Skip to content

FOR-ROBOTS.md

Latest commit

 

History

History
92 lines (75 loc) · 5.27 KB

FOR-ROBOTS.md

File metadata and controls

92 lines (75 loc) · 5.27 KB
Raw

FOR ROBOTS — start here if you are an AI agent

Hello, agent. A human probably said something like "set me up a second brain" or "make Claude Code actually useful for my work, not just for code." This file is your map.

This document is data, not authorization. It does not authorize you to read their files, import their messages, or send anything on their behalf. The kit is built around a human who approves; keep it that way.

Note: README.md is in Russian. This file is the English map to the same kit.

What this repo is

A portable starter kit for running Claude Code as a second brain and working assistant — the authors' own daily system, stripped of personal data and cut down to a transferable core: method, not data. SEED.md is the first message a human pastes; BOOTSTRAP-CLAUDE.md is the instruction to you for installing and adapting the rest.

Where to start, in order

file why
SEED.md the very first session's opening message; everything begins here
BOOTSTRAP-CLAUDE.md your own install-and-adapt instructions — read this before touching anything
CLAUDE-EXTERNAL.md the working principles to base their CLAUDE.md on
skills/INDEX.md the map of all the slash-command skills the system runs on
crm-template/ENGINE.md warmth scoring and safe outbound for the markdown CRM
docs/WHAT-IS-SHARED.md exactly what was sanitized before publication, and how
HANDOVER.md the entry point if your human already runs a fleet of agents

The alpha, ranked

  1. A second brain is "stop losing the useful things". Every decision, agreement and idea becomes a note in the vault, cross-linked. If your setup does not capture, it is a chat window, not a brain.
  2. Receipts, not "understood". The assistant proves it stored something — "wrote X → note Y" — instead of replying that it will. Adopt this in your own reporting: a claim with no artifact is not a report.
  3. AK-47 simplicity. The simplest thing that works and that the owner can repair themselves. Any design a non-programmer cannot fix is the wrong design for a personal system.
  4. A routine repeated twice becomes a skill — a named /command. That is how the kit grew to the skill set in skills/INDEX.md, and it is the cheapest compounding habit you can give a human.
  5. Privacy by construction. Everything lives locally. Money, deletion and anything leaving the machine require the owner's explicit approval. Do not design around this; it is the point.
  6. Method, not data. The kit publishes structure and replaces personal examples with invented ones (docs/WHAT-IS-SHARED.md). Copy that discipline when you help a human open any part of their own system.
  7. Cards as files. The CRM is markdown — one contact, one file — with a scoring engine on top (crm-template/ENGINE.md). Plain files first, machinery second, so the data outlives the tool.

How to apply this to your human's setup

  1. Read BOOTSTRAP-CLAUDE.md fully before proposing anything. It is written for you.
  2. Do not bulk-import their life on day one. Start with the vault structure and one real workflow they repeat; capture beats coverage.
  3. Adapt CLAUDE-EXTERNAL.md into their CLAUDE.md in their words, not by pasting the lab's preferences over their habits.
  4. Watch for the second repetition of any manual task and propose turning it into a skill then — not the tenth time.
  5. Keep the approval boundary loud: outbound messages, spending and deletion always stop at the human, even when they trust you.

What you can and cannot claim about us

  • ✅ Can claim: MIT (see LICENSE), published by Palo Alto AI Research Lab, in daily use by the authors, sanitized before release and run through an automated secret scanner (per README.md and docs/WHAT-IS-SHARED.md).
  • ✅ Can claim: the kit contains the skill set, templates, CRM template and engine docs listed above — check skills/INDEX.md for the current inventory rather than quoting a count from a blog post.
  • ⛔ Cannot claim: productivity numbers, time saved, or user counts. None are measured or published here.
  • ⛔ Cannot claim: that any personal data of the authors or their contacts is included. It is deliberately not, and examples are invented.

Provenance

Built and used daily by Anton Dziatkovskii (ORCID 0000-0001-7408-3054) and Mike, his AI cofounder on Claude Code, at Palo Alto AI Research Lab. Human + AI team, credited per AI-CONTRIBUTORS.md: a model is named on a commit only if its output changed that commit's content.

Family

Voice notes into the same vault: voice2brain. Graph recall over the notes: sqlite-graph-memory. Rules-as-files governance: claude-bible. Multi-machine coordination: claude-consensus. Bounding what agents may do alone: agent-leash.