Contribute
Tell us where the econometrics is wrong.
This is guidance an AI will follow literally. A rule stated too strongly becomes a rule applied where it does not hold; an exception left out becomes one the model never considers. So the most valuable contribution is not a missing skill — it is "this skill says X, and X is false when Y."
What helps most
Corrections
A rule that is wrong, overstated, or missing a well-known exception. A correction with a citation is merged faster than one with an argument.
Failure cases
A transcript where a model followed a skill and still produced bad econometrics. Guidance that is present but not operative needs a different fix from guidance that is missing.
Software idioms
An R or Stata idiom that is outdated, has a better modern equivalent, or breaks in a common version.
Local practice
Conventions differ by field and country. Where a skill states one as universal and it is not, say so.
Before you open a pull request
- Read CONTRIBUTING.md and, if you are editing a skill, the house style in skills/README.md.
-
Run the validator —
python tools/validate_skills.py. It checks the frontmatter contract, cross-references, and length limits, and CI runs it on every push. - Describe the failure your change prevents. “Adds a note about few clusters” says less than “a model following this skill would have used cluster-robust errors with six clusters and reported a spurious result.”
- One logical change per pull request. A correction and a new skill are two.
What not to send
Policy positions. The project has views about method and none about outcomes, and a skill that acquires an opinion about what should be done is a defect — governance explains the line. Also: skills for a proprietary tool or dataset most readers cannot access, and padding. Length is a real cost here, because a skill competes for the model's context window and every sentence that is not load-bearing dilutes the ones that are.
Licensing
Contributions are licensed under MIT for code and CC BY 4.0 for written material, matching the rest of the repository. There is no contributor licence agreement and no copyright assignment — see licences.