Content
56%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 content is highly actionable with concrete tool calls, executable script gates, and well-sequenced workflows backed by validation. Its main weakness is conciseness and progressive disclosure: the body is large and inlines dated case-study tables and industry-comparison material that would be better placed in separate reference files.
Suggestions
Move the 反面案例速查 and 业界对照验证 tables plus dated '来源实证' prose into a separate references/case-studies.md, leaving SKILL.md as a concise rule + gate overview that points to it.
Trim repeated '根因/来源' narrative blocks under each rule to one line; the audit-proven rationale can live in the external reference rather than inline.
Convert remaining rule-statement recovery guidance (e.g. 遇阻自救三连, Recovery Ladder) into explicit numbered validate→fix→retry steps so the feedback loop is unambiguous.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is noticeably verbose at ~34KB / 285 lines—several dated prose case studies, repeated root-cause blocks, and lengthy反面案例 tables re-explain context Claude could infer, padding the token budget beyond what each rule earns. | 2 / 5 |
Actionability | Provides concrete executable guidance—named tool calls (`job_output(wait=true)`, `queue_cli_request` with `target_model`+`effort`), real script invocations (`python <skill>/scripts/flow_check.py --check-flow`), and fixed CLI command strings—with only minor gaps where guidance stays at rule-statement level. | 4 / 5 |
Workflow Clarity | Multi-step processes are clearly sequenced (ten rules, 遇阻自救三连, Recovery Ladder, 流程完整执行 with --selftest validation and FAIL-stops-implementation gates); most validation checkpoints are present, though a few recovery loops are stated as rules rather than explicit validate→fix→retry feedback steps. | 4 / 5 |
Progressive Disclosure | Two real scripts exist (flow_check.py, takeover_check.py) and are referenced by path, but the SKILL.md is a near-monolithic inline document—large case-study tables, dated empirical notes, and 业界对照 content that clearly belongs in separate reference files are inlined rather than split out, so structure is present but not cleanly one-level-deep. | 3 / 5 |
Total | 13 / 20 Passed |