Writing New Code
Using Verdent to write and generate code effectively
Learn how to write effective prompts, use @-mentions for context, and select the right model preset for code generation.
What You'll Learn
- Compose effective prompts for code generation
- Use @-mentions to provide context
- Leverage Desktop-specific features for parallel development
- Review generated code with Code Review
Prompt Composition
Effective Prompt Structure
For straightforward code generation:
Create a function that [does X]Example:
Create a function that validates email addresses using regexFor multi-step or complex code:
I need to [goal]. The context is [background].
Requirements:
- [Requirement 1]
- [Requirement 2]
Consider [constraints or preferences]Example:
I need to build a user authentication system. We're using Express.js with PostgreSQL.
Requirements:
- JWT-based authentication
- Refresh token support
- Password hashing with bcrypt
Follow the existing patterns in our auth middlewareDo:
- Be specific about desired behavior
- Mention relevant files or patterns
- Specify error handling expectations
- Include edge cases to handle
Don't:
- Use vague language ("make it better")
- Assume context from previous sessions
- Leave out important constraints
- Request multiple unrelated changes
Iterative Development
Build features incrementally:
Start Simple
Request the basic structure or core functionality first
Review and Refine
Review generated code, request specific improvements
Add Complexity
Layer in error handling, edge cases, validation
Test and Polish
Request tests, documentation, or final polish
Using @-Mentions
@-mentions provide context by referencing specific files, folders, or symbols.
Syntax
| Mention | Scope | Example |
|---|---|---|
@filename | Single file | @userService.ts |
@folder/ | Directory | @src/components/ |
When to Use @-Mentions
Look at @authMiddleware.ts and create a similar middleware for rate limitingRefactor @userController.ts to use the patterns from @productController.tsSearch @src/utils/ for any duplicate utility functionsDesktop-Specific Code Writing
Parallel Feature Development
Build unrelated features simultaneously in separate workspaces:
Create First Workspace
Click New Workspace in the Top Bar to create an isolated workspace for your first feature
Name the Workspace
Give the workspace a descriptive name (e.g., "feature-auth")
Start a Task
In the input box, describe what you want to build and send your message to start a new task
Create Additional Workspaces
Repeat the above steps for each additional feature you want to work on
Work in Parallel
Each agent works independently in its own workspace without conflicts
Review Changes
When a feature is complete, review changes
Rebase to main branch
Click Workspace Actions → Rebase to main branch to apply changes
Example: Build an authentication system in one workspace while another agent builds the dashboard UI in a separate workspace.
Fix urgent bugs without interrupting ongoing feature development:
Continue Feature Work
Keep your current feature work running in its workspace
Create Bug Fix Workspace
Click New Workspace in the Top Bar to create a separate workspace for the bug fix
Name the Workspace
Name the workspace descriptively (e.g., "fix-login-error")
Describe the Bug
In the input box, describe the bug and how to fix it to start a new task
Parallel Work
The bug fix agent works independently while your feature agent continues
Ship the Fix
When the fix is ready, click Workspace Actions → Rebase to main branch to ship the fix
Resume Feature Work
Your feature workspace continues uninterrupted with the fix now in master
Benefit: Ship critical fixes immediately without pausing feature development or losing context.
Try risky refactors or experimental approaches without affecting stable code:
Create Experiment Workspace
Click New Workspace in the Top Bar to create an isolated workspace for your experiment
Name the Workspace
Name it clearly (e.g., "experiment-new-architecture")
Describe the Experiment
In the input box, describe the experimental approach you want to try to start a new task
Let Agent Explore
Let the agent explore and implement the changes freely
Review Results
Review the results to see all modifications
Keep or Discard
If successful, click Workspace Actions → Rebase to main branch to keep changes. If failed, click Delete Workspace in the Workspace Bar to discard all changes.
Benefit: Your base workspace remains stable and unchanged regardless of experimental outcomes.
Plan Mode for Complex Features
For complex code generation:
- Press
Shift+TaborCtrl+.to switch to Plan Mode - Describe the feature you want to build
- Review the plan Verdent creates
- Refine with follow-up questions
- Switch to Agent Mode to execute the plan
Benefit: Review approach before consuming credits on execution.
Reviewing Generated Code
After Verdent generates code, use Code Review for quality assurance:
View Code Review
After Verdent completes changes, the Code Review panel appears inline within the chat
Address Critical Issues
Review Must Fix items first, these are security, performance, or stability issues
Consider Suggestions
Review Suggestions for recommended improvements you may want to apply
Request Fixes
Ask Verdent to fix any issues by describing what needs to change in the input box
View Diff
Click Task Changes in the middle panel to see all file modifications
Code Review analyzes:
| Category | Focus |
|---|---|
| Security | Vulnerability scanning, injection risks, credential exposure |
| Accessibility | Accessibility compliance and best practices |
| Performance | Algorithmic complexity, resource usage |
| Call-chain Impact | Downstream effects on dependent code |
Enable Auto Generate Code Review in Settings → Chat to automatically generate reviews after every change.
FAQs
How do I get more accurate code generation?
- Be specific - Include exact requirements
- Provide context - Use @-mentions for relevant files
- Specify patterns - Reference existing code as examples
- Iterate - Build incrementally rather than all at once
Should I use Plan Mode for all code writing?
Use Plan Mode for:
- Complex multi-file changes
- Unfamiliar patterns or codebases
- When you want to review approach first
Skip Plan Mode for:
- Simple, straightforward tasks
- Well-understood changes
- Quick iterations
Can I work on multiple features at the same time?
Yes. Create separate workspaces for each feature by clicking New Workspace in the Top Bar. Each workspace has its own agent working in isolation.
How do I undo changes Verdent made?
You don't need to delete the entire workspace to undo changes. You have several options:
- Rollback: Hover over any message in the chat and click Rollback to this point to revert to an earlier state.
- Use Source Control: Open Source Control (
Ctrl+Shift+G/Cmd+Shift+G) to view changed files and revert specific changes. - Delete workspace: Click Delete Workspace in the Workspace Bar to discard all changes in that workspace. Your base workspace remains unaffected.
Does Code Review use credits?
Yes. Code reviews consume credits based on the amount of code analyzed. Disable auto-review in Settings → Chat if you want to save credits.
Can Verdent generate tests for my code?
Yes. Ask Verdent to write tests by describing what to test and referencing files with @-mentions (e.g., "Write tests for @userService.ts").