What does this skill do, and when should you use it?
file-itr is an Agent Skill for resident Indian individual income tax filing, distilled from a real end-to-end ITR-3 filing (salary + 44ADA content-creator presumptive income + listed-share STCG) and generalised across filer types. It builds an FY employment timeline first, reconciles every income head to source documents (Form 16, 26AS, AIS, bank and broker statements), independently computes tax under both the old and new regime (115BAC/Form 10-IEA) to pick the cheaper one, then fills the e-filing portal schedule-by-schedule and clears validation defects before handing payment, submission, and e-verification back to the user. The stated goal is the lowest legal tax — proactively chasing every deduction the user genuinely has — never fabricating or inflating. Non-resident/RNOR returns, GST, TDS returns, and company returns are explicitly out of scope.
- Derives how many Form 16s to collect from the FY employment timeline (job change, concurrent employers, mid-year join/exit, and gap months each have distinct rules).
- Reconciles each income head (salary, house property, 44ADA/44AD presumptive, capital gains, interest/dividends, crypto VDA) to a source document — Form 16, 26AS, AIS, bank statements, broker or platform payout files.
- Computes total income and tax in a script under both regimes and picks the cheaper one, handling 87A rebate, surcharge, and the 4% cess.
- Selects the correct ITR form (ITR-1/2/3/4) from income composition, including AY 2026-27 boundaries like the ₹1.25L Section 112A limit.
- Fills the e-filing portal schedule-by-schedule, applying known workarounds (logout pop-ups, mat-select dropdowns, the trailing-zero bug, silent schedule un-confirmation) until validation errors reach zero.
- Proactively asks for 80C/80D/NPS/HRA/home-loan-interest/80G proofs so no legitimate deduction is left unclaimed in the regime comparison.
- A salaried employee who changed jobs mid-year: multiple employers mean one Form 16 each, with TDS merged across Schedule S rows.
- A content creator or freelancer with side income: 44ADA presumptive taxation (CBDT code 16021), including the no-account balance-sheet validation defect.
- An investor who sold listed shares or mutual funds this year: 111A/112A special rates, quarterly breakup for 234C, and the ITR-2 vs ITR-3 call.
- A crypto trader on Indian exchanges: Schedule VDA at the flat 30% under 115BBH, with no loss set-off and 1% TDS under 194S.
- A salaried person with 80C, a home loan, and HRA: wants the old-vs-new regime answer computed on real numbers, not guessed.
- A retiree over 60 with pension and FD interest: needs the senior-citizen slabs and 80TTB applied correctly.
- Non-residents (NR/RNOR): DTAA relief, Form 67 foreign tax credit, and Schedule FSI/TR are explicitly out of scope.
- Company returns, GST, or TDS returns (24Q/26Q): the skill excludes these scenarios outright.
- Users who want the agent to pay or submit on their behalf: by design it never touches passwords/OTPs, payments, or the submit/e-verify actions — those are legally the user's.
How do you install this skill?
- This is a static source review; nothing was executed or independently tested; confidence is low.
- Tax rules change every assessment year — re-verify rates, limits, and form eligibility for the target AY before use.
- The engine is scoped to AY 2026-27 and fail-loudly refuses out-of-scope cases (HUF, non-residents, foreign income); do not force-fit inputs.
- Payment, submission, and e-verification remain the user's acts; the skill is not a substitute for a chartered accountant and the taxpayer stays responsible for filed figures.
- Publisher is unverified (identity unknown); no versioning or changelog — track the repository yourself for updates.
- Shell / CLI
- Network access
- Local filesystem
Python (for the independent tax-computation scripts and engine/ helpers)
Claude Cowork (desktop): download itr-india.skill from the repo, drag it into the chat and click "Save skill", or add it via Settings → Capabilities → Skills.
Claude Code (CLI, project-level): clone the repo and copy the skill into your project's skills folder.
git clone https://github.com/shivprime94/file-itr.git
mkdir -p .claude/skills
cp -r file-itr/skills/itr-india .claude/skills/Claude Code (CLI, user-level, all projects):
git clone https://github.com/shivprime94/file-itr.git
mkdir -p ~/.claude/skills
cp -r file-itr/skills/itr-india ~/.claude/skills/How do you use this skill?
Once installed, send your agent any of these to trigger it:
- Help me file my ITR for FY 2025-26 — I'm salaried and also do freelance content work, and I sold some shares this year.
- Reconcile my 26AS and AIS and tell me my total income and tax.
- I'm stuck on a validation error on the income tax portal for my 44ADA return — help me fix it.
- I traded crypto on an Indian exchange — made profit on some coins, lost on others, 1% TDS was deducted. How is it taxed and which ITR form?
