Indian Stock Analyst

Indian Stock Analyst

Run every unique holding through the TradingAgents graph, then present its five-tier ratings and the project's Buy/Hold/Avoid mapping. This workflow performs research only and never places orders.

Step 1: Detect the Runtime and Authentication

Check both the installed CLI and Python package rather than assuming either exists:

command -v indian-stock-analyst || echo "CLI_NOT_FOUND"
command -v pi && pi --version || echo "PI_NOT_FOUND"
python3 -c "import tradingagents, indian_stock_analyst; print('PYTHON_READY')" 2>/dev/null || echo "PYTHON_NOT_READY"
python3 -c "import sys; print(sys.version_info >= (3, 10))"
for key in OPENAI_API_KEY ANTHROPIC_API_KEY GOOGLE_API_KEY OPENROUTER_API_KEY; do
  [ -n "$(printenv "$key")" ] && echo "$key=SET"
done
printf 'PI_BACKBONE=%s/%s\n' "$PI_PROVIDER" "$PI_MODEL"

Use this decision tree:

1. CLI found + an LLM key is set → use the direct-provider CLI path in Step 3.
2. CLI found + Pi is authenticated → use --provider pi; Pi may use OAuth without exposing a key.
3. Python packages found → use python -m indian_stock_analyst with the matching provider path.
4. Project checkout found but packages are absent → with user permission, create a Python 3.10+ virtual environment and run pip install -e '.[engine]', then repeat detection.
5. Neither a provider key nor working Pi authentication exists → guide setup; never print secrets.
6. Engine remains unavailable → report the blocker. Do not silently replace TradingAgents with a different verdict engine.

Exit gate: a Python 3.10+ runtime can import both packages and either a direct provider or Pi is authenticated.

Step 2: Resolve Inputs and Defaults

Locate the uploaded CSV/XLSX/XLSM and use these defaults unless the user overrides them:

The parser accepts symbol columns named Instrument, Trading Symbol, Symbol, Ticker, Security, or Stock. It converts NSE symbols to .NS, BSE symbols to .BO, honors an Exchange/Segment column, strips broker prefixes such as NSE: and series suffixes such as -EQ, and analyzes duplicate normalized symbols once.

For a portfolio with more than 10 stocks, warn that the full graph makes many LLM calls. Unless the user already approved a full run, recommend validating configuration with --max-stocks 1 first.

Exit gate: identify one readable portfolio file and the intended analysis date/exchange.

Step 3: Run TradingAgents

Preferred installed-CLI method:

indian-stock-analyst "<portfolio-path>" \
  --date "<YYYY-MM-DD>" \
  --exchange NSE \
  --output "<output-directory>"

Python-module fallback:

python3 -m indian_stock_analyst "<portfolio-path>" \
  --date "<YYYY-MM-DD>" \
  --exchange NSE \
  --output "<output-directory>"

For a direct API provider, add explicit model flags only when supplied:

--provider <provider> --deep-model <model> --quick-model <model>

For a Pi OAuth backbone, use:

--provider pi --pi-provider <pi-provider> \
--deep-model <pi-deep-model> --quick-model <pi-quick-model>

The Pi adapter launches strict-LF RPC sessions with no Pi tools, extensions, skills, prompts, or context files. TradingAgents remains responsible for tool execution. OAuth credentials stay inside Pi's credential runtime.

For the initial cost/config check, append --max-stocks 1. After it succeeds, remove the limit to run the approved portfolio. Keep checkpointing enabled unless the user requests --no-checkpoint.

The adapter configures Yahoo Finance data, Nifty/Sensex benchmark resolution, and India-specific RBI, SEBI, rupee, crude, GDP, FII, and monsoon macro queries before invoking TradingAgentsGraph.propagate().

Exit gate: the command writes portfolio-report.md and results.json. A per-stock failure is acceptable only if recorded; continue analyzing the remaining holdings.

Step 4: Validate the Artifacts

Read the generated portfolio-report.md and results.json. Verify:

• Every parsed ticker has either a TradingAgents rating or an explicit error.
• Every completed stock has a linked artifacts/<ticker>/complete_report.md.
• Ratings are only Buy, Overweight, Hold, Underweight, Sell, or REVIEW.
• REVIEW remains REVIEW; never turn an unparseable output into HOLD.
• "Sentiment confidence" is described as source/data quality, not prediction confidence.
• No report implies that an order was or will be placed.

If all stocks failed, summarize the common provider, key, rate-limit, model, or data error and stop. If only some failed, clearly separate them from completed analyses.

Exit gate: summary counts match results.json, links exist, and mapping is correct.

Step 5: Respond to the User

Use this exact structure:

1. Run summary — date, number completed/failed, engine/provider, and output path.
2. Portfolio decisions — table with ticker, TradingAgents rating, mapped verdict, and status. Include sentiment confidence only when the social analyst was explicitly enabled and returned reliable data.
3. Key portfolio observations — concise themes grounded in completed TradingAgents decisions; distinguish portfolio-manager conclusions from your own aggregation.
4. Failures or limitations — missing symbols/data, stale inputs, disabled/unavailable sources, and provider errors.
5. Artifacts — link the consolidated report, JSON, and per-stock complete reports.
6. Disclaimer — AI-generated research is not financial advice; verify data and consult a SEBI-registered adviser.

Do not invent confidence scores, news, prices, or rationale beyond the generated artifacts. Preserve the TradingAgents five-tier rating alongside the simplified verdict so users can see the original engine output.