A Claude Code skill that takes a user from raw documents (Form 16, AIS, Form 26AS, broker Tax P&L, payslips) to a fully reconciled, schedule-by-schedule computation and a portal-ready data-pack JSON, designed to be handed to another agent (or a human) that.
Claude Code
$ npx -y skills add NidheeshJain/itr-prep-skill --agent claude-code
Codex
$ npx skills add NidheeshJain/itr-prep-skill -a codex
Cursor
$ npx skills add NidheeshJain/itr-prep-skill -a cursor
Gemini CLI
$ npx skills add NidheeshJain/itr-prep-skill -a gemini-cli
GitHub Copilot
$ npx skills add NidheeshJain/itr-prep-skill -a github-copilot
OpenCode
$ npx skills add NidheeshJain/itr-prep-skill -a opencode
Windsurf
$ npx skills add NidheeshJain/itr-prep-skill -a windsurf
A Claude Code skill that takes a user from raw documents (Form 16, AIS, Form 26AS, broker Tax P&L, payslips) to a fully reconciled, schedule-by-schedule computation and a portal-ready data-pack JSON — designed to be handed to another agent (or a human) that fills the e-filing portal.
Rates/limits in the reference files are treated as shape, not truth — the skill mandates verifying
every rate against official sources (incometaxindia.gov.in / incometax.gov.in) per assessment year and
recording them in a rules_verification block in the output.
No coding needed. You'll copy-paste a few commands into the terminal and then just talk to Claude.
1. Install Claude Code (needs a Claude subscription — sign in when it asks):
curl -fsSL https://claude.ai/install.sh | bash
irm https://claude.ai/install.ps1 | iex
2. Make a folder for your tax documents (e.g. ITR on your Desktop) and drop everything in:
Form 16 from every employer this year, AIS (JSON preferred) + Form 26AS from incometax.gov.in,
broker Tax P&L (Zerodha Console → Reports → Tax P&L), last year's ITR if you have it.
Don't worry about completeness — Claude tells you what's missing.
3. Start Claude in that folder. In the same terminal:
cd ~/Desktop/ITR
claude --dangerously-skip-permissions
(The flag skips per-action approval prompts for a smoother run. It disables all permission prompts session-wide — this skill only reads your documents, verifies rates online, and writes the data pack locally — but skip the flag if you prefer approving each step; everything still works.)
4. Install this skill by just asking. Type into Claude:
Install the skill from https://github.com/NidheeshJain/itr-prep-skill into my Claude skills folder, then prepare my income tax return from the documents in this folder.
Claude fetches the skill, reads your documents, asks you simple questions (job changes? sold shares?),
verifies current tax rules online, and produces ITR_datapack_<PAN>_AY<year>.json — every number
traced to your documents, with anything unresolved flagged for you.
5. Pay and file. Claude will tell you if tax is still payable (pay it on the portal when told, before filing). Then either fill the portal yourself using the data pack as your cheat-sheet, or hand it to a browser agent ("Handing off" below). You always do the payment, the final Submit, and e-verify.
The data pack is designed to be consumed by a second agent with browser access — e.g. Claude in Chrome (install the Claude Chrome extension) or a cowork session. Attach the data-pack JSON (and optionally the raw documents for spot-checks). A prep run ends by showing you this prompt pre-filled for your case — just copy-paste it; the generic version:
I need you to fill my Indian Income Tax Return on https://eportal.incometax.gov.in using the attached
data pack: ITR_datapack_<PAN>_AY<year>.json. It was prepared by another agent after full reconciliation
of my Form 16s, AIS, 26AS and broker statements — treat its figures as authoritative and fill each
schedule with these EXACT numbers. The JSON's comments explain where each figure goes.
Rules:
1. I will log in myself — never ask for or handle my password or OTP.
2. Work schedule by schedule in the portal's Return Summary; after editing any schedule, re-confirm
downstream schedules (the portal silently un-confirms them).
3. The portal pre-fills from AIS — where pre-fill disagrees with the data pack, the data pack wins;
tell me about any disagreement instead of silently accepting either.
4. Resolve every item in `verification_checklist_before_submit` and every TODO/VERIFY flag with me
before proceeding.
5. When validation shows 0 errors, STOP. Show me the final computed tax vs the pack's
`tax_computation` and `balance_and_interest` figures. I will make the payment, click Submit,
and e-verify myself.
6. For Schedule 112A, don't type rows manually: use the portal's "Download CSV Template" and fill it
from the pack's `csv_ready_112A` block (follow `csv_generation_notes`). If any field is missing or
you need to reconcile a figure at runtime, use the attached broker tax report.
7. Once everything is filled and the portal validates with 0 errors, download the portal's JSON so I
can take it back to my prep agent for a final line-by-line reconciliation and sign-off before I
pay and submit.
The prep agent stays with you to the end: bring the portal JSON back to it for the final reconciliation, and if validation errors come up, either let the filling agent fix them or send the prep agent a screenshot — it will fix the pack or tell the filling agent exactly what to do.
If anything in the pack looks wrong mid-fill, the filling agent should stop and flag it — the pack carries source attribution for every figure so discrepancies can be traced.
SKILL.md — 8-phase workflow orchestratorreferences/01–10 — per-domain depth, loaded lazily by profilescripts/decrypt_ais.py — decrypts the portal's encrypted AIS JSON (PAN + DOB)Manual install: git clone https://github.com/NidheeshJain/itr-prep-skill.git ~/.claude/skills/itr-prep-skill
(Python deps self-install on first run.)
This is a careful preparer, not a chartered accountant. Users remain legally responsible for filed figures. Audit cases, HUF/deceased filings, and contested notices are explicitly escalated to a CA.
Field reports from real filings are the most valuable contribution — see CONTRIBUTING.md. Licensed under MIT.
Stars on GitHub
Reads text from scanned pages, photos and PDFs and returns it as structured Markdown or JSON, so you can pull tables, formulas and headers out of a document without retyping them. Covers many languages and works as input for AI tools.
@paddlepaddleGives your AI assistant a memory that persists across sessions, so it remembers your preferences and earlier work in this repository instead of starting fresh. Search, forget, pause and resume memories with simple commands.
@mem0aiGives AI agents a library of ready-made skills for scientific work such as metabolic flux analysis, protein experiment design, time series modeling, and analytical method validation, so research tasks need less hand-built code. It works with any agent that supports the open Agent Skills standard.
@k-dense-aiGives Claude Code skills, agents, and commands for PostHog analytics, session replay, feature flags, error tracking, and LLM observability. Includes MCP server setup and project access checks.
@posthogA Claude Code skill that takes a user from raw documents (Form 16, AIS, Form 26AS, broker Tax P&L, payslips) to a fully reconciled, schedule-by-schedule computation and a portal-ready data-pack JSON, designed to be handed to another agent (or a human) that. Install it with the command on its page. It includes itr-prep-skill. Its skills do not fire on their own yet. Free and open source.
Is this plugin yours?
Claim it with GitHubSubmit a pluginPromote it