Content
96%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: token-efficient, fully executable commands, and a sequenced workflow with an explicit review_required validation gate and a feedback loop. The only notable weakness is navigation — the Resources section lists reference files as bare paths rather than labeling what each contains and when to consult it.
Suggestions
Annotate the Resources entries with what each file contains and when to read it (e.g., '**Stagnation triggers**: see references/stagnation_triggers.md for trigger thresholds and edge cases') so navigation is signposted like the anchor-5 pattern.
Replace wildcard arguments in the generate command (reports/pivot_diagnosis_*.json) with a concrete example filename to keep every command literally copy-paste runnable.
Move the cryptic profit-factor edge-case note in workflow step 2 into references/stagnation_triggers.md, keeping the workflow line as a pointer.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and assumes competence: a two-sentence Overview, terse bullet lists for When to Use / Prerequisites / Output, a compact 7-step workflow, and three command blocks with no padding or explanation of concepts Claude already knows. Every section earns its place, matching the anchor-5 example of lean, efficient structure. | 5 / 5 |
Actionability | Three complete, copy-paste-ready bash commands with realistic arguments (--append-eval with a concrete --changes example, --strategy-id, --max-pivots) cover the common cases of appending history, detecting stagnation, and generating pivots, plus concrete output file paths. This matches 'fully executable; copy-paste ready; specific examples cover the common cases'. | 5 / 5 |
Workflow Clarity | The 7-step workflow is clearly sequenced with an explicit validation gate ("If the diagnosis says review_required, repair or re-evaluate the underlying metrics first; do not generate pivots") and a feedback loop (step 7 feeds the selected pivot back into backtest-expert), mirroring the validate-then-proceed pattern of the anchor-5 example. | 5 / 5 |
Progressive Disclosure | Content is appropriately split: detailed triggers, archetypes, techniques, and schema live in real one-level-deep reference files (verified present in references/ with no nested references), while commands stay inline. However, the Resources section is a bare list of paths without labels, when-to-read guidance, or links, which is the minor organization gap described by anchor 4 rather than the well-signaled navigation of anchor 5. | 4 / 5 |
Total | 19 / 20 Passed |