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.
The body is highly actionable, with complete copy-paste templates, concrete quantitative thresholds, and explicit output requirements, and its workflows are clearly sequenced with validation on the risky steps. Its weaknesses are token efficiency (sections restating concepts Claude already knows, like the abbreviations table and generic number-formatting rules) and the absence of any progressive disclosure — everything, including pure reference material, is inlined in one long file.
Suggestions
Remove or drastically trim the "Common Abbreviations" table and generic number-formatting rules (e.g., YYYY-MM-DD dates, X.X% percentages) — Claude already knows these conventions; keep only the house-specific choices.
Move the terminology guide, rating tables, and formatting standards into a references/ file (e.g., references/style-guide.md) and keep SKILL.md as an overview with well-signaled one-level-deep pointers.
Add a final validation checkpoint to the writing workflow, e.g. a step that re-checks the finished report against the Output Format checklist (rating present, ≥3 risk warnings, disclaimer) before delivery.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Most of the body is genuinely house-style convention Claude would not know (rating thresholds, structure templates, terminology preferences), but several sections restate generic knowledge: the "Common Abbreviations" table (YoY, QoQ, ROE, EPS, FCF), number-formatting rules like "Date | YYYY-MM-DD", and a fully worked 15-line equity-curve example table. Not 2 because the padding is confined to identifiable sections rather than pervasive; not 4 because multiple sections (abbreviations, number formatting, parts of the chart-substitute example) could be trimmed or cut outright. | 3 / 5 |
Actionability | For an instruction-only skill the guidance is fully actionable: copy-paste-ready Markdown templates for the standard report and backtest report, concrete rating thresholds ("Sharpe > 1.5, max drawdown < 20%"), explicit terminology mappings ("pullback / correction / drawdown" vs "dropped / fell"), a 5-type classification with trigger scenarios, an explicit output checklist, and concrete PDF delivery steps via `write_file` with fields to check. The only placeholder is "[Insert output from the `risk-analysis` skill]", which is a deliberate delegation rather than a gap. | 5 / 5 |
Workflow Clarity | "Preparation Before Writing" gives a clear 4-step sequence (clarify type → collect data → determine core view → choose rating), the backtest flow specifies reading engine-written `metrics.csv`/`equity.csv` with grounding rules, and the PDF flow has an explicit checkpoint ("Check its returned `status`, `path`, and `bytes_written`") with error handling ("If rendering fails, report that tool error instead of claiming the PDF was created"). Not 5 because the main report-writing flow itself has no verify/recover checkpoint after generation (e.g., a final consistency check against the output checklist); not 3 because the sequences that exist are clear and the riskier operations (PDF write, metric grounding) do have validation. | 4 / 5 |
Progressive Disclosure | There are no bundle files at all; the entire skill is a ~280-line SKILL.md. Sections are well organized with a clear H2/H3 hierarchy, but reference-type material that clearly belongs in separate files — the terminology guide, common abbreviations, number-formatting standards, and rating tables — is fully inlined. Not 4 because at this length that material should be split into references/ files; not 2 because the body is cleanly sectioned and navigable rather than a wall of text. | 3 / 5 |
Total | 15 / 20 Passed |