B
bitcoinvsalts/finnegans-wake-skill
A Claude skill that borrows Finnegans Wake's compression devices: same verdict on 'why' questions, about a third fewer output tokens.
updated 2026-11-19
View on GitHub ↗Direct link · no redirect page
- Star count
- 0 ★ · bitcoinvsalts/finnegans-wake-skill · read from the GitHub search API · (checked 2026-11-19 20:00:48)
- Source
- read directly from its GitHub page · type: open-source project
- Harvest history
- first seen 2026-11-20 · last seen 2026-11-20 · seen 1× · listed by 2 sources
- Score
- score 203 · 2 sources · repo
FAQ (answers come from our records)
Who is it for?
We filed it under these tags: Open Source. The tags come from the entry itself, not a taxonomy we imposed.
Who maintains it?
repository
bitcoinvsalts/finnegans-wake-skill;first catalogued 2026-11-20;last seen 2026-11-20。There is no official certification or quality rating here — we record sources and dates only.README excerpt (English, untranslated) · Read the full README on GitHub →
# Hypothesis Engine, Wake edition
Excerpted from the repository README (abridged)
A Claude skill for uncertain "why did this happen?" and "does X cause Y?" questions. It borrows
compression devices from James Joyce's Finnegans Wake so Claude reaches the same verdict in far
fewer words.
## What you get
Ask Claude a cause-and-effect question with messy evidence: an incident, a metric that moved, an
A/B test, a health claim, sources that contradict each other. With the skill, the answer is:
The verdict first: supported, refuted or underdetermined, and how sure.
The strongest rival explanation, and the one fact that defeats it.
One next check: the test whose result would most change the conclusion.
That's five short lines at most, unless you ask for depth.
## Measured results
Claude Opus 5.5 at high effort. Each question was answered with and without the skill, and a
separate model judged the answers blind.
| Test | Cases | Correct verdicts | Output tokens |
| Blind holdout, written by an agent that never saw the skill | 24 | 100% both ways | −37% |
| Validation set | 17 | 100% both ways | −28% |
| Fresh generated cases, outcome rule fixed before the run | 32 | 100% both ways | −34% |
For comparison, in a later run on the validation and fresh cases together (49 cases), adding the
single line "Be concise and lead with the answer." to the prompt saved 26% of output tokens, against
34% for the skill. The skill does better, but most of the gain comes from asking for brevity at all.
These numbers come from a private test harness that isn't part of this repository.
## When it helps, and when it doesn't
It helps when answers to analysis and "why" questions make up a large share of the output
you pay for.
It doesn't help with replies that are already short, with extraction or JSON output, or with
agent loops where most tokens are tool calls.
It only cuts output. The skill adds about 720 input tokens to every call. On the API, put it
in a cached system prompt, or it pays off only on long answers.
## Install
Claude Code:
bash
git clone https://github.com/bitcoinvsalts/finnegans-wake-skill.git
cp -r finnegans-wake-skill/hypothesis-engine ~/.claude/skills/
Claude loads it on its own for the questions in its description. You can also call it with
/hypothesis-engine.
Claude app (claude.ai or desktop): zip the hypothesis-engine folder and upload it as a skill.
Claude API: paste the body of SKILL.md (everything below the --- header) into your system
prompt, and cache it.
## Where the rules come from
| Wake device | In the book | Rule in the skill |
| riverrun / ricorso | The last sentence runs into the first; Vico's cycle ends in a ricorso that starts the next round | Verdict in the first line; end with the one check that starts the next round |
compression devices from James Joyce's Finnegans Wake so Claude reaches the same verdict in far
fewer words.
## What you get
Ask Claude a cause-and-effect question with messy evidence: an incident, a metric that moved, an
A/B test, a health claim, sources that contradict each other. With the skill, the answer is:
The verdict first: supported, refuted or underdetermined, and how sure.
The strongest rival explanation, and the one fact that defeats it.
One next check: the test whose result would most change the conclusion.
That's five short lines at most, unless you ask for depth.
## Measured results
Claude Opus 5.5 at high effort. Each question was answered with and without the skill, and a
separate model judged the answers blind.
| Test | Cases | Correct verdicts | Output tokens |
| Blind holdout, written by an agent that never saw the skill | 24 | 100% both ways | −37% |
| Validation set | 17 | 100% both ways | −28% |
| Fresh generated cases, outcome rule fixed before the run | 32 | 100% both ways | −34% |
For comparison, in a later run on the validation and fresh cases together (49 cases), adding the
single line "Be concise and lead with the answer." to the prompt saved 26% of output tokens, against
34% for the skill. The skill does better, but most of the gain comes from asking for brevity at all.
These numbers come from a private test harness that isn't part of this repository.
## When it helps, and when it doesn't
It helps when answers to analysis and "why" questions make up a large share of the output
you pay for.
It doesn't help with replies that are already short, with extraction or JSON output, or with
agent loops where most tokens are tool calls.
It only cuts output. The skill adds about 720 input tokens to every call. On the API, put it
in a cached system prompt, or it pays off only on long answers.
## Install
Claude Code:
bash
git clone https://github.com/bitcoinvsalts/finnegans-wake-skill.git
cp -r finnegans-wake-skill/hypothesis-engine ~/.claude/skills/
Claude loads it on its own for the questions in its description. You can also call it with
/hypothesis-engine.
Claude app (claude.ai or desktop): zip the hypothesis-engine folder and upload it as a skill.
Claude API: paste the body of SKILL.md (everything below the --- header) into your system
prompt, and cache it.
## Where the rules come from
| Wake device | In the book | Rule in the skill |
| riverrun / ricorso | The last sentence runs into the first; Vico's cycle ends in a ricorso that starts the next round | Verdict in the first line; end with the one check that starts the next round |
Listed badge (optional)
Copy this line into the README if you want the listing mark. Optional, and it does not affect being listed.
[](https://anclum.com/en/s/bitcoinvsalts-finnegans-wake-skill/)
- Source
- github · https://github.com/bitcoinvsalts/finnegans-wake-skill
- Browse by field
- Open source
- Listed by
- 2 independent sources
- added
- today
More entries
Filed under
Every tag opens the full list for that filter. Counts are the live catalogue figures.