finance-research-agent: Skill for Claude Code

Financial forensics

A skeptic's audit — not a summary of pre-built flags. The goal is to surface everything questionable — accounting quality, earnings quality, leverage, working-capital, dilution, governance — using the same named, formula-precise tests professional forensic accountants use (Altman, Piotroski, Sloan — all defined with exact formulas in investing-principles), computed from raw statement pulls, then sanity-checked against what management and rating agencies say. A "the flags look clean" conclusion without having actually run these numbers is not a completed audit.

Procedure

Pre-computed flags (financial_health, forensic_checks) are a starting hypothesis list, not the audit. They're multi-year trend detectors — the exact kind of thing clever accounting is designed to slide past (a one-off gain smoothed into a 12y trend, a bad quarter buried inside a good year, a related-party deal that never shows up as a "flag" at all). Steps 3b and 3c below are mandatory on every run, whether or not steps 2-3 fired anything — a clean flag read is not a clean company, it's an unconfirmed one.

1. Resolve the symbol; note the company type (bank/NBFC vs non-financial) — it changes which flags apply.

2. Trend flags — financial_health. Read its concern/watch flags first; these seed the hypothesis list:

   • earnings quality: cumulative operating cash flow vs profit (accruals red flag if CFO lags PAT badly)
   • margin compression, interest-coverage weakness, doubled borrowings
   • genuine per-share dilution (EPS lagging PAT), debtor-day creep, ROCE decline (For banks/NBFCs these cash/coverage/WC flags are auto-suppressed — pivot to asset quality, NIM, growth vs capital.)

3. Forensic checklist — forensic_checks. Read the negatives in each topic bucket (Accounting Quality, Promoter & Governance, Balance Sheet & Debt, Growth & Returns, Valuation & Sentiment) and the explicit promoter-pledge reading.

3b. Named quantitative scores — mandatory, computed not eyeballed. Pull 2+ years of financial_statements(profit_loss) + (balance_sheet) + (cash_flow) and compute, using the exact formulas defined in investing-principles' Quantitative scoring section:

• Altman Z''-Score (EM variant) (distress/solvency, non-financials only) — flag if < 2.6, hard flag if < 1.1. Skip for banks/NBFCs. State plainly that the A-term (working capital) is an estimate from Other Assets − Other Liabilities, not the textbook figure — see investing-principles' data-honesty note.
• Piotroski F-Score (0-9 fundamental-trend score) — flag if ≤ 3, especially alongside a "statistically cheap" valuation (value-trap signature). If the current-ratio test isn't computable (same current/non-current split gap as Altman's A-term), say so and score out of 8 rather than silently treating it as a pass or fail.
• Sloan accrual ratio ((NI-CFO)/Avg Total Assets) — trend it across as many of the 12 years as financial_statements covers; flag a persistently high or rising ratio even before financial_health's qualitative CFO-vs-PAT flag would trigger. Show every input ratio (the five Altman components; the eight or nine Piotroski tests pass/fail) — not just the final score — so the computation is auditable, not asserted. State the two (or more) fiscal years/quarters used for each score explicitly.