Content
85%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 highly actionable, well-sequenced research workflow with concrete thresholds, tool invocations, search-query templates, and built-in validation gates. Its main weakness is structure: everything lives inline in one 185-line file with no progressive disclosure to reference material.
Suggestions
Move the per-trend layer decomposition maps (AI infrastructure example, plus templates for other trends) into references/layer-maps.md and link to it from Step 2, keeping only the layer taxonomy inline.
Extract the output artifacts — the ranking table schema, one-pager template, and signal-strength scale — into references/output-templates.md so they are loaded only at Step 6.
Trim the rhetorical framing in "Core Idea" and deduplicate the valuation principles in "Core Principles" against Step 4's valuation gate, which already states the same PS/PE rules.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Mostly lean: dense criteria tables, numeric thresholds, and no explanation of concepts Claude already knows. Not 5 because the "Core Idea" section is partly rhetorical and "Core Principles" restates Step 4's valuation gates (e.g., "at PS>30x or still loss-making... not a buy" vs. the red-light rule). | 4 / 5 |
Actionability | Fully executable instruction guidance: copy-paste search queries ("{trend} supply chain bottleneck", "{trend} sole source supplier"), named tool commands with parameters (financial_rigor command=verify_valuation / three_scenario, report_audit command=extract → verdict), numeric gates throughout (>30% revenue share, <$10B, PS>30x, <10%/yr), a one-pager template, and a save-path convention. Even fallback behavior is specified (missing financials → signal ≤★★). | 5 / 5 |
Workflow Clarity | Steps 1–7 are explicitly sequenced with validation checkpoints: "Valuation Gate (mandatory, never skip)", a mandatory sanity check, a report_audit quality gate after drafting (extract → verify → verdict), and a bias self-check checklist. This is analysis (not destructive/batch) work, and the reverse-check (Munger inversion) step plus the audit loop provide feedback for error recovery. | 5 / 5 |
Progressive Disclosure | Sections are well organized under ## Step N headings, but the skill is monolithic: no references/scripts/assets files exist, and domain reference material (the ~25-line AI-infra layer map, the one-pager template, the bias table) is inlined in SKILL.md where per-trend layer maps in a references/ file would keep the overview lean. Not 2 because it is clearly sectioned rather than a wall of text. | 3 / 5 |
Total | 17 / 20 Passed |