GPT Org-roam

General technical knowledge graph and tutorial repository maintained as Org-roam documents.

The repository is concept-first, not tutorial-only. Reusable concepts and source-grounded research form the backbone; architectures compose those concepts; tutorials provide long-form learning paths through the graph; projects anchor the ideas in concrete systems.

StarIntel-specific research, approvals, and design authority belong in the dedicated StarIntel research repository. Existing StarLang/StarIntel material may remain here as project and tutorial material, but it is not the authority for StarIntel design.

Repository layout

• concepts/ — focused reusable concepts, distinctions, invariants, and models.
• research/ — source-grounded research, evidence, synthesis, hypotheses, and open questions.
• architecture/ — system-level compositions and implementation models.
• tutorials/ — first-class long-form technical tutorials intended to be read from zero knowledge through practical use.
• projects/ — concrete project notes and entry points.
• index.org — top-level Org-roam navigation index.
• AGENTS.md — contribution, research, linking, verification, and merge contract.

Graph model

Prefer durable Org-roam links over duplicated prose. A tutorial can explain a concept in context, but reusable theory should live in a concept/research node and be linked by stable Org ID.

The initial general agent-systems backbone covers:

• frozen acceptance specifications;
• proof-carrying execution;
• symbolic verification with Prolog;
• symbolic agent memory;
• RLM-driven exploration;
• mutation testing as adversarial evidence;
• spec-carrying agent loops.

Personal LLM memory and Gemini Notebook

This repository also has a local-first LLM journal/export path. Raw provider history is not committed by default: ChatGPT conversations import into the gitignored private/ tree, where they can still participate in the local Org-roam graph.

python3 tools/llm_roam.py import-chatgpt ~/Downloads/conversations.json

Build a Gemini Notebook source set with:

python3 tools/llm_roam.py build-gemini
python3 tools/llm_roam.py verify-gemini

build-gemini automatically includes sibling starintel-auto-research and starintel-server checkouts when present. You can pin exact local checkouts instead:

python3 tools/llm_roam.py build-gemini \
  --repo auto-research=../starintel-auto-research \
  --repo starintel-server=../starintel-server

The generated _exports/gemini/ directory contains uploadable Markdown source packs, CODE-INDEX.csv, notebook instructions, and a hash manifest. StarIntel material is a read-only projection: Auto-Research remains the research/design authority, and StarIntel Server plus its schema lock remain runtime/schema authority.

See the Org-roam architecture node LLM Org-roam with Gemini Notebook Code Execution for the full model.

Contribution and merge policy

Changes should be made on a feature branch and submitted through a pull request.

Merge automatically only when all required CI/CD checks are green.

A missing, skipped, cancelled, pending, or failing required check is not green and must not be treated as approval to merge. If a repository change requires a new validation rule, add that validation to CI/CD rather than relying on a manual convention.

Documentation changes must preserve valid Org syntax, stable Org-roam IDs and links, source grounding where claims require it, and the navigability of index.org. See AGENTS.md for the full contract.