A Markdown company brain for solo founders working with AI agents.
AI tools are powerful. Real AI work gets messy fast.
- 🤯 decisions disappear inside chat history
- 🤖 different AI windows do not know what others did
- 📝 finished tasks leave no durable worklog
- 🔁 useful workflows are taught again and again
- 🧪 AI says "done", but nobody checks the evidence
Solo-AI-Company-OS helps you turn scattered AI sessions into reviewable, reusable operating memory.
中文文档 · Start Here · AgentOps Doctor
Do not read the whole architecture first. Run one small loop:
Human decides.
AI executes.
Worklog remembers.
Handoff coordinates.
Skill evolves.
Doctor checks risk.
Markdown keeps the operating memory.
- Pick one real task.
- Choose an AI worker.
- Copy its
START_PROMPT.mdinto your AI tool. - Finish the task.
- Create a worklog.
- Review what should become a reusable skill.
That is enough to start building durable AI memory.
| Part | Purpose |
|---|---|
| 🧭 Founder Decision Log | Records human decisions, tradeoffs, and why they were made |
| 🤖 AI Employees | Defines AI worker roles, boundaries, and start prompts |
| 📝 Worklogs | Keeps task history, evidence, results, and open questions |
| 🔁 Handoffs | Lets agents, windows, and future sessions continue work |
| 🧰 Skills | Turns verified workflows into reusable operating knowledge |
| 📊 Dashboards | Summarizes current state without replacing source records |
| 🩺 AgentOps Doctor | Checks context, claims, evidence, and workflow risk |
This is not a prompt collection. It is an operating memory system for human-AI work.
Use this if you:
- work with Codex, Claude Code, Cursor, ChatGPT, Gemini, or other AI tools
- run projects across multiple AI windows or agents
- keep losing context between sessions
- want AI work to become reviewable, transferable, and reusable
- think of AI as a small team, not just a chat box
This may be too heavy if you only want a list of prompts.
Windows:
powershell -ExecutionPolicy Bypass -File .\INSTALL.ps1Cross-platform helpers:
python scripts/os-helper.py new-worklog --agent-id AI-01 --title "first setup loop"
python scripts/os-helper.py new-handoff --from-agent AI-01 --to-agent AI-02 --task-title "evidence check"
python scripts/os-helper.py doctor-example --case handoff-expiredThe scripts only create Markdown drafts. The real source of truth is still the operating memory you maintain.
Do not put credentials, private customer data, unpublished strategy, or raw founder reflections into public outputs.
Core rule:
Founder decisions outrank AI suggestions.
Verified truth outranks narrative.
Operating memory beats scattered chat history.