Chase Wang

Advisory skills for founders evaluating token launches. Built from 1,000+ project reviews and ~100 token launches in 2025. The tool thinks with you, not for you — it uses forcing questions to surface what you actually know before recommending anything.

• /why-token Answers the most important question first: why a token at all? Routes pure memes to a quick checklist; deep-dives value accrual for everyone else.
• /tokenomics Turns the thesis into numbers. Build from scratch or start from a real token ($UNI, $JUP, $SOL, $LINK, $ENA, $ONDO). Outputs an Excel release schedule with vesting + buyback simulation.

Answers "What does Buffett say about <X>?" with synthesized takeaways grounded in verbatim quotes from his shareholder letters (1956–2024). No paraphrase from memory — every claim traces back to a specific letter, with year and URL.

• 98 letters · 117 entities · 5,659 edges Bundled corpus: every Partnership + Berkshire shareholder letter, plus pre-aggregated dossiers for 49 concepts (moat, intrinsic value, compounding, …), 61 companies (Coca-Cola, GEICO, Apple, …), and 7 people (Munger, Graham, Abel, …). Sourced from learnbuffett.com.
• Autoresearched · 92.96% Karpathy-style iterative refinement against a 5-dimension rubric, with programmatic verbatim-quote checking. The corpus, the rubric, the citation checker, and the round log all ship in the repo — anyone can reproduce.

Turns a folder of declassified UAP/UFO documents — war.gov "PURSUE" releases, FBI Vault tranches, NARA boxes, AARO publications — into an inventory, an entity surface, a per-file digest, and a standardized 11-section REPORT.md you can read in ten minutes. Designed to be honest about what it can't read: scanned PDFs, image-only files, and pypdf-unreadable structures each get their own caveat instead of being silently lumped together.

• 132 files · 4,157 pages · 5 agencies Tuned against the May 2026 war.gov tranche (FBI / DOW / NASA / NARA / DOS). The corpus is mirrored at uap-release-01 (Git LFS, 2.4 GB) so anyone can clone and rerun the same pipeline against the same input.
• Eval-driven · 97% / +23pp Skill-creator iteration loop: 4 eval cases × with-skill vs baseline × 8 subagents in parallel. With-skill mean pass rate 96.9% vs baseline 74.4%; ~12 min vs ~26 min wall-clock; cheaper on tokens too. ARTICLE.md · 中文版