Hand off a task to Claude Code

Connect Claude Code to your workspace, hand it a task, watch the loop run, grade the output. Twenty minutes, most of it Claude working.

Before you start

  • A task in Ready with at least one acceptance criterion (why that matters)
  • Claude Code installed and signed in

1. Connect

bash
claude mcp add --transport http masaro https://masaro.ai/api/mcp/mcp

First use triggers OAuth: browser opens → sign in → consent screen (pick a workspace if you have more than one) → approve. Once per client; revocable from Masaro any time.

Verify: run /mcp inside Claude Code; masaro should show as connected.

2. Hand off

In your repo, in Claude Code:

text
> Get the next task from Masaro and work it. Report progress as you go,
> and raise a blocker instead of guessing if anything's ambiguous.

That last clause matters. You’re setting the rule of engagement: ambiguity is a blocker, not a coin-flip. Masaro’s tooling backs it: report_blocker pings you with specific questions instead of the agent quietly picking a lane.

3. Watch the board, not the terminal

Switch to Masaro. You’ll see, in real time: the task jump to In Progress, claimed by the agent; then progress updates land on the activity feed as it clears real chunks (“schema migration written and applied”, “core path passing tests”). You don’t need to babysit the terminal; the board is the status report. Go do something else.

4. The hand-back

When Claude finishes, it calls complete_task with a summary and a per-criterion verdict, and the task moves to In Review. If it got stuck instead, you’ll have a blocker notification with concrete questions. Answer in a comment and tell Claude to continue.

5. Grade the work

Open the task in review. Walk the criteria: met or not, checked against the actual diff and behavior. This is deliberately not a vibe-check: the contract you wrote up front is doing the reviewing with you.

  • All met, code’s right → Done. Anything the completion unblocked is now waiting in Ready.
  • Something’s off → comment precisely (“criterion 2: the 429 is missing Retry-After”), move it back, and hand it off again. The feedback rides along with the task.

Where this gets good

One hand-off is a demo. The habit (a Ready column agents drain while you review and steer) is the product. Three tasks queued before lunch; three hand-backs to grade after. Reviewing agent work →