Skip to content

CONTRIBUTING.md

Latest commit

 

History

History
40 lines (29 loc) · 3.53 KB

CONTRIBUTING.md

File metadata and controls

40 lines (29 loc) · 3.53 KB
Raw

Contributing

This doc is an index that points you at the right place to read. Substantive content lives in the linked files.

I want to… → read this

What you want to do Where to start
Add a new LLM model or provider benchmark/README.md § Adding New LLM Providers
Tweak the verification prompt benchmark/README.md § Where the prompt lives — core/prompts.js is the single source of truth
Fix a bug you see in the Wikipedia sidebar UI The UI is hand-maintained in main.js (outside the <core-injected> markers). CLAUDE.md § Key Methods in main.js is the function-by-function index
Improve how the tool extracts claim text from articles README.md § core/ and the sync script — logic lives in core/claim.js, tests in tests/claim.test.js. The "between adjacent citations" design constraint is in README's Constraints section
Add test cases or extend the ground-truth dataset benchmark/README.md § Workflow + § Dataset Versions
Run the benchmark to evaluate your change benchmark/README.md § Workflow
Spot-check a single citation from the command line README.md § Command-line interface — npx ccs verify <url> <n>
Test your changes on Wikipedia before they ship README.md § Development — load via the User:Alaexis/AI_Source_Verification_test.js page

Before you open a PR

  • npm test — runs the core/ unit tests
  • npm run build -- --check — confirms the live main.js (the file Wikipedia loads) is in sync with core/. Fails if you edited core/ but didn't rebuild.
  • If your change touches the prompt or claim extraction, run the benchmark and check that metrics didn't regress: cd benchmark && npm run benchmark && npm run analyze (see benchmark/README.md § Workflow). This runs the full 189-row dataset across all configured providers and takes a few minutes; a faster smoke-set workflow is on the roadmap. Running it requires at least one API key — see benchmark/README.md § Run Benchmark for details on each. If you don't have a key for financial reasons, ping the maintainers — we'll happily set you up with access.

Repository layout (one-line orientation)

  • main.js — the userscript Wikipedia loads (single-class IIFE, no build step on the install path)
  • core/*.js — pure-logic modules; spliced into main.js by scripts/sync-main.js
  • bin/, cli/ — npx ccs verify CLI
  • benchmark/ — eval suite, ground-truth dataset, frozen snapshots
  • tests/ — node:test suite covering core/
  • docs/ — reference docs and design plans (see docs/README.md)
  • Benchmarking_data_Citations.csv — single source of truth for the dataset

Working with Claude or another LLM

If you're using an AI assistant to contribute: start with CLAUDE.md for architecture and conventions, then read the relevant benchmark/README.md section for the area you're changing. Run npm test and npm run build -- --check before declaring work complete.

Questions

Open an issue at https://github.com/alex-o-748/citation-checker-script/issues.