Skip to content

cyberagent 0.1.0

Choose a tag to compare

View all tags
@CyberK13 CyberK13 released this 10 Jun 00:08
· 7 commits to main since this release
v0.1.0
d0b4692

The idea

cyberagent distills two bodies of work into one reproducible, falsifiable agent chain:

  • Leopold Aschenbrenner — Situational Awareness (the why). AI scaling is a
    massive industrial process, not a software one — bottlenecked by physical
    inputs: power, transformers, HBM, CoWoS packaging, specific materials. Effective
    compute compounds ~1 OOM per year; conviction is built from the physical
    trendlines before consensus prices them.
  • Serenity (@aleabitoreddit) & Crux — the
    bottleneck method     (the how). Don't ask "which stock goes up?" — take the
    machine apart and find the chokepoint. Which part of the BOM is hardest to
    replace? If one supplier stopped shipping tomorrow, how long would the whole
    chain wait?

The chain

Every asset runs through a fixed, falsifiable telescope — zooming from physical
reality down to an actionable name:

Positioning → Physical World → Human Development → Economics → Company Financials → Leaders & Verdict

executing the five-step method: physical bottleneck → uniqueness →
commercialization → financial elasticity → consensus correction.

The discipline

  • Real-time grounding — search why a price moved; never trust model memory.
  • Price-action guardrail — a parabolic move on one headline is AVOID / observe, never a buy.
  • Evidence ladder — every claim tagged Confirmed / Inferred / Weak; a load-bearing inference caps confidence.
  • Two independent axes — "is it a bottleneck" and "should you buy it here" are never conflated.
  • Honest "too late" — parabolic + extreme valuation + loud consensus ⇒ labeled a top, not an opportunity.

What's in 0.1.0

  • Phase-0 positioning + 5-department chain: physical · human_dev · economics · financials · leaders
  • Markets: A-share / HK / US stocks + crypto (token / EVM contract)
  • LLM: Gemini (real-time grounding, default) / OpenAI / Claude / DeepSeek / offline MockLLM
  • CLI (cyberagent) + local web UI (cyberagent serve) + Claude Skill (SKILL.md)
  • Prompts & methodology fully open-source — no paywall

We borrow these thinkers' method — not their words, positions, or endorsement.
Educational and research use only; not financial advice.

Assets 2
Loading