First release of uap-release-analyzer.
What this skill does
Inventory + analyze a folder of declassified UAP/UFO documents (war.gov PURSUE releases, FBI Vault, NARA, AARO) and produce an inventory.csv, an analytics/ folder, and a standardized 11-section REPORT.md.
Eval results (iteration-1)
| Eval | with skill | baseline | Δ |
|---|---|---|---|
| Full-tranche walkthrough | 100% | 60% | +40 |
| Single-file summary | 100% | 100% | 0 |
| Scanned-tranche honest caveats | 100% | 88% | +12 |
| Fresh-tranche bootstrap | 88% | 50% | +38 |
| Mean | 97% | 74% | +23 |
Wall-clock: with-skill ~12 min; baseline ~26 min. Tokens: with-skill 239K; baseline 259K. Skill is faster, cheaper, and more accurate.
What v1.0.0 fixes (over the iteration-1 draft)
- Per-file digest now distinguishes scanned PDFs vs PNG/JPG images vs unreadable PDFs (each gets its own status message)
- ICAO airbase codes added to the LOCATIONS vocab (OMAM, OMDB, OEDR, OBBI, OKBK, OAIX, LCRA, HEDC, HEMM)
- SKILL.md now surfaces the agency_vocab extension workflow when answering bootstrap questions
- SKILL.md now tells subagents to stay foreground + use [start end] chunking instead of backgrounding extract_text.py
Install
Download uap-release-analyzer.skill and install via Claude Code's skill UI, or extract under ~/.claude/skills/.
Source: this repo. Build notes: ARTICLE.md. 中文版: ARTICLE_CN.md. Live writeup: https://cbti.club/#essay-uap-skill