Equity Research — Productivity skill for Claude Code

Indian Equity Knowledge System

Point-in-time knowledge system for NSE/BSE equity research. Rules and scope live in [CLAUDE.md](CLAUDE.md); build order in [getting-started-plan.md](getting-started-plan.md).

Setup (Windows / PowerShell)

py -3.13 -m venv .venv          # CLAUDE.md targets 3.12; any >=3.12 works
.\.venv\Scripts\python.exe -m pip install -r requirements.txt   # exact pinned rebuild
.\.venv\Scripts\python.exe -m pip install -e . --no-deps
docker compose up -d --wait db blob   # Postgres 16 + pgvector on :5433, MinIO on :9000
docker compose run --rm blob-init     # write-once buckets (object lock); safe to re-run
.\.venv\Scripts\alembic.exe upgrade head
.\.venv\Scripts\python.exe -m pytest -x --tb=short

Tests recreate the `equity_test` database from scratch and run migrations up → down → up on every run. Tests that need no services: `pytest -m "not db and not blob"`.

Ingest adapters

.\.venv\Scripts\python.exe -m ingest list            # adapters, source class, stores, switch state
.\.venv\Scripts\python.exe -m ingest run NAME        # exit 0 = succeeded or disabled
.\.venv\Scripts\python.exe -m ingest canary          # daily shape check; disabled adapters skipped
.\.venv\Scripts\python.exe -m ingest reparse NAME    # re-derive from stored bytes; never re-fetches

Every adapter is off until `EQUITY_SOURCE__ENABLED=true`; `web_scrape` adapters also need `EQUITY_WEB_SCRAPING_ENABLED=true`. With `EQUITY_DEPLOYMENT_MODE=commercial`, any scraping switch refuses startup, and so does a switch that names no adapter.

Raw bytes go to blob storage (keyed by SHA-256, write-once) before parsing; `raw_source_file` records source URL, publication time (`as_of`) and fetch time. Files downloaded by hand go in `EQUITY_DROP_FOLDER//`, each with a `.meta.json` sidecar:

{"source_url": "https://…", "published_at": "2024-03-01T23:59:59+05:30", "media_type": "text/csv"}

Index constituent lis