Content
81%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 rigorous, project-grounded specification with excellent workflow sequencing, explicit validation and crash-recovery handling, and highly concrete APIs and file conventions. Its main gaps are pseudocode rather than executable examples, some repetition of key invariants across sections, and a monolithic structure that could offload detail to reference files.
Suggestions
Add one minimal executable TypeScript example — a registry entry (with progressSteps, i18nT keys, mode flags) and a skeletal task using context.getCheckpoint/reportFailedRecords/saveCheckpoint — so the guidance is copy-paste ready.
State the checkpoint-vs-error-snapshot persistence ordering rule once in a canonical section and reference it elsewhere, instead of repeating it in 状态能力, 分批断点续跑, and 交付前检查.
Move the detailed test-scenario matrix and the scheduling-mode/lease semantics into a reference file (e.g., references/testing.md, references/scheduling.md) to keep SKILL.md closer to an overview.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense, project-specific rule-writing with no padding of concepts Claude already knows (no explanation of idempotency, Mongo, or leases), but the checkpoint/error-snapshot ordering invariant is restated in "脚本可使用的状态能力", "分批断点续跑" 关键约束, and again in "交付前检查". This matches the 4 anchor (efficient with minor instances that could be trimmed) rather than 3, since the repetition is deliberate emphasis of safety invariants rather than unnecessary explanation. | 4 / 5 |
Actionability | Concrete guidance throughout: exact paths to inspect ("projects/app/src/migration/registry.ts"), exact Context APIs ("context.saveCheckpoint", "context.assertActive"), a mandated directory layout, specific constraints ("_id 或其他不可变、唯一且有索引的游标;不要使用 offset", "SYSTEM_MIGRATION_DELAY_SECONDS"), and verification commands ("App typecheck、相关 ESLint 和 git diff --check"). It falls short of 5 because the two recommended loops are pseudocode and there is no complete copy-paste TypeScript example of a registry entry or a task implementation. | 4 / 5 |
Workflow Clarity | Multi-step processes are explicitly sequenced with validation checkpoints: the loop embeds "assertActive", "校验该批结果并提交事务", and "执行全局完成校验"; crash-recovery feedback loops are specified ("进程可能在业务提交后、checkpoint 保存前退出"), and "交付前检查" provides a final checklist. Validation is thoroughly present, so the destructive/batch cap of 3 does not apply; this matches the 5 anchor (explicit validation, error-recovery loops, checklist). | 5 / 5 |
Progressive Disclosure | No bundle files exist, so all content is inline in a single well-organized file with clear sections, a state-ownership table, and code blocks; navigation is easy and there are no nested or buried references. It stays at 4 rather than 5 because the ~250-line body inlines the full test-scenario matrix and detailed scheduling-mode semantics that could live in reference files, and the file is not the lean overview-plus-references shape the 5 anchor describes. | 4 / 5 |
Total | 17 / 20 Passed |