A model-neutral Markdown workflow for persistent project knowledge maintained with help from LLM agents.
This public repository contains portable rules and templates only. Private facts, sources, decisions, code, and paths stay in separate project instances.
- Create a private entry page from project-entry.md.
- Create its canonical status page from status.md.
- Add only project routes and project-specific constraints.
- Follow the Fast Path:
entry → status → at most one relevant page.
No registry, skill installation, custom agent, fixed directory tree, or copied protocol file is required.
| Place | Owns |
|---|---|
| This repository | Portable read, status, write-back, template, and publication rules. |
| Private instance | Private sources, facts, routes, decisions, and project constraints. |
| Optional adapter | Host-specific skills, retrieval, automation, or multi-instance lookup. |
Link across these roles instead of copying generic rules or private content.
- LLM_WORKFLOW.md: the small protocol Interface.
- BOOTSTRAP_NEW_INSTANCE.md: private-instance setup and smoke test.
- GPT_FIRST_PROFILE.md: concise GPT-family responses and optional ChatGPT Project instructions.
- DOCUMENT_LAYERING_AND_READ_POLICY.md: optional topic routing and source/digest split.
- SKILLS_INTEGRATION.md: optional host and retrieval Adapter contract.
- WORKFLOW_LINT_CHECKLIST.md: maintenance checks.
- PUBLICATION_GUARD.md: public/private review before publication.
Start with project-entry.md and status.md. Add the expanded layout, topic index, source digest, or meeting summary only when the project needs them. Remove unused placeholders.
- Skills repository: https://github.com/Counull/UniSkills.git
- Runtime skill: llm-workflow-maintainer
- Workflow repository: https://github.com/Counull/LLM-Workflow.git
Inspired by Andrej Karpathy's LLM Wiki.