Content
82%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 disciplined, well-structured instruction skill: terse checklists, a clear mode-to-output-to-quality-gate flow, and no padding. Its main gaps are the absence of concrete research mechanics (named sources or search strategies for each mode) and no error-recovery guidance when the Quality Gate fails, plus a modest opportunity to split per-mode detail into reference files rather than keeping everything in SKILL.md.
Suggestions
Actionability: add one line of concrete research mechanics per mode — e.g. named sources for fund diligence (fund website, portfolio pages, SEC/crunchbase-style databases) and for market sizing (analyst reports, census/public datasets) — so 'Collect' becomes executable rather than a topic list.
Workflow clarity: extend the Quality Gate with a short recovery loop (e.g. 'If a number cannot be sourced, label it an estimate and note it in risks; if data is stale, state the as-of date') so failures have defined next steps.
Progressive disclosure: consider moving the four research-mode checklists into one-level-deep reference files (e.g. references/modes.md) or a per-mode deep-dive with source lists, keeping SKILL.md as the lean overview.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and efficient — terse bullet checklists ('fund size, stage, and typical check size', 'product reality, not marketing copy') with zero padding and no explanation of concepts Claude already knows. 'Produce research that supports decisions, not research theater' earns its tokens by setting the standard in one line. This matches the anchor 'lean and efficient; assumes Claude's competence; every token earns its place' and is not a 4 because there is no over-explanation to trim. | 5 / 5 |
Actionability | The guidance is concrete for an instruction-only skill: each research mode has a specific collection checklist (e.g., Market Sizing specifies 'top-down estimates from reports or public datasets' and 'bottom-up sanity checks'), a default output structure is given as a numbered list, and the Quality Gate is a concrete pre-delivery checklist. It falls short of 5 because nothing names actual executable research moves — specific sources (SEC filings, Crunchbase, industry report outlets) or search strategies — leaving 'Collect' and 'Use' at the level of well-specified hints rather than fully executable steps. | 4 / 5 |
Workflow Clarity | A clear sequence is present — activation triggers, research standards, per-mode collection lists, output format, then a Quality Gate that functions as an explicit validation checkpoint before delivering. This matches 'clear sequence with most checkpoints present; minor validation gaps' — the gate says what to check but gives no recovery loop (what to do when a check fails, e.g. where to go find a source or how to flag an unfixable gap). It is not a 3 because validation is explicit rather than absent, and not a 5 because there is no error-recovery guidance. | 4 / 5 |
Progressive Disclosure | The body is well-sectioned (When to Activate, Research Standards, Common Research Modes with four compact sub-modes, Output Format, Quality Gate) and no content is inlined that clearly belongs in a separate file — each mode's checklist is appropriately brief for an overview. It does not reach 5 under the guideline reserving that score for short skills (under 50 lines) with well-organized sections alone: at ~70 lines with four research modes, the modes are natural candidates for one-level-deep reference files (e.g., a per-mode deep-dive or source list), which are absent along with any bundle. | 4 / 5 |
Total | 17 / 20 Passed |