Content
71%Weight 40%Scale 1-5Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
A dense, highly actionable body whose module-specific conventions (sign discipline, seed requirements, stderr-based tail typing) are exactly what a skill should add, but it carries a layer of textbook finance theory Claude already knows and inlines reference-style catalogs that belong in separate files. The workflow is clear with embedded sanity checks, though they are not packaged as an explicit validate/fix/retry loop.
Suggestions
Trim the textbook sections — the VaR/CVaR definitions, the three-method advantages/disadvantages table, the subadditivity/Basel row, and kurtosis/skewness explanations — to one line each or drop them; keep only the module-specific conventions and gotchas.
Move the historical scenario table and the hypothetical STRESS_SCENARIOS catalog into a references/ file (e.g. references/scenarios.md) and link to it from the stress-testing section.
Promote the embedded sanity checks (cvar >= var, shape_xi stability across thresholds) into an explicit validation step in 'Analysis Steps' with a fix-and-retry instruction.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Excellent module-specific guidance ('A loss is a positive number', the 2σ tail_type rule, the parametric-vs-historical direction table) sits alongside textbook material Claude already knows, e.g. 'VaR (Value at Risk) ... the maximum expected loss over a given horizon', the three-method advantages/disadvantages table, the VaR-vs-CVaR subadditivity/Basel comparison, and kurtosis/skewness explanations. Not 4 because these known-concept sections are a noticeable share of the body; not 2 because the bulk is genuinely non-obvious implementation knowledge, not padded filler. | 3 / 5 |
Actionability | Concrete, copy-paste-ready calls throughout: 'historical_var(returns, confidence=0.99, horizon=10)', 'monte_carlo_gbm(s0=100.0, ..., seed=42)', 'fit_gpd_tail(returns, threshold_pct=5.0)', plus exact return shapes ('paths.shape # (10000, 253)') and failure behavior ('fewer than 2 exceedances raises'). The common cases are covered by specific examples; nothing is pseudocode. | 5 / 5 |
Workflow Clarity | 'Analysis Steps' gives a concrete 7-step sequence with numeric parameters ('compare three methods at both 95% and 99%', '10,000 paths', 'at least 3 historical scenarios + 2 hypothetical'), and there are real checkpoints ('If you ever compute a CVaR below its VaR, the tail mask is wrong', 'Check that shape_xi is stable across a few nearby threshold_pct values'). Not 5 because the checkpoints are embedded in method notes rather than an explicit validate-and-retry loop tied to the step sequence; not 3 because validation guidance is present, just not formatted as a loop. | 4 / 5 |
Progressive Disclosure | No bundle files exist (no references/, scripts/, or assets/), and the body is a single ~300-line file with good section headers. The historical scenario catalog, EVT theory, and method-comparison tables ('2008 financial crisis ... -65%', the GPD/POT explanation) are reference-style content that clearly belongs in a separate one-level-deep file. Not 2 because structure and navigation within the file are solid with consistent headers; not 4 because substantial separable content is inlined in SKILL.md itself. | 3 / 5 |
Total | 15 / 20 Passed |