Agent Skills

Seven Skills for Indian Markets

A skill is a folder of instructions an AI assistant loads when the task calls for it. These seven carry the parts of Indian market practice a general-purpose model gets wrong.

Install

npx skills add https://github.com/ArthaGH/artha-skills-hub

MIT licensed. Read the repository

What they are

Instructions, Not Access

A general-purpose model reasons about Indian markets with US defaults. It annualises on the wrong day count, it reads a split as a crash, and it quotes a metric the sample cannot support. These seven skills carry the conventions, the adjustments, and the places where a US default quietly produces an Indian mistake.

They hold no data and no credentials. A skill changes how an assistant reasons about the figures in front of it, and nothing else.

Three steps

Quick Start

  1. Install them

    Run this in a terminal.

    npx skills add https://github.com/ArthaGH/artha-skills-hub
  2. Point your assistant at them

    Set the assistant’s skills folder to the folder the command wrote.

  3. Ask a question about Indian markets

    The assistant loads the skill the question calls for. You do not name it.

Four layers

What They Give You

The seven sit in four layers. Each layer assumes the one beneath it is already right.

Foundation

Getting the Data Right

Beneath everything else. Every figure in the layers above is wrong if the return series it rests on is wrong.

indian-market-data-literacy

Reading the bhavcopy. Bhav means price, so a bhavcopy is the day's copy of prices: the file NSE publishes after every session, and the one the market and Artha both read daily prices from. What delivery percentage tells you. Why a bonus or a split hands you more shares and cuts the quoted price to match, so a series that does not know it happened shows a crash that never happened and every return built on it is wrong. The .NS and .BO suffixes, why NSE and BSE quote different prices for the same company, and what a circuit limit does to a candle.

Layer 1

What You Own

The portfolio you already hold, measured honestly.

portfolio-pnl-review

Realised against unrealised. XIRR against absolute return against CAGR, and when each is the right question. FIFO on partial sells, and cash-flow-weighted returns when money went in over time.

Prevents reading a 40 percent gain over four years as a 40 percent return.

portfolio-concentration-review

How concentrated the portfolio is, measured rather than counted. Take each holding's share of the portfolio, square it, add the squares, and invert the total. What comes back is the effective number of holdings: how many equal-sized positions the portfolio actually behaves like. Twenty names with half the money in one behave like about four. Sector concentration measured against a benchmark rather than against zero. SEBI and AMFI market-cap bands as they are actually defined. Promoter-group cross-exposure, so four companies in one house register as one bet.

Prevents believing twenty holdings means twenty bets.

portfolio-risk-metrics

Sharpe, Sortino, beta, R-squared, maximum drawdown, volatility, on India-correct inputs: 252 trading days and a risk-free rate near 6.5 percent rather than the US default a model assumes. The minimum sample below which each metric should not be quoted.

Prevents importing US defaults, and trusting metrics the sample cannot support.

Layer 2

What You Are Buying

The instrument and the company behind it, read closely.

mutual-fund-overlap-review

Computing overlap from published holdings. Declared scheme category against actual holdings. Direct against regular expense ratio compounded over a real holding period, expressed in rupees rather than basis points.

Prevents six funds that are one position wearing six names.

reading-indian-filings

Forensics on the India-specific disclosure set: related-party transactions and their approval trail, promoter pledge and what a pledge percentage does not mean, insider and SAST disclosures, auditor qualifications and emphasis-of-matter paragraphs, contingent liabilities, and the gap between standalone and consolidated accounts.

Layer 3

What You Keep

The part that decides what the return was actually worth.

indian-investment-taxation

Structure rather than rates. Indian tax law sorts every rupee into one of five heads of income, and the head decides the rate, what a loss can be set off against, and which form you file. A share you held and sold is capital gains. Futures and options are business income. Bank and bond interest is income from other sources. Holding-period boundaries and where they differ by instrument, the 31 January 2018 grandfathering for equity, STT, dividend TDS, and buyback treatment. The skill asks you to confirm the assessment year before it states any rate, because the Income-tax Act, 2025 renumbers almost every section from 1 April 2026.

Boundaries

What They Can and Cannot Do

They work without Artha

Each skill declares its inputs in three tiers: what it needs, where to read it if the Artha MCP server is connected, and what to ask you to paste if it is not. A skill that only works for Artha users cannot travel.

They state their limits

Every skill says what its output cannot tell you. A concentration number computed from an incomplete holdings list is a number with a caveat attached, and the skill carries the caveat.

They do not recommend

Skills compute, structure, and explain. None of them returns a buy, sell, or hold judgement, and none of them is a substitute for advice from someone registered to give it.

The skills read best alongside the Artha MCP server, which gives them your own figures to work on instead of numbers you paste in by hand.

Artha is operated by Quantrast Technologies Private Limited, CIN U66190MR2026PTC475667. Artha is not a SEBI-registered Investment Adviser or Research Analyst: no advice, no target prices, no recommendations. The skills teach method; they name no security. Read the risk disclosure.

Something not working? A tool that answered wrongly, a client that will not connect, or a step above that does not match what you see.

We use analytics cookies to understand how you use Artha and improve your experience. No data is sold to third parties.