Content
93%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, information-dense body: every section teaches non-obvious, tool-specific facts and each common question has a concrete call-and-present recipe. The only weaknesses are the thrice-repeated 'not advice / cannot copy' caveat and the absence of explicit validation checkpoints in the workflow sequences.
Suggestions
State the 'positions are evidence, not recommendations; this connector cannot copy or trade' caveat once in the Conduct section and trim its near-duplicate phrasing from the intro paragraph and the 'Should I copy trader X' workflow.
Fold the freshness check into each workflow as an explicit step (e.g., 'before answering, check the newest openTime against the feed delay; if stale, say the feed is delayed') rather than leaving it as a separate note in 'Reading the data'.
The symbol-filter guidance is excellent but split between two paragraphs — merging the 'BTC matches both BTC and BTCUSDT exactly' rule with the tokenized-stock exception into one compact rule would save a few tokens.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Nearly every paragraph carries connector-specific knowledge Claude cannot know (per-exchange symbol formats, refresh cadences, smart-score weights, Hyperliquid's null winRate), but the 'positions are evidence, not advice / no copying tool / user's decision' caveat is repeated almost verbatim in the intro, the 'Should I copy trader X' workflow, and the Conduct section — minor trimmable repetition. Not 5 because a few tokens are spent restating the same caveat three times; not 3 because there is no explanation of concepts Claude already knows. | 4 / 5 |
Actionability | Fully actionable for an instruction-only skill: exact tool names with argument values and bounds ('status open / closed / all', 'timeRange 24h / 3D / 7D / 30D', 'limit ≤ 50'), per-question call recipes, concrete computations ('win rate from closed rows, typical hold time (closeTime − openTime)'), and a ready-made presentation table. Not 4 because the guidance covers the common cases completely with no gaps. | 5 / 5 |
Workflow Clarity | The four named workflows are clearly sequenced with concrete first calls and conditional follow-ups ('If open rows are few, add status: closed'; wrong-exchange ids 'return nothing' → take source from the originating row), and empty-result handling is specified ('say the feed is delayed rather than "no activity"'). Not 5 because no explicit validate→fix→retry checkpoints structure the workflows; not 3 because sequencing, branching, and failure-mode handling are far more complete than 'checkpoints missing'. | 4 / 5 |
Progressive Disclosure | A well-sectioned overview (Tools, Reading the data, Workflows, Presenting, Conduct) with exactly one clearly signaled one-level-deep reference — 'For exchange-specific quirks and refresh cadences read references/exchange-notes.md' — and the referenced file exists (40 lines, on-topic sections, no nested references). Core operational knowledge is inline where it belongs; exchange-specific detail is correctly pushed out. | 5 / 5 |
Total | 18 / 20 Passed |