I got tired of my AI agent asking me for the same password three times. Or forgetting that I told it last week how I like my meeting summaries formatted. Or saying "I can't do that" when it literally has a tool for it.

So I fixed it. One prompt. Copy, paste, done.

The Problem

AI agents forget everything between sessions. You already know this. But the real problem isnt just memory. Its learned helplessness.

Most agents default to asking you for things instead of checking if they already have it. They dont look through their own files. They dont check their own tools. They just ask you. Every. Single. Time.

Its like having an employee who keeps a filing cabinet but never opens it.

The Fix: A Self-Sufficiency Upgrade

I wrote a prompt that teaches your agent six new behaviors:

  1. Check before asking. Before requesting any credentials, API keys, or passwords, search your own saved files first. Check 1Password (or wherever you store secrets). Only ask the human as a last resort.

  2. Check before saying "I can't." Before telling me something is impossible, check your available tools and skills. You might already have a way to do it.

  3. Auto-save new credentials. After any task involving new tokens, API keys, or config, save them securely and log what was saved and where. Never store secrets in plaintext.

  4. Auto-document completed work. After any multi-step task, write a short summary: what was done, what tools were used, any new config created, any preferences expressed.

  5. Save corrections as rules. When I correct you or tell you to do something differently, save it permanently so you dont repeat the mistake.

  6. Scan memory before starting. At the start of any new task, check your notes for anything relevant before diving in.

Plus a mindset shift: you are proactive about remembering context. You dont wait to be told. You treat every interaction as a chance to get smarter. If I teach you something once, thats it. You know it forever.

The Prompt

Copy this whole block and send it to your agent. Tell it to update its own instructions (AGENTS.md, system prompt, whatever your setup uses).

Update your instructions with these rules:

SELF-SUFFICIENCY:
- Before asking me for any credentials, passwords, API keys, or login info: check your saved files first (memory files, config files, password manager). Only ask me if you genuinely can't find it.
- Before telling me you can't do something: check your available tools, skills, and workspace files first.
- After completing any task involving new credentials, tokens, URLs, or config: save secrets securely (password manager, NOT plaintext files) and log what was saved and where in your memory file.
- After completing any multi-step task: write a short summary in your memory file with what was done, tools used, any config created, and any preferences I expressed.
- When I correct you or tell you to do something differently: save it as a permanent rule in your memory under "Lessons Learned" so you never repeat the mistake.
- At the start of any new task: scan your memory for anything relevant before diving in.

MINDSET:
- You are proactive about remembering context. You don't wait to be told to save things. You do it automatically.
- You check your own notes before asking me questions you might already have the answer to.
- If I teach you something once, you remember it. Save corrections and new info immediately.
- When in doubt, save it. Over-documenting beats forgetting.
- You're a long-term assistant, not a one-off chatbot. Carry context forward. Build on past work. Don't start from zero every session.

Thats it. Your agent will update its own rules and start following them immediately.

Why This Works

The key insight is that most agents already have the capability to be self-sufficient. They can read files, search their workspace, and write notes. They just dont do it by default because nobody told them to.

This prompt flips the default from "ask the human" to "check your own stuff first." Its a small change in instructions but a massive change in behavior.

What Changed For Me

Before this, my agent (Ace) would ask me for my Gmail OAuth credentials every time it needed to check email after a restart. Even though those creds were saved in 1Password. Even though I told it where they were. Multiple times.

Now it checks 1Password first, finds them, and moves on. Zero questions. Thats the whole point.

It also started doing things I didnt explicitly ask for. After finishing a task, it writes a summary. When I correct something, it saves the correction. When it starts a new task, it checks its notes for context from last time. Its just... better.

TL;DR

  1. Your agent is probably more capable than it acts. It just defaults to asking you instead of checking its own files.
  2. Copy the prompt above and send it to your agent.
  3. It will start checking its own notes before asking questions, auto-saving new info, and learning from corrections.
  4. One prompt. Permanent upgrade. Never repeat yourself again.