Content
88%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.
An exemplary orchestration skill body: fully specified MCP calls, field tables, bounded feedback loops, and validation checkpoints for a batch ticket-creation workflow, with details correctly pushed to reference files and sub-skills. The only deductions are redundant restatements of threshold rules and a slightly incomplete reference-file index.
Suggestions
Add references/size-difficulty-standard.md to the 参考文件 table (it currently is only discoverable via a pointer inside rice-scoring-standard.md).
State the 24-人时 / size=13 / size=8 threshold rules once in a single authoritative section and reference it from §2.3, §2.5, and §4 instead of repeating them four times.
Consolidate the repeated "子 skill 的详细流程见 …" pointers into the existing 子技能 table to remove four redundant lines.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and operational — parameter priority rules, field tables, thresholds (24 人时, size 13, 3 retries), and an error-handling table with no filler explaining concepts Claude already knows. It is not a 5 because the same threshold rules (size=8 advisory, size≥13 forced rollback, 24-hour cap) are restated in §2.3, §2.5 twice, the 职责边界 blockquote, and §4, and "子 skill 的详细流程见 …" is repeated four times — trimming opportunities beyond "minor". | 4 / 5 |
Actionability | Concrete MCP calls with named parameters (stories_get with workspace_id/id/with_v_status, stories_create, stories_update, tapd_id_get, story_category_get), an exact required-field table with value sources, explicit decision pseudocode for category mapping, and worked usage examples. The control loop is explicitly flagged "伪流程，非代码" and is justified as orchestration over sub-skills and MCP calls, so the instruction-only skill's guidance is fully actionable. | 5 / 5 |
Workflow Clarity | The process is clearly sequenced (参数收集 → 逐需求处理 2.1–2.5 → 四阶段创建协议 → 汇总输出) with explicit validation checkpoints at every stage: v_status check with user decision, user confirmation loop with a 3-round termination condition, pre-check before creation, post-verify of all written fields, an auto-fix loop via stories_update, and a bounded (3×) rollback loop for oversized sub-stories — exactly the validate→fix→retry pattern the rubric rewards, with an error table and status checklist on top. | 5 / 5 |
Progressive Disclosure | Good structure against the actual bundle: the 参考文件 table lists three references/ files with a "何时读取" column, §-level pointers ("见 references/requirement-splitting-guide.md §4"), and detail is properly delegated to sub-skill SKILL.md files. Minor gaps keep it at 4: references/size-difficulty-standard.md exists in the bundle but is absent from the reference table (only reachable through rice-scoring-standard.md, a nested hop), and the referenced sub-skill paths tapd-story-breakdown/SKILL.md and tapd-story-score/SKILL.md are not present in this bundle to verify. | 4 / 5 |
Total | 18 / 20 Passed |