Content
92%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
This is a highly actionable, well-sequenced instruction skill with concrete commands, thresholds, templates, and a validation feedback loop. Its main weakness is monolithic inline content with no actual bundled reference files to progressively disclose into.
Suggestions
Move the detailed per-company analysis template (第三/四步) and the four-master rubrics into a referenced file (e.g. references/analysis-templates.md) so SKILL.md stays a lean overview.
Extract the AI-research-bias checklist and the 5-hard-indicator definitions into references/ so the core funnel flow stays scannable.
Verify the referenced tools (tools/report_audit.py, tools/financial-data.md) are discoverable from the skill's own bundle or clearly documented as repo-root dependencies.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense with operational specifics — funnel tables, hard thresholds, output templates, and bias checklists — and avoids explaining concepts Claude already knows; the only mild padding is the Codex adapter note, but overall every section earns its place. | 3 / 3 |
Actionability | It gives executable commands (e.g. 'python3 tools/report_audit.py extract --report ...'), exact quantitative criteria (ROE > 15%, 负债率 < 60%), and ready-to-use table and report templates, making the guidance copy-paste ready. | 3 / 3 |
Workflow Clarity | The four-layer funnel is explicitly sequenced with per-layer keep/drop rules, mandatory elimination records, and a final 数据抽检 pass/fail gate with a fix-and-re-audit feedback loop. | 3 / 3 |
Progressive Disclosure | Sections are well organized and external targets (tools/report_audit.py, sibling skills) are named, but no bundle files exist in references/scripts/assets and the bulk of the workflow is inline rather than split into one-level-deep reference files. | 2 / 3 |
Total | 11 / 12 Passed |