This skill is prompt + reference markdown, not code — contributions are mostly tax knowledge and field reports, and the bar is accuracy, not style.
- Field reports. Ran the skill on a real return (yours!)? Open an issue with: profile shape (e.g. "salary + RSUs, ITR-2"), what worked, where it gave wrong/thin guidance. Redact all PII. This is the #1 need — most reference content is statute-derived and needs battle-testing.
- Annual refresh (~April–June). Each new AY: new ITR utility schemas (schedule/field names in
references/10), changed limits/rates flagged in the references, new portal validation quirks. - Corrections with citations. Tax content fixes must cite an official source (incometaxindia.gov.in Act/circular, or the e-filing portal). Secondary sources (ClearTax etc.) alone don't clear the bar.
- Depth in thin areas. Currently thinnest: HUF/representative filings, NR/RNOR returns, post-filing notice handling beyond 143(1)/139(9)/154.
- No PII ever — no real PANs, names, account numbers, or actual return figures in issues, PRs, or examples. Use synthetic values.
- Keep the numbers "shape, not truth". The skill mandates runtime verification of every rate against official sources; hardcoded figures in references are illustrations. Don't add content that encourages trusting a written number over a verified one.
- Respect the scope boundary: this skill ends at the data-pack JSON. Portal automation / data entry belongs in a downstream agent, not here.
- No aggressive-tax-position content. Anything that over-claims exemptions or hides income gets rejected regardless of cleverness — see the red lines in SKILL.md.
Fork → branch → PR against main. For reference-file changes, say in the PR description which
official source backs the change. One topic per PR.