Feat: Adding support for equity markets in india

[muku314115]

Summary

• Add first-class Indian equity (NSE/BSE) support across the full stack: data ingestion → dedicated backtest engine → equity_in factor universe → live/paper broker + mandate wiring.
• New IndiaEquityEngine modeling T+1 delivery, no overnight shorts (allow_short opt-in), configurable circuit bands, 1-share lots, and the STT / stamp-duty / exchange / SEBI / GST cost stack.
• Route .NS/.BO symbols to Yahoo (native NSE/BSE coverage), with yfinance and an opt-in Shoonya/Dhan broker data bridge as fallbacks.

Why

Indian symbols previously had zero research support: they silently misrouted to China infrastructure (tushare loader + ChinaAEngine), producing wrong fills and costs. India was the only market with
live broker connectors (Shoonya/Dhan) but no data/backtest/factor path. This wires the broken middle so the existing strategies, factors, and metrics work on NSE/BSE.

Changes

• Data ingestion: .NS/.BO routing in market_data.py + _market_hooks.py (new india_equity market type); extended yahoo_loader/yahoo_client and yfinance_loader to accept the suffixes;
  india_equity fallback chain in registry.py; runner.py source map. Regexes allow &/- (e.g. M&M.NS, BAJAJ-AUTO.NS).
• Engine: new backtest/engines/india_equity.py; wired into composite.py and runner.py single-market routing (precedes source-based routing so India's yahoo source can't fall through to the
  crypto default).
• Factors: added equity_in to the Universe literal and Market enum (vwap uses the typical-price branch — India bars carry raw price/volume, no Tushare scaling); tagged all 101 Alpha101 and 154
  QLib158 factors for equity_in. GTJA191 left China-only (Tushare-scale math).
• Live/paper: added AssetClass.IN_EQUITY and wired it into the mandate-enforcement loader chain; new backtest/loaders/india_broker_loader.py bridges Shoonya/Dhan get_historical_bars into the
  backtest layer (opt-in india_broker source, unavailable without broker creds). Live order placement stays structurally disabled (those brokers expose no paper/live switch).
• Docs: SKILL.md updated (8 engines, 19 sources, India symbol convention, factor coverage, paper/read-only-live limitation).
• Cost-stack rates and circuit band are config-driven defaults (documented), not magic constants.

Test Plan

• Existing tests pass (pytest --ignore=agent/tests/e2e_backtest --tb=short -q)
• New tests added — test_india_equity_engine.py, test_india_backtest_smoke.py, test_india_mandate.py, test_india_broker_loader.py, tests/factors/test_india_universe.py, plus India cases in
  test_market_data.py, test_market_detection.py, test_yahoo_loader.py, test_registry.py
• Tested manually — registry loads 456 factors (0 failures), 255 listed for equity_in; representative Alpha101/QLib158 factors compute on a synthetic NSE panel; end-to-end India backtest emits metrics

• run card and shows India costs applied vs the zero-commission US engine; factor purity + lookahead gates pass (911); tools/ci_grep_gates.sh all gates pass; ruff check clean across the contribution
  footprint

Checklist

• No changes to protected areas (src/agent/, src/session/, src/providers/) without prior discussion
• No hardcoded values (API keys, file paths, magic numbers) — India cost/band defaults are config-overridable
• Code follows CONTRIBUTING.md guidelines — ruff check (the configured gate) passes on all new/changed files; public signatures type-annotated; Google-style docstrings
• Documentation updated (if user-facing change) — SKILL.md

---

Reviewer notes

• The first Test Plan box is left unchecked: I verified targeted suites + ruff locally, but not the full pytest --ignore=agent/tests/e2e_backtest run (the sandbox lacks some optional deps, e.g.
  fastmcp). Please let CI confirm the full suite.
• Exact SEBI/STT/exchange tariffs and any F&O lot sizes should be verified against a current broker schedule before relying on absolute cost figures (rates change periodically); they're exposed as config
  knobs.
• The india_broker bridge returns a bounded window of recent bars (broker API limit) — use Yahoo for deep history.
• black was not run: the repo is not black-conformant (untouched core files fail black -l120), so formatting only these files would diverge from house style. Files conform to the configured ruff gate
  instead.

[warren618]

Thanks for the PR — we’ve seen it and will review as soon as we can.

For market-support changes, the main things we’ll check are symbol conventions, loader/data-source behavior, deterministic tests, and whether the first version stays narrowly scoped enough to maintain.