Your agent needs a task handoff document — here's the template that prevents dropped work

Your agent finishes a task, you close the chat, and three days later you realize it never actually shipped the final piece. Sound familiar?

The problem isn't your agent's memory or intelligence. It's that you're treating task completion like a conversation ending instead of a formal handoff.

Every task needs a handoff document. Not a chat summary — an actual deliverable that captures what got done, what's left, and what you need to know to pick it up later.

Here's the template I use:

## Task Handoff: [Task Name]
**Date:** [Today's date]
**Agent:** [Agent name/session]
**Status:** [Complete/Partial/Blocked]

### What Got Done
- [Specific deliverable 1]
- [Specific deliverable 2]
- [File locations, URLs, commit hashes]

### What's Left
- [Next concrete step]
- [Known blockers or dependencies]
- [Estimated effort remaining]

### Context You Need
- [Key decisions made]
- [Approaches that didn't work]
- [Important constraints or requirements]

### Files/Resources Modified
- [Full file paths]
- [URLs or account info]
- [Credentials or access notes]

### Next Session Starter
"[Exact prompt to resume this work]"

The magic is in that last section. Your agent should write the exact prompt that would let you (or another agent) pick up exactly where it left off.

Instead of ending with "The API integration is mostly done," you get:

Next Session Starter: "Continue the Stripe integration in /src/payments/stripe.js. The webhook endpoint is built and tested, but we still need to add error handling for failed charges and update the user dashboard to show payment history. The test API keys are in .env.example."

This isn't just documentation — it's a forcing function. If your agent can't write a clear handoff, it probably doesn't actually understand what it accomplished.

I make this automatic by adding this to every agent's instructions:

When a task reaches a natural stopping point, always create a handoff document using the Task Handoff template. Save it to HANDOFFS/[date]-[task-name].md and present the key points in our conversation.

The result? No more "what was I working on" moments. No more rediscovering the same problems. No more half-finished work that never gets shipped.

Your agent becomes accountable for real completion, not just conversation completion.

If you're running coding sessions that span multiple days or involve complex multi-step work, you need this discipline built into your workflow.