Content
82%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.
The body is a well-engineered, highly actionable read-only status skill: exact thresholds, priority-ordered trigger rules with executable commands attached, explicit no-side-effect and don't-guess rules, and a concrete output template. Its weaknesses are modest — a somewhat long mock output, under-specified judgment heuristics, and reliance on shared-reference files outside the bundle.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and functional — tables for inputs and derived metrics, constants named once, no explanations of concepts Claude already knows. Minor trimming is possible: the ~40-line mock dashboard and the verbose default/earlier/later sub-bullets on trigger rules 7 and 19 could be tightened. | 4 / 5 |
Actionability | Guidance is fully concrete: exact thresholds ("SQLITE_UPGRADE_THRESHOLD = 30", "buffer_days = buffer_count × target_publish_cadence_days", "<1 红 / 1-2 橙 / 3-5 绿 / >5 蓝"), exact commands attached to every suggestion ("/cheat-bump --propose", "/cheat-seed", "/cheat-trends"), and a complete copy-paste output template covering the common case. | 5 / 5 |
Workflow Clarity | A clear 4-phase sequence (read state → derive metrics → detect triggers by priority → output) is presented with an explicit flowchart, and checkpoints exist for missing state ("你还没初始化。请先跑 /cheat-init") and missing fields ("显式标'未知'，不猜"). Minor gaps remain — e.g., undefined behavior for malformed predictions files and the 'Claude 判断' heuristics on rules 7/19 lack verification steps — though as a read-only skill there are no destructive/batch operations requiring a validation cap. | 4 / 5 |
Progressive Disclosure | Structure is good — Overview, Constants, Inputs, phased Workflow, Key Rules, Refusals, Integration — with clearly signaled one-level-deep links ([cadence-protocol.md](../../shared-references/cadence-protocol.md), [state-management.md](../../shared-references/state-management.md)); no bundle files exist in this skill and the inline trigger rules are its core working content. Minor gap: the long example dashboard could arguably live in a reference file, and the referenced shared-reference paths sit outside the skill bundle. | 4 / 5 |
Total | 17 / 20 Passed |