Skip to content

Instantly share code, notes, and snippets.

@gnapse

gnapse/README.md

Last active April 8, 2026 22:03
Show Gist options

  • Select an option

    Learn more about clone URLs
  • Save gnapse/e5628e6f219811e7afd46bb95098f0d6 to your computer and use it in GitHub Desktop.

Select an option

Learn more about clone URLs
Save gnapse/e5628e6f219811e7afd46bb95098f0d6 to your computer and use it in GitHub Desktop.
Download ZIP
AI personal assistant starter prompt
Raw
README.md

Starter prompt for building a repo-based personal assistant

Use this in Claude Code, Codex, or a similar coding-agent harness.

  • Paste starter-prompt.md into an empty folder and let the agent interview you.
  • Use connectors-and-adjacent-tools.md as a menu of systems worth connecting first, plus a few more advanced adjacent tools.
  • Prefer a private Git repository from the beginning so the assistant's memory stays inspectable and synchronized.
Raw
connectors-and-adjacent-tools.md

Useful connectors and adjacent tools

This is a menu, not a checklist. You do not need all of these. Two or three good connections are enough to make the assistant genuinely useful.

Good first connectors

  • Task manager: Todoist, Linear, Jira, or whatever system holds commitments and follow-ups.
  • Chat: Slack, Twist, or Microsoft Teams for inbox triage, thread summaries, and draft replies.
  • Google Workspace: Gmail, Calendar, Drive, Docs, and Contacts for day-to-day work context.
  • Code hosting: GitHub or GitLab for pull requests, issues, notifications, and repo context.
  • Meeting notes: Granola or another meeting-notes system for notes, transcripts, and follow-ups.
  • Internal docs: Notion, Outline, or Confluence for handbook content, project docs, and decisions.

Useful personal systems

  • Personal email: Fastmail, IMAP mailboxes, or another inbox outside work.
  • Finance: hledger, ledger, CSV exports, or bank statements for expense tracking and bookkeeping.
  • Notes and journal: markdown files, Obsidian vaults, or plain folders of documents.
  • Health and habits: simple trackers, spreadsheets, or exports from systems you already use.

Adjacent tools

  • qmd: useful once the repository has enough notes, transcripts, and docs that grep stops being enough. It adds local keyword, semantic, and hybrid search over your markdown collections.
  • OpenClaw: a more advanced next step if you want the assistant to live outside the coding-agent harness, for example on a cloud machine or a headless computer in your office, and be reachable from Telegram, Slack, or WhatsApp.
  • Local scripts: useful for deterministic pieces like imports, cleanup, normalization, and reporting.
  • MCP servers: useful when there is no good CLI or when a service is easier to access through a structured server.

How to prioritize

  1. Start with one memory-rich system and one action system.
  2. Prefer read-only access first.
  3. Prefer inspectable interfaces like CLIs, local files, and small scripts.
  4. Add chat gateways last.
Raw
starter-prompt.md

I want you to help me create a personal assistant for myself inside this empty folder.

Your job is not only to answer questions in chat. Your job is to build and maintain a working system for helping me with my real work over time.

This folder is going to become your memory and workspace. You should organize it in whatever structure makes sense for you, and you should keep that structure updated as you learn more about me and how I work.

This should also become a git repository. If it is not already one, help me initialize it. Encourage me to create a private GitHub repository for it, add it as a remote, and keep the system safely synchronized there over time.

Start by interviewing me. Ask me one question at a time. Your goal is to understand:

  • who I am and what kind of work I do
  • what recurring workflows I want help with
  • what external systems I use
  • what kinds of information I need to remember over time
  • what outputs would be most useful to me
  • what permissions and safety boundaries I want

In particular, ask me about the external systems I use for work and life: task managers, chat tools, calendar, docs, email, code hosting, notes, finance, health, or anything else relevant.

For each system I use, investigate whether there is:

  1. a good command-line interface
  2. an MCP server
  3. some other practical integration path

Prefer stable and inspectable integration paths when possible.

Do not wait for me to design the whole system up front. Propose a first version yourself, then improve it iteratively as you learn more.

As part of setting this up, you should create and maintain your own operating instructions for this environment. If this agent harness supports a project instruction file such as AGENTS.md, CLAUDE.md, or an equivalent, create or update it so future sessions know how this system works.

You should also create reusable skills for recurring workflows. If this harness supports a skills folder or skill installation mechanism, use the appropriate location and conventions for this environment. You are responsible for figuring out how this harness expects skills to be stored and maintained.

When a workflow has a mechanical or deterministic part, consider implementing that part in a small script and using the skill for orchestration, reasoning, and output formatting.

Your first phase is:

  1. interview me
  2. inspect this environment and determine what kind of agent harness you are running in
  3. propose the initial folder structure, memory model, and workflow model
  4. set up git and propose a private GitHub remote for this folder
  5. identify likely external integrations
  6. create the first version of your instructions and supporting skills

After that, continue improving the system as we use it.

One important rule: I should not need to manually maintain your memory. I will talk to you naturally, and you should take responsibility for persisting and organizing the information that matters.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment