terminal command, browser action, and MCP call to show you a diff or confirmation prompt. Auto-approve lets you grant standing permission for specific categories so the agent can proceed immediately.

Permissions are configured in the DevcoreAI settings panel (gear icon in the sidebar). Each toggle is independent — you can allow file reads but still require approval for commands.

Permissions

Permission

Read project files

Allow reading any file in the workspace. Recommended — enables full codebase awareness.

Edit files (workspace)

Allow writing files inside the workspace without diff approval prompts.

Edit files (outside workspace)

Allow writing files anywhere on the filesystem. Use with caution.

Execute safe commands

Run commands classified as read-only or low-risk (e.g. npm test, git status, ls).

Execute all commands

Run any terminal command, including destructive ones. Requires explicit trust.

Browser actions

Allow the agent to launch a browser, navigate URLs, and interact with pages.

MCP tool calls

Allow calling MCP server tools without individual approval.

✓ Generally safe to enable · ⚠ Enable only for trusted workflows

Safe vs. Approval-Required Commands

Safe commands

Auto-approved when "Execute safe commands" is on:

npm run buildnpm testgit statusls -lacat package.json

Requires approval

Always pauses unless "Execute all commands" is on:

npm install <pkg>rm -rf <path>mv <a> <b>git pushsed -i ...

YOLO Mode

YOLO mode enables all permissions at once: file writes anywhere, all terminal commands, browser actions, and MCP tools — no prompts. The agent runs continuously until the task is done or an error stops it.

What gets auto-approved in YOLO mode

All file operations anywhere on the filesystem
All terminal commands including destructive ones
Browser navigation and interactions
All MCP server tool calls
Mode transitions (Plan → Act)

Best practices for YOLO mode

Use isolated environments or throwaway projects first. Be specific with your request — vague instructions with unlimited permissions produce unpredictable results. DevcoreAI checkpoints are your safety net.

Enable YOLO mode in DevcoreAI Settings → Auto-Approve → toggle Enable YOLO Mode. You can also enable OS-level notifications so you know when the agent is waiting or when a long command finishes.

Recommended Starting Config

Read project files

Always on — needed for full codebase awareness

ON

Edit files

Leave off until you trust the task scope

OFF

Execute safe commands

Enable for CI-like workflows

OFF

Execute all commands

Enable only in sandboxed environments

OFF

Browser / MCP

Enable per-task as needed

OFF