A-Share Financial Forensics: Agent Skill for Earnings Quality Audits

Where this fits

You're building an agent that screens A-share companies for earnings quality. You feed it a ticker, and it returns a verdict. But the agent keeps confusing aggressive accounting with outright fraud, or it misses that the company quietly changed its revenue recognition policy. You need a skill that enforces audit discipline: check the three statements, reconcile cash flow to profit, spot disclosure drifts, and separate evidence from red-flag counting. a-share-financial-forensics does exactly that—it's a structured workflow for financial statement forensics, built for agents that must stay factual and avoid overclaiming.

Why agents benefit

• Rigorous workflow: Enforces a step-by-step process—lock evidence, fetch PIT data, reconcile statements, analyze 3–5 year cash-profit divergence—so your agent doesn't jump to conclusions.
• Disclosure drift detection: Explicitly checks for changes in KPI definitions, segment reporting, and accounting policies, and quantifies the impact using old vs. new calculations.
• Risk classification without accusations: Outputs clean, watch, concern, material concern, or insufficient evidence—never a probabilistic fraud score. Keeps your agent legally and ethically safe.
• Anti-pattern guardrails: Hard constraints prevent your agent from using red-flag counts to infer fraud, from ignoring historical/peer comparison, and from giving trade recommendations.
• Verifiable next steps: Each anomaly includes normal explanations, risk explanations, counter-evidence, and next-period metrics, making your agent's reasoning auditable and actionable.

Practical scenarios

1. Pre-investment earnings quality screening

Your agent reviews a potential portfolio company. It pulls the last five years of financials, runs the three-statement reconciliation, and notices operating cash flow consistently lags net income. The skill directs it to dig into receivables and inventory buildup, and flags a concern about revenue recognition timing. No fraud accusation—just a clear risk rating with supporting evidence.

2. Monitoring a portfolio for accounting red flags

You run a monthly alert system. The skill detects that a company changed its depreciation method mid-year without a restatement. It quantifies the impact on net income, marks a watch for disclosure quality, and suggests next-quarter metrics to confirm whether the change is cosmetic or substantive.

3. Post-earnings response assistant

After an earnings release, your agent answers analyst questions. It uses the skill to check whether the company's adjusted EBITDA definition shifted, recalculates the old-metric value, and reports the variance. This prevents your agent from accidentally comparing apples to oranges in its commentary.

Add it to your agent workflow

Install the skill via npx:

npx skills add cyijun/china-financial-services --skill a-share-financial-forensics

Then invoke it in your agent's toolchain. For example, in a Python agent:

result = skill.run(
    company="600519",
    period="FY2024",
    data_sources=["a-share-research-evidence", "china-market-data"],
    output_format="structured"
)
print(result.risk_ratings)  # {'revenue_recognition': 'concern', 'cash_flow': 'watch'}

The skill automatically pulls the required evidence, runs the reconciliation, and returns a structured report following the output contract. For a deeper look, see the skill page for full documentation and the reference checklist.

Compare before adopting

Before integrating, compare against your current approach. Check whether the skill's data sources align with your PIT requirements—it strictly prohibits downgrading to AKShare data without announcement dates. Evaluate the maintenance freshness (last commit, issue resolution) and whether the author responds to feedback. Look at alternatives like generic financial-analysis prompts; they lack the hard constraints and output contract that prevent false fraud claims. Also consider workflow fit: does your agent already have a research-evidence step? This skill expects that as a prerequisite.

Why it is worth tracking

With 18 stars and an active repository, this skill stands out for its disciplined, evidence-first methodology—rare among agent skills that often overpromise. Its clear output contract and explicit anti-accusation rules make it a model for responsible financial AI. Track it if you process A-share financial data; even if you don't adopt it, its workflow design is worth studying. Re-evaluate when the author adds new data sources or expands coverage to other Asian markets—signals of growing community trust and utility.