Performance & Cost
Measured savings, not estimates
Token counts use tiktoken. Discovery figures come from tools/discovery_tax.py. A full skill costs roughly $1 once, far less than rereading the PDF into context every session.
View source on GitHubExtraction on real books
| Book | Format | Pages | Tokens | Chapters |
|---|---|---|---|---|
| Think Python 2 | 244 | 119K | 19 | |
| Working Backwards | 371 | 175K | 10 | |
| Pro Git | 501 | 229K | n/a † | |
| Moby-Dick | EPUB | n/a | 301K | 133 |
† Pro Git uses section titles rather than Chapter N, so it does not auto-segment. Extraction still works.
Technical vs text heavy PDF
On a 103-page technical PDF:
| Method | Time | Tables | Code blocks |
|---|---|---|---|
| pdftotext | 0.1s | 0 | 0 |
| Docling (technical) | 164s | 48 | 36 |
Pick text mode for prose. Use technical mode when you need tables and code preserved as markdown.
The Discovery Loop Tax
Tokens entering context to answer one targeted question. book-to-skill loads a resident core (~4K) plus one compiled chapter (~1K) ≈ 5,000 tokens.
| Book | Context dump | Discovery loop | book-to-skill | vs dump / loop |
|---|---|---|---|---|
| Think Python 2 | 119,264 | 12,152 | ~5,000 | 24× / 2.4× |
| Working Backwards | 175,253 | 33,444 | ~5,000 | 35× / 6.7× |
| AI Engineering | 256,287 | 77,866 | ~5,000 | 51× / 15.6× |
The context dump advantage (24x to 51x) recurs on every conversation turn. The discovery loop advantage scales with chapter size.
Generation cost
One pass full conversion estimated on Claude Sonnet 4.5 ($3 / $15 per MTok input/output):
| Book | Input | Output | ~Cost |
|---|---|---|---|
| Think Python 2 | 155K | 28K | $0.88 |
| Working Backwards | 228K | 19K | $0.96 |
| Pro Git | 298K | 23K | $1.23 |
| Moby-Dick | 391K | 17K | $1.42 |
Roughly $1 per book for a full skill, paid once.
See how conversion works
SKILL.md covers formats, modes, and the files each conversion produces.
Open SKILL.md