CLI · Best Practices · AI · Workflow
Mastering VEM CLI: A Guide to 10x AI-Assisted Development
Learn the task-driven, agent-first workflow that keeps your AI agents perfectly aligned with your codebase context.
The Core Philosophy: Task-Driven Development
Power users never start by coding. They start by defining the Task. In VEM, a task is more than a todo item; it is a context anchor.
By adding validation steps up front, you are explicitly telling the CLI (and eventually your AI agent) exactly what 'done' looks like. This prevents the 'I think I finished' hallucination that often happens with AI.
Create a task with built-in validation
vem task add "Implement user profiles" --validation "pnpm build, pnpm test"The Hero Feature: vem agent
The vem agent command is the single most important tool in your belt. Instead of opening Claude or Gemini and copy-pasting code, you let VEM wrap the agent session.
What happens under the hood? VEM pulls the latest project memory from the cloud, generates task-specific context files, and launches your agent tool with these files auto-injected.
Launch an agent tied to a task
vem agent --task TASK-001Closing the Loop: vem_update & finalize
The biggest failure in AI-assisted work is forgetting to record why a change was made. VEM enforces 'Strict Memory' via the vem_update block.
When your agent finishes a task, it should provide a structured update block. The CLI detects this automatically and prompts you to apply it. If you miss the prompt, use the finalize command.
Apply memory updates manually
vem finalizeSemantic Discovery: ask and search
Stop using grep to find out why a weird bug exists. VEM indexes your commits, decisions, and tasks semantically.
Ask questions about your project's history or search for specific architectural decisions without needing to know the exact keywords.
Query your project memory
vem ask "Why did we switch from Clerk to Supabase?"
vem search "auth model"The Pro-User Daily Workflow
If you want to maximize your value, follow this 5-minute cycle as a habit:
- Define: vem task add your goal.
- Work: vem agent --task <id> to do the heavy lifting with AI.
- Verify: Complete the validation steps prompted by the CLI.
- Finalize: Apply the vem_update block to update your local memory.
- Push: vem push to sync your progress to the cloud.