Content
100%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.
This is top-tier skill authorship: a lean overview that routes by deal type to one-level-deep references, an adversarial rule set expressed as concrete failure modes rather than generic risk language, and a fully specified 10-section output contract with an explicit self-check before emitting. Time-sensitive figures are confined to the reference snapshots with a LAST_UPDATED stamp rather than inlined. No weaknesses worth a deduction were found under this rubric.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and telegraphic with zero concept explanations — it assumes competence on LP/GP, waterfalls, J-curves, and cap rates — and every line carries an instruction (routing tables, nine one-line skepticism rules, six anti-patterns, ID-cited triggers). Nothing can be cut without losing function, so it sits at anchor 5 rather than anchor 4's 'minor instances of over-explanation'. | 5 / 5 |
Actionability | Guidance is fully executable for an instruction-only skill: the routing table maps each deal type to the exact reference section, flag prefixes, and comparator; the 10-section output schema specifies each section's content; scripts are named with a concrete usage constraint ('pass 0 for undisclosed fees'); and the Artifacts table plus enumerated JSON fields cover the common request cases. The only unshown detail is full script argparse syntax, which the skill's own progressive-disclosure model delegates to the script — consistent with anchor 5's 'specific examples cover the common cases'. | 5 / 5 |
Workflow Clarity | The three-step sequence (Classify & parse → Route by deal type → Analyze & assemble) is clear, with branch tables covering all five deal types, an explicit verdict decision tree ('too little to underwrite the core return story? Then Pass as presented — insufficient disclosure'), and a pre-emit validation checkpoint ('Self-check before emitting: every flag family touched in §§2–3 appears by ID in §5, and every stated pref / waterfall / debt term has its 04 probe cited by ID in §8'). This matches anchor 5's clear sequence with explicit validation and a checklist; the destructive/batch cap does not apply to an analysis-only skill. | 5 / 5 |
Progressive Disclosure | SKILL.md is a lean process overview, and the five reference files are one level deep, clearly signaled by the routing table's 'Load when / Gives' columns with per-deal-type slices (all five files and both scripts verified to exist). The references/data/ snapshot layer is a provenance chain cited by 05 and the scripts rather than instructional nesting Claude must traverse, so it does not constitute the 2+-level nesting of anchor 1–3 examples. | 5 / 5 |
Total | 20 / 20 Passed |