Your AI doesn’t level up the longer you play. It quietly forgets the quest.
1. Start Fresh
Most people leave AI sessions running far too long.
It really… irks me when I see people in the Claude app have dozens of sessions dating back weeks and then going back to them because “that’s where I was working on the project last…”
As AI conversations stretch on, token usage climbs, the AI starts repeating itself, missing instructions, or hallucinating details it was getting right an hour ago. That’s the nature of AI. Autocompact is supposed to help, but in practice, it’s unreliable — and it quietly drops context you actually needed.
The fix is intentional session hygiene.
For long sessions, keep an eye on how much of your context window is used. When you hit about 70%, ask the AI to create a HANDOFF.md file — a structured document that captures everything from the session: what the goal is, what’s been done, what worked, what didn’t, and what’s left to do.
This also applies to if you’re finishing up for the day.
Then exit the session, start a fresh one, point it at the handoff file, and tell it to continue.
OR better yet, use a dedicated /handoff skill so the process is one command. The AI reads the conversation, writes the handoff, and tells you the file path. Fresh session, full context window, no degradation.
2. The Handoff Skill
So how do you actually build a /handoff skill?
You don’t. You steal an existing one, like mine:
---
name: handoff
description: Use when the user wants to create or update a handoff document for continuing work in a fresh conversation, or when saying "handoff", "hand off", "save context for next session", or "write a handoff".
---
# Handoff
Write or update a handoff document so the next agent with fresh context can continue the current work.
## Process
```mermaid
flowchart TD
A(["User invokes /handoff"]) --> B{"HANDOFF*.md exists?"}
B -- yes --> C["Read existing HANDOFF*.md"]
B -- no --> D["Gather context from conversation"]
C --> E{"Related to current work?"}
E -- yes --> D
E -- no --> D
D --> F{"Related or new?"}
F -- "related" --> G["Update existing HANDOFF.md"]
F -- "new" --> H["Create HANDOFF-N.md"]
G --> I["Tell user the file path"]
H --> I
I --> J(["Done"])
```
### Step 1: Check for existing handoff documents
Glob for `HANDOFF*.md` in the project root. If any exist, read them.
### Step 2: Determine related or new
If an existing `HANDOFF.md` (or `HANDOFF-N.md`) covers the same goal or work area as the current conversation, update that file. If the existing handoff is completely unrelated, create a new file with the next increment:
- No existing file: `HANDOFF.md`
- Existing `HANDOFF.md` is unrelated: `HANDOFF-2.md`
- Existing `HANDOFF.md` and `HANDOFF-2.md` are unrelated: `HANDOFF-3.md`
### Step 3: Gather context
Review the current conversation to understand:
- What the overall goal is.
- What has been accomplished.
- What approaches worked and which failed.
- What remains to be done.
### Step 4: Write the handoff document
Create or update the handoff file with these sections:
```markdown
# Handoff
## Goal
What we're trying to accomplish. Be specific enough that a fresh agent can understand the objective without prior context.
## Current Progress
What's been done so far. Reference specific files, commits, or changes made.
## What Worked
Approaches and decisions that succeeded. Include enough detail to preserve useful context.
## What Didn't Work
Approaches that failed and why. This prevents the next agent from repeating dead ends.
## Next Steps
Clear, actionable items for continuing. Order by priority. Be specific — file paths, function names, concrete tasks.
```
**Writing guidelines:**
- Write for an agent with zero prior context.
- Be concrete: file paths, function names, commit hashes, not vague descriptions.
- Keep "What Didn't Work" honest — include the reasoning for why it failed, not just that it did.
- Next Steps should be actionable without re-reading the entire codebase.
- If updating an existing handoff, preserve relevant history from prior sections and mark what's new.
### Step 5: Confirm
Tell the user the absolute file path so they can start a fresh conversation with:
> Handoff written to `/path/to/project/HANDOFF.md`. Start a new conversation and point the agent at that file to continue.You can either save it in your .claude/skills/ folder (or equivalent) as handoff/SKILL.md or ask Claude Code to set it up for you. One command, clean break, no lost context.
3. Courses
I have a range of online courses that teach business people what they need to know about productivity and AI:
1. Next Level Productivity
A practical, straightforward course that teaches you how to achieve elite-level personal productivity in today’s constantly interrupted world.
2. The AI Agents, Automations and Agentic Workflows Guide
This non-technical course shows business people and non-coders how they can build and use AI agents in ChatGPT and Zapier.
3. The Complete Claude, Claude Code & AI for Work Productivity
A comprehensive course taking you from beginner AI concepts like prompt and context engineering, to cutting-edge AI productivity using terminal-based AI tools like Claude Code for non-coding office work. Usable with ChatGPT, Claude, Gemini and other current LLMs.
4. The AI Playbook
This is my longer premium course on how businesses can deploy AI tools and technology across their processes and teams.
5. ChatGPT for Managers
See how AI can solve complex management challenges in less than 30 seconds. Full prompt library and examples included.
That’s it for this week!
— Aaron