Content
71%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 delivers a well-structured, schema-first CLI orchestration guide: a disciplined capability-discovery workflow with error-recovery loops, concrete commands, per-command reference files, and useful constraints (GUID vs display number, due/repeat_rule ordering). Its weaknesses are redundancy — the search-vs-list decision logic repeated three times and an inlined 40-line permission table — plus the absence of any caution/validation rule for destructive delete operations.
Suggestions
Merge the three near-duplicate search-decision blockquotes (任务搜索技巧 / 任务清单搜索技巧 / 意图区分补充) into a single decision rule stated once, noting it applies to both tasks and tasklists