Command Reference

source-to-skill is a conservative CLI. Most workflows should start with analyze or demo, not build.

Install from GitHub:

python -m pip install "source-to-skill @ git+https://github.com/GeoLab-org/source-to-skill.git@v0.3.0"

Check the installed version:

source-to-skill --version

Commands

Command	Purpose	Typical output
demo	Run the bundled article and transcript workflow	demo-report.md plus example artifacts
analyze	Score one source before generating files	readiness report or JSON
build	Generate the recommended artifact level, or a forced level	note, seed, mini skill, or full skill
fold	Add a weak source as a seed under an existing skill	seed file in an existing skill folder
evolve	Compare a source with an existing skill before merging	pending update report or JSON
split-source	Split a long source into topic candidates	topic-report.md and segment files
build-segment	Build artifacts from one split segment	segment-derived artifact
fold-segment	Fold one split segment into an existing skill	segment seed in an existing skill folder
clean-transcript	Normalize SRT, VTT, or transcript text into Markdown	cleaned Markdown transcript
transcribe-audio	Run an installed Whisper-compatible CLI on audio	transcript file
eval-skill	Compare generated guidance against evidence	evaluation report or JSON

Common Workflows

Start with the bundled demo:

source-to-skill demo --out out/demo

Score a local source:

source-to-skill analyze examples/article.md
source-to-skill analyze examples/article.md --json

Score a local EPUB:

source-to-skill analyze path/to/book.epub

Build only after reading the score:

source-to-skill build examples/article.md --level auto --out out

Compare a new source with an existing skill before changing the skill:

source-to-skill evolve examples/article.md \
  examples/existing-skill/review-playbook-skill \
  --out out/evolution-demo

This writes a pending update under out/evolution-demo/pending-updates/. It does not edit the target skill automatically.

Clean and score a transcript:

source-to-skill clean-transcript examples/meeting-transcript.vtt \
  --out out/clean-meeting.md \
  --title "Meeting Transcript"

source-to-skill analyze out/clean-meeting.md

Split a long transcript or article into candidates:

source-to-skill split-source out/clean-meeting.md --out out/topics
source-to-skill build-segment out/topics 2 --level seed --out out

Evaluate a generated skill:

source-to-skill eval-skill out/review-playbook
source-to-skill eval-skill out/review-playbook --json

Notes

transcribe-audio does not ship a speech model. It calls an installed Whisper-compatible command and writes the transcript to the path you choose.

Remote URL intake fetches one public text or HTML page. It does not crawl, log in, or render JavaScript pages.

Weak material is expected to stop at Note or Skill Seed. That is a feature, not a failure.