Content
78%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 operational body: copy-paste commands, precise metric semantics, honest reliability gating, confounder checks, and a handoff format make it highly actionable, and it wastes few tokens on things Claude already knows. Weaknesses are navigational and structural: two referenced files are missing from the bundle, caveats are duplicated, and the multi-tool workflow is never given an explicit order.
Suggestions
Fix the dangling references: either add references/papers/INDEX.md and docs/OTHER-WATCHES.md to the bundle or repoint those mentions at files that exist (e.g., fold the papers index into references/library.md, which already tracks read depth per paper).
Deduplicate repeated caveats — state 'coaching conventions, not published cut-offs' and 'compare only with the athlete's own similar sessions' once and cross-reference from the other section — and split the single-paragraph 'Other watch brands and Strava' section into short labeled rules.
Add a brief explicit step order for the workflow (e.g., 1. analyze_activity.py for the session in question, 2. aerobic_trend.py for context if 4+ reliable sessions, 3. hold_check.py when durability is the question, 4. apply confounder checks, 5. emit handoff block) so the tool-selection sequence is stated rather than inferred.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Efficient and assumes Claude's competence — no basic-concept explanations, and operational content (thresholds, reliability flags, fallbacks) dominates — but it could be tightened: the caveat 'coaching conventions, not published cut-offs' appears in both the decoupling bullet and the All sports section, 'compare only with the athlete's own similar sessions' is likewise stated twice, and the 'Other watch brands and Strava' paragraph is a single dense wall. Not 5 because of this duplication and overlong compound sentences; not 3 because nothing explains what Claude already knows. | 4 / 5 |
Actionability | Fully executable: copy-paste commands with flags for all three scripts ('python -X utf8 "<skills>/rampwise-aerobic-base/scripts/analyze_activity.py" ACTIVITY_ID --json', '--days 56 --sport all', '--sport bike --days 30 --ceiling 150'), exact metric definitions with thresholds and units, the exact Strava stream field names to save, a literal handoff block template, install instructions, and download-failure fallback. Covers the common cases completely. | 5 / 5 |
Workflow Clarity | The flow (run script → check reliability flags → check confounders via sibling skills → emit handoff block) is discernible with real validation checkpoints ('A flagged number is **not a fitness signal**; say "no valid reading"', 'fewer than 4 comparable, reliable sessions there is **no trend**'), an error-recovery loop ('If a download fails, report it and fall back to the summary only'), and an output checklist. Falls short of 5 because the sequence is never made explicit — no ordered steps or decision guidance for which of the three scripts to run in what situation, leaving the reader to assemble the order from prose. | 4 / 5 |
Progressive Disclosure | Good instinct — computation is delegated to scripts (all three referenced scripts exist) and evidence detail lives in references/evidence-map.md — but scored against the actual bundle, two referenced paths dangle: 'references/papers/ (INDEX.md lists the read depth)' and 'docs/OTHER-WATCHES.md' do not exist in the bundle, and the body never mentions the existing references/library.md while evidence-map.md chains into it (a second reference level). Missing referenced files break navigation, which is more than the 'minor organization gaps' of anchor 4; the clear signaling and appropriate splitting keep it above anchor 3's muddled-structure profile, so it sits at 3. | 3 / 5 |
Total | 16 / 20 Passed |