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.
An exemplar instructional skill body: executable commands with real backing scripts, precise decision thresholds, a concrete handoff contract, explicit edge-case and safety rules, and disciplined evidence-strength labeling — all in about 50 lines with no filler. The only defect is a broken bundle reference: `references/papers/` (and its `INDEX.md`) is cited in the Papers section but absent from the shipped bundle.
Suggestions
Fix the Papers section's broken reference: either ship `references/papers/` with the promised `INDEX.md` (listing read depth per paper, as library.md's status column already does) or point the section at the existing `references/library.md`.
Add a one-line link from the body to `references/library.md` (e.g. in the Papers section) so the citation index is discoverable from the top level instead of only via evidence-map.md.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The ~50-line body is dense with content Claude cannot be assumed to know — exact flag thresholds, the baseline band definition, the verdict rule, the handoff block, and evidence-strength limits — with zero padding and no explanation of basic concepts (it never explains what HRV or sleep debt are). Re-checked against anchor 4: the only candidate for trimming is the triple-restated `<skills>` path note, but that ambiguity across harnesses justifies it. | 5 / 5 |
Actionability | Four copy-paste command variants (`--brief`, `--json`, `--stress 4`) plus `--date/--days/--trend` options, all pointing at a script that exists in `scripts/readiness_check.py`; the handoff block is a concrete fill-in template with enumerated flag codes and confidence conditions. This matches the fully-executable anchor covering the common cases; no pseudocode or vague direction remains. | 5 / 5 |
Workflow Clarity | Single-task skill (run the readiness check, emit the handoff) with an unambiguous action, and the edge cases that matter are explicitly handled: "Missing data for today (watch not synced) = say so; use the latest day and label it", confidence-downgrade conditions, the AMBER-one-flag escalation prompt, and the stop-and-refer medical rule. The operation is read-only, so the destructive/batch validation cap does not apply. | 5 / 5 |
Progressive Disclosure | Good structure overall: the body is a lean overview, evidence detail is correctly pushed to `references/evidence-map.md` (which exists and is clearly signaled), and the bulk paper data stays out of SKILL.md. Not 5 because the "Papers" section points to `references/papers/` with an `INDEX.md` that does not exist in the bundle, and `references/library.md` is only discoverable via evidence-map.md rather than from the body — minor organization/navigation gaps per the anchor-4 description. | 4 / 5 |
Total | 19 / 20 Passed |