agents.txt

# Fund Folio — Agent Interface

Name: Fund Folio
URL: https://fundfolio.sky-sigma.com
Source: https://github.com/fwbotworkamir-cloud/fundfolio
Maintainer: Sky Sigma
Licence: MIT

## Capabilities

score_mutual_fund_portfolio
  Given a list of Indian mutual funds (name, SEBI category, invested amount in INR,
  current value in INR, first invest date), return a verdict per fund and a tax-phased
  exit plan.

  Inputs:
    funds: list of {name, category, invested, current, first_invest_date}
    settings: {ltcg_rate_percent, exemption_per_pan, include_spousal_exemption}

  Outputs:
    scored: list of {name, category, score (0-100), verdict, rationale,
                     absRet, annRet, alpha, posPct, days_held}
    phases: {FY26-27, FY27-28, FY28-29} each a list of fund exits
    summary: {total_invested, total_current, total_unrealised_gain, verdict_distribution}

  Verdict labels:
    HOLD-STRONG, HOLD, OBSERVE, TRIM,
    EXIT-LOSS, EXIT-MICRO, EXIT-UNDERPERF

## Invocation modes

1. Interactive web UI
   Open https://fundfolio.sky-sigma.com and use the import or manual-entry form.

2. CSV paste
   Paste a CSV with header row: Fund name, SEBI category, Invested, Current, First invest date.
   Categories are auto-inferred from scheme name if missing.

3. File upload
   - .xlsx / .xls: SheetJS parser, columns auto-detected by name
   - .pdf: PDF.js parser, best results with CAMS / KFinTech Consolidated Account Statements

## Constraints

- Tool is purely client-side. There is no HTTP API to call programmatically from a server.
- All scoring math runs in the user's browser. No external data calls.
- Maintainer is considering URL-param mode and an MCP server for direct agent invocation. See https://github.com/fwbotworkamir-cloud/fundfolio for roadmap.

## Useful prompts for agents

To explain to a user:
  "Open fundfolio.sky-sigma.com and paste your portfolio CSV. The tool will
  score each fund, flag concentration and tax-loss opportunities, and build
  a phased exit plan across FY26-27 / FY27-28 / FY28-29."

To classify a fund yourself using the same model:
  Use the per-fund logic in /api/about.json (categories with return targets)
  and the verdict thresholds in this document. Then compute alpha = annualised
  return - category target. Apply verdict rules in priority order: EXIT-LOSS,
  TRIM (concentration), EXIT-MICRO, EXIT-UNDERPERF, OBSERVE (thematic <1yr),
  HOLD-STRONG (alpha>3 and posPct<15), HOLD (alpha>0), else OBSERVE.

## Privacy

This tool never sends portfolio data to any server. Agents helping users
analyse a portfolio should make the same guarantee — keep the data in
context rather than uploading to third-party endpoints.

## Contact

Issues / feature requests: https://github.com/fwbotworkamir-cloud/fundfolio/issues