Skip to content
Back to Blog

How-to · Workflow · Teams

The full vem workflow: from task intake to verified memory

February 1, 202610 min read

A practical, end-to-end workflow for teams using vem across CLI, agents, and the web dashboard.

Workflow map connecting tasks, agents, and Git
A single workflow that keeps humans, agents, and Git aligned.

Start with a verifiable task

Every workflow begins with a task that is durable, trackable, and verifiable. Use the CLI or MCP to create tasks so they receive a stable TASK-XXX ID and land inside .vem.

Add validation steps up front to make sure every completion is backed by a real check, like a build or test run.

Create a task with validation steps

vem task add "Improve email audit log UI" --priority high --validation "pnpm build, pnpm test"

Kick off work in the CLI

The CLI is the authoritative write path for memory artifacts. Start the task, collect context, and keep task context synchronized while you work.

When you launch an agent through the CLI, vem injects task context and asks for evidence and validation steps before completion.

Move a task into progress

vem task start TASK-101

Launch an agent with context

vem agent --task TASK-101

Sync context before work begins

Before work starts, capture a snapshot so agents and teammates see the same context. The snapshot becomes the shared contract for the session.

Use vem pack for manual workflows or MCP tools for real-time agent access.

Pack context for an agent

vem pack

Agents complete work with signatures

Agents should always sign their updates so the UI can attribute who completed a task. That signature is captured in the task action history alongside evidence.

When tasks include validation steps, agents confirm each step before closing the task.

  • Include actor on completion
  • Provide evidence tied to validation steps
  • Keep reasoning concise and specific

Close the loop with evidence

Completing a task requires evidence. Evidence can be a file path, test command, or a short log excerpt.

vem treats evidence as a guardrail against truth drift: it becomes part of the permanent audit trail.

Mark a task done with evidence + actor

vem task done TASK-101 --evidence "pnpm build" --reasoning "Updated audit log UI states" --actor "Claude Desktop"

Push pending snapshots, verify in Git

Once the work is done, push a snapshot so the cloud index can serve it immediately. The snapshot is marked pending until the associated code lands in Git.

Verification happens when the Git webhook confirms the commit. That final step turns memory into the canonical, verified source of truth.

Send the snapshot

vem push

Review in the dashboard

The web app surfaces validation steps, evidence, and actor signatures so reviewers can confirm what shipped. It also centralizes org settings, seats, and security policies.

Use the task detail view to audit evidence quickly and spot any missing validation, and keep a weekly cadence of reviewing drift, redactions, and task hygiene.

  • Task detail view for evidence + validation
  • Drift panel for suspicious completions
  • Project settings for redaction reports and GitHub sync

Repeat the workflow as a habit

Teams that stick to the full workflow see faster onboarding, fewer regressions, and more reliable agent output.

Make validation steps explicit, keep tasks updated, and treat memory updates like production code.

  • Define validation steps for critical tasks
  • Require evidence on completion
  • Verify memory with Git commits