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 well-engineered skill body: every command is concrete and runnable, errors have a lookup table with recovery actions, and detail is properly pushed to one-level-deep reference files. The only deductions are minor redundancy (inline links repeated in Resources, an output excerpt duplicating examples.md) and one referenced path (tests/) that is missing from the bundle.
Suggestions
Ship the tests/ directory referenced by "python3 -m unittest discover -s tests -v" (or drop that line from Resources) — the path does not resolve in the bundle, so the command fails as delivered.
Cut the seven-line funding output excerpt from Examples and point directly to references/examples.md; it duplicates that file's content, which the body already cites twice.
Drop either the inline reference links in Output/Error Handling or the Resources bullet list that repeats them, keeping one navigation path to each reference file.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and assumes competence — no explanation of what a perpetual swap is, only the interpretive rules Claude cannot guess ("Positive funding means longs pay shorts", hourly venues multiplied by 8). Minor trimmable redundancy keeps it off the top anchor: the funding output excerpt duplicates references/examples.md, and the Resources section re-lists the three reference links already given inline in Output and Error Handling. Not 3, since nothing explains concepts Claude already knows and every section carries operational weight. | 4 / 5 |
Actionability | All seven commands are shown as complete, copy-paste-ready invocations with realistic symbols and flags ("funding BTC ETH", "options ETH --expiry 27NOV26 --large-trades 1000000" with the unit documented), plus --format/--demo/--timeout. The gap versus the top anchor: "Tests: python3 -m unittest discover -s tests -v" points at a tests/ directory that does not exist in the bundle, and the `venues` subcommand is never shown as a full runnable line. Not 3, because every user-facing command is fully executable and covers the common cases. | 4 / 5 |
Workflow Clarity | Each of the seven commands is a single unambiguous read-only action, so the simple-skill exception applies, and the Error Handling table adds explicit feedback loops the rubric rewards: rate limited → "Wait a minute and retry", timeout → "Retry, or raise --timeout", region block → "Use --venues to skip it", plus exit code 2 semantics and a named "Unavailable" convention so failures are never silent. This is read-only research, so the destructive/batch cap at 3 does not apply. | 5 / 5 |
Progressive Disclosure | Verified against the actual bundle: all three referenced files exist (implementation.md 70 lines, examples.md 365, errors.md 29), the demo recording referenced by --demo exists in assets/, and the split is exactly right — endpoints/formulas and full captured outputs live in references while the body keeps only overview, commands, and an error table. References are one level deep and doubly signaled (inline links in context plus a Resources section with one-line descriptions). The dangling tests/ mention is a maintenance command, not a disclosure pointer, and is scored under actionability. | 5 / 5 |
Total | 18 / 20 Passed |