Model Checker for Financial Audits
Systematically audit financial models for circular refs, broken formulas, hardcoded overrides, and balance sheet imbalances before client or IC delivery.
The skill clearly instructs a read-only audit on user models: 'Don't change the model without asking' and warns not to modify the user's model, aligning with least privilege. It specifies input from work directory and output to same. However, it does not explicitly mention user confirmation (though implied), nor does it address sensitive data handling or data-flow disclosure. Dependency security or rollback is absent. So main risks visible but permissions and confirmation incomplete, deduct to 14.
The skill provides clear, detailed, step-by-step audit procedures (structural, integrity, logic, common bugs) that are self-consistent. It describes failure modes (BS imbalance, formula errors) but no execution evidence, and static review only, so happy path plausible but tests and error handling thin. Deduct to 8.
The skill's purpose is clear: model auditing for financial modellers. Trigger phrases and scenarios are well-defined. Boundaries are stated (e.g., note about VBA macros, reference to Excel formatting standards). However, environment fit is not discussed—it expects Excel model input from a work directory, possibly a cloud sandbox, but not specified, and does not address mainland China or related service reachability. So clear scenario but boundaries and triggers limited evidence, deduct to 11.
Documentation is structured well with step-by-step and common-bug tables. License is clearly stated as derived from anthropics/financial-services-plugins (Apache-2.0). No version, changelog, or maintenance responsibility. Parameter naming is consistent ('$WORK_DIR/work/{task}/'). Examples are descriptive, not code. Known limitation (VBA) is mentioned. So readable and usable but hidden assumptions, troubleshooting gaps, or incomplete versioning, deduct to 10.
The skill seems likely to complete the model audit task, generating a report with issue log. But no execution evidence, so direct usability of output unverified. Marginal value is high since manual audit is time-consuming, but evidence of effectiveness is limited. So completes core task but output completeness, direct usability evidence limited, deduct to 6.
There are no tests or CI evidence available, so only author claims or descriptive content. No independent verification possible. Deduct to 4.
- The skill is described as 'derived from anthropics/financial-services-plugins' with modifications, but degree of modification unknown; verify license compliance.
- The skill does not mention any dependencies or external tools other than Excel itself, so dependency security or supply-chain risks are unassessed.
- No versioning or changelog, may not be maintained over time.
- This skill is designed to audit financial models; ensure only user-authorized models are audited; the skill itself provides appropriate instructions.
What does this skill do, and when should you use it?
This skill provides a structured QA workflow for financial models. It ingests an Excel workbook (.xlsx or .xlsm), identifies the model type (DCF, LBO, merger, 3-statement, comps, returns), and maps its structure. It performs structural checks (tab layout, color coding, hidden tabs, formula consistency), integrity checks (balance sheet balance, cash flow tie-out to balance sheet, retained earnings rollforward), logic checks (reasonableness of growth, terminal value dominance, sensitivity), and model-type-specific bug catalogs (DCF, LBO, merger, 3-statement). The output is an audit report with severity-ranked issue log (Critical/Warning/Info) and suggested fixes, saved to the designated work directory.
Reads a user-provided Excel financial model from the file path $WORK_DIR/work/{task}/. Determines the model type and tab structure. Checks for input/calculation separation, color conventions, hidden tabs, hardcoded numbers inside formulas, inconsistent formulas across ranges, and formula errors (#REF!, #VALUE!, #N/A, #DIV/0!). Verifies balance sheet identity for every period. Reconciles cash flow statement to balance sheet (ending cash, D&A, capex, working capital changes) and income statement (tax expense, share count). Detects circular references and checks iteration settings. Performs reasonableness checks (growth rates, margins, terminal value share, hockey-stick projections) and edge cases (zero/negative growth, negative EBITDA, leverage bounds, divide-by-zero). Ensures cross-tab consistency (linked cells, date headers, units). Applies model-type-specific bug catalogs for DCF, LBO, merger, and 3-statement models. Produces an issue log table, annotated model (if user provides file), and summary assessment with fix priority, saving to the same work directory.
- An investment banking analyst checks a 3-statement model before sending it to a client, ensuring the balance sheet balances and cash flow ties out.
- A PE professional spots a discrepancy in the cash sweep mechanics of an LBO model and uses the skill to pinpoint the issue.
- Before an IC meeting, an analyst runs an M&A model through the audit to verify accretion/dilution, purchase price allocation, and synergy phasing.
- A new hire takes over a colleague's DCF model and wants to find why FCF includes interest expense and terminal value is discounted incorrectly.
What are this skill's strengths and limitations?
- Comprehensive coverage: structural, integrity, logic, and model-type checks provide a complete audit workflow.
- Structured output: severity-ranked issue log and fix suggestions make prioritization easy.
- Clear triggers: multiple trigger phrases for easy activation.
- Practical focus: built around common model errors (hardcodes, sign conventions) rather than theory.
- Dependent on a specific file path ($WORK_DIR/work/{task}/) — may need adjustment in other environments.
- Requires ability to open Excel files; the skill itself doesn't include built-in Excel read/write (references another skill).
- For models with VBA macros, macro-driven calculations can't be audited from formulas alone — a limitation.
How do you install this skill?
Place the SKILL.md file from skills/check-model/ into your agent skills directory (e.g., ~/.claude/skills/ or project .agents/skills/). The skill is part of the LangAlpha repo but can be used standalone; copy the file to the appropriate location.
How do you use this skill?
Mention a trigger phrase like "check my model" or "debug model" in a conversation, and provide the Excel file path. The skill reads the file from $WORK_DIR/work/{task}/, performs the audit, and outputs the issue log, annotated file, and summary. Ensure the file is in the expected directory.