Two lines of YAML:
uses: omrdev1/[email protected]
with:
plan-file: plan.json
github-token: ${{ secrets.GITHUB_TOKEN }}
No AWS credentials. No SaaS signup. Works in air-gapped envs.
The Principle (Before the Tool)
Every AI step that causes a side effect must have:
1. A defined success boundary
2. A compensating action (undo)
3. A timeout
4. A deterministic rollback order
This is the Saga pattern. AI workflows need it by default.
Why Existing Tools Fail:
Workflow engines ≠ transactional safety. Queues retry too much. State machines don’t undo side effects. Timeouts are bolted on.
None of this answers the core question: “What state is the system in after failure?”
I deeply regret my participation in the board's actions. I never intended to harm OpenAI. I love everything we've built together and I will do everything I can to reunite the company.
The best book I’ve ever read, and yet I feel like I’ve only touched the surface.
I will reread this book every year so that I can continue to learn from it.
It’s jam packed with incredible knowledge.
Thanks @DavidDeutschOxf for writing this and @naval for recommending it.
"I think business networking is a complete waste of time. The reality is if you’re building something interesting, you will always have more people who will want to know you."
@naval