Content
100%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 an exemplar instruction-orchestration skill: lean and free of padding, with concrete script interfaces, a copy-paste JSON payload, and exact enum values making the guidance fully executable. The multi-step lifecycle has explicit approval gates, validation checkpoints, and error-recovery loops, and detail is properly split into conditionally-loaded, one-level-deep reference files that all exist in the bundle.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body contains only load-bearing, harness-specific rules — e.g., "原样使用返回值,不根据名称或路径推算 ID", "计划已保存不代表用户已批准", "只有...明确批准...才可进入构建" — with zero explanation of concepts Claude already knows and no filler. Every token earns its place, matching anchor 5; anchor 4 would require identifiable over-explanation to trim, and none is present. | 5 / 5 |
Actionability | Fully executable guidance: an exact script-interface table ("create_plan.py → plan_id、规范化 plan" etc.), the precise invocation rule ("通过 execute_shell_command 运行 <python> scripts/<script>", "<python> 使用环境变量 QWENPAW_DESKTOP_PY_RUNTIME"), a complete copy-paste JSON plan payload, explicit enum mappings ("instruction/template/workflow、true/false、foreground/background 和 off/smoke/eval"), and a YAML frontmatter template — copy-paste ready with specific coverage of common cases, matching anchor 5. | 5 / 5 |
Workflow Clarity | The plan → approve → build → validate/test → publish lifecycle is clearly sequenced with explicit validation checkpoints and feedback loops: "执行任何 draft script 或 batch 前,先对初始化后的草稿运行 validate_skill.py", "按静态或安全错误修复 draft 并重新校验", "更新返回 missing-plan 时,返回规划和批准流程,不继续构建", and digest verification on publish ("将校验结果中的 digest 作为 expected_digest 传入"). This matches anchor 5's explicit validation steps with error-recovery loops; validation is present, so the destructive/batch cap does not apply. | 5 / 5 |
Progressive Disclosure | The body is an overview that delegates detail to three real, one-level-deep bundle files (references/type-and-package.md, references/behavior-testing.md, references/run-batch.md — all verified present, none nesting further), each clearly signaled with conditional-loading rules: "计划的测试模式不是 off 时...阅读[行为测试]", "只有选择 batch: true 后,才...阅读[运行 Batch];batch: false 时不要读取". Clear overview, well-signaled references, easy navigation — anchor 5. | 5 / 5 |
Total | 20 / 20 Passed |