Welcome 👋 — if you're reading this, you're part of the experiment.
Sefi is not "just another library." It's a semantic governor: a system that filters floods of agent emissions into a handful of actionable signals. Because it's experimental, your role isn't only to add code — it's to challenge the design.
If something feels unclear, backwards, or overcomplicated — say so. Your questions are as valuable as your pull requests.
Every line should be understandable. Boutique philosophy: if we don't know what it does, it doesn't belong.
Expect the architecture to evolve. We prefer tight loops of feedback to over-specified blueprints.
-
Read the README — that's the north star. Don't assume the details are final.
-
Ask first — open an issue or comment to probe assumptions.
-
PRs welcome — but small, clean, self-explanatory changes.
-
Tests are mandatory — TDD is our safety net. Every new component should ship with its own checks.
-
Language matters — be precise in code and in docs. If a metaphor (valley, ridge, peak) confuses, propose a better one.
- What concrete problem does this solve?
- Why this algorithm, not something simpler?
- How does feedback actually get consumed?
- What happens when thresholds are wrong?
- Is this efficient at scale, or just pretty in theory?
If you don't find yourself asking at least one of these, you may not be looking closely enough.
This is not a finished product — it's a living lab.
Your job is to question, refine, and stress-test the design.
If all you do is implement, you're missing half the fun.
💎 Think twice, code once.