Once installed, just talk about Indian tax filing naturally — the skill triggers via its YAML frontmatter description. The flow: confirm the Assessment Year and residential status and check the due date; build the FY employment history; collect Form 16/26AS/AIS/bank/broker documents and reconcile; run the independent two-regime computation; pick the form; fill the portal schedule-by-schedule. You provide the documents yourself: Form 16(s), 26AS, AIS/TIS, FY bank statements, broker/capital-gains statements, and platform payout files. For browser work the skill prefers precise DOM/JS clicks over coordinate clicks; after editing any schedule it re-confirms everything downstream. Payment, submission, and e-Verify (Aadhaar OTP or pre-validated bank) are yours — "e-Verify Later" must be completed within 30 days or the filing is void.
What are this skill's strengths and limitations?
- Distilled from a real end-to-end ITR-3 filing, with a concrete catalogue of portal defects and workarounds rather than generic tax advice.
- Insists on independent script-based computation for both regimes and ties every rupee of income to a document, so it catches the portal's own errors.
- Proactively chases deduction proofs toward the lowest legal tax, with built-in safeguards against false-premise filings (salary relabelled as consulting, AIS omissions, undocumented 80G, etc.).
- Clean boundaries: never handles passwords/OTPs, never pays, never submits or e-verifies — responsibility is explicit.
- Indian tax rules change every assessment year and AIS/26AS can be incomplete or wrong; the skill requires the user to re-confirm current slabs and limits and remain responsible for the figures.
- Resident individuals only — no non-resident/RNOR, company, GST, or TDS-return support.
- It will not give forward-looking investment/policy recommendations to save future tax, and is not a substitute for a chartered accountant.
- Some reference files listed in SKILL.md (e.g. truthful-filing-safeguards.md, huf-filing.md) don't appear in the README's repo layout diagram, so the full directory may differ from what's shown.
How does this skill compare with similar options?
Side by side with related skills; every score comes from the same FSRS standard.
| Skill | FS score | Stars | Last updated | License |
|---|---|---|---|---|
| ITR India Filing Assistant this page | 67 · Recommended | ★ 540 | 20d ago | MIT |
| EdgarTools — SEC EDGAR Filing Analysis Skill | 67 · Recommended | ★ 2.8k | 11d ago | MIT |
| Income Investment: Durable & Opportunistic Distribution Analysis | 59 · Recommended | ★ 17k | 13d ago | MIT |
| Canadian Personal Finance Planner Skill | 58 · Recommended | ★ 63 | 3mo ago | MIT |
| Reconciliation Break Tracer ✓ Anthropic · Official | 53 · Use with care | ★ 40k | 1mo ago | Apache-2.0 |
The skill positions itself as a filing assistant distilled from real filing experience, with portal-specific traps and a full reconciliation-to-submission workflow — not a generic AI tax Q&A and not a replacement for a licensed CA. The source names no direct competitors.
How did FollowSkills review this skill?
SKILL.md sets hard boundaries: never touches credentials/OTPs, never pays, never submits or e-verifies — the user retains all legal acts; SECURITY.md declares sensitive-data red lines and git-history risk; no covert exfiltration or disproportionate permissions; MIT license. Deducted for: unverified publisher identity, no systematic statement of legal-consequence recovery or rollback, and text-only constraints on browser-automation risk surface.
Instructions are self-consistent; the engine is an independently testable fail-loud module, and the repo contains a CI workflow (engine-tests.yml) plus a committed test suite, so a score above 10 is statically permitted. Deducted for: cannot statically confirm the claimed 116 tests cover all key paths; the referenced references/ files were not fully provided; failure-feedback quality on abnormal input is inferred, not verified.
Trigger keywords in the description are precise, in/out-of-scope boundaries (non-resident, RNOR, GST, other countries) are explicit, and scope is AY 2026-27 specific. Deducted for: annual rule-change staleness risk; the reference files themselves were not supplied to verify the boundary claims; no documented fallback for environments that cannot reach the Indian e-filing portal.
Well-layered docs (SKILL.md → references → engine README), MIT, SECURITY.md, CI-kept skill bundle, install instructions for three environments. Deducted for: no version numbers or CHANGELOG, maintenance concentrated in a single unverified maintainer, update path depends on a personal repo, and unpackaged-engine assumptions left to the user.
Clear goal (lowest legal tax, both-regime comparison, rupee-level reconciliation); the independent computation engine adds value over manual methods; README gives concrete use cases. Deducted for: static review cannot confirm outputs are directly usable; the CI+tests evidence permitting >7 exists but the shown test file is truncated and representative outputs are unverified.
Rule engine carries citations per statutory value, CI badges, a pytest suite, and a conftest reproducibility entry point — more than author claims alone. Deducted for: static read cannot independently reproduce conclusions; test code only partially shown; no third-party corroboration.
Open a dimension to read why it scored that way
Evidence confidence：Low — Mostly static review, author material or a limited demo; useful for discovery, not high-risk decisions.
See the full review method →