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 strong, dense instruction body: every tool call, formula, and the report template are concrete and executable, with conduct boundaries stated crisply. The only weakness is mild duplication between the inline math section and references/risk-math.md, and edge-case handling that is present but not framed as explicit workflow checkpoints.
Suggestions
Deduplicate the math: keep either the compact formulas and bands inline or in references/risk-math.md, not both — e.g. reduce 'The math' to the report-critical formulas and point to the reference for bands and worked examples.
Add explicit workflow checkpoints for data gaps, e.g. after the tool calls: 'if a symbol returns no data[] entry, report it as untracked and proceed with the rest' rather than only mentioning it in Conduct.
Surface the reference earlier with a markdown link (e.g. in the '## The math' heading) so navigation to the worked examples is available at the point of use, not only in the final line.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and assumes Claude's competence (no concept explainers, no padding), but 'The math' section and the liquidation bands (≤ 5% / ≤ 15%) restate formulas and thresholds that references/risk-math.md also carries — a minor trim opportunity that keeps it below anchor 5. | 4 / 5 |
Actionability | Fully executable instruction-only guidance: exact tool calls with parameter values ('intervals: ["1h","4h","1d"]', 'status: "open"', 'limit: 20'), batching limits ('up to three in ONE call'), concrete symbol-mapping examples (kPEPE → 1000PEPEUSDT), exact formulas, and a copy-paste report template covering the common cases. | 5 / 5 |
Workflow Clarity | Clear ordered sequence ('A position check answers, in order: ...') with an explicit call budget and quota-skip fallback, plus an untracked-coin case. It falls short of anchor 5 only because the error/edge cases are scattered in Conduct rather than framed as explicit checkpoints in the workflow itself. | 4 / 5 |
Progressive Disclosure | Well-organized sections with a clearly signaled, verified one-level-deep reference ('Formulas and worked examples: references/risk-math.md'). The split is not fully clean — the body duplicates the reference's formulas and bands — leaving a minor organization gap versus anchor 5. | 4 / 5 |
Total | 17 / 20 Passed |