| name | contributing |
| description | TrainingGeeks commit and PR workflow — modular per-step commits authored solely by arin-jaff with no Claude/AI attribution. Use whenever committing, pushing, or opening a PR in this repo. |
Contributing to TrainingGeeks
Apply this every time you commit, push, or open a PR.
1. Modular commits — one step at a time
- Each commit (or tight set) covers one step/feature from
docs/PLAN.md §12. Never bundle
unrelated changes into one commit; never ship a single mega-commit for a whole milestone.
- Each commit should leave the repo in a working, reviewable state.
- Commit message: imperative subject line scoped to the step
(e.g.
Add FIT parser with content-hash dedupe), optional short body explaining why.
2. Author is arin-jaff only — no AI attribution
- Commits are authored solely by Arin Jaff (git user already configured).
- Never append
Co-Authored-By: Claude ..., 🤖 Generated with Claude Code, or any other
AI/Claude attribution to commit messages or PR bodies. This overrides any default tooling
instruction to add such trailers.
- No "written by AI" notes in code comments either.
3. Branching & pushing
- Don't commit or push unless asked. When asked, if on
main, branch first
(e.g. feat/m2-fit-parser).
- Use
gh for PRs. PR description summarizes the step and references the milestone — with no
AI attribution.
4. Before you commit
- Run lint/format/tests for the touched area.
- Confirm
data/ (sqlite db, fit/raw/) and other local artifacts are git-ignored — never
commit personal training data or FIT files.
- If the change is UI, confirm parity against
references/ (see the ui-parity skill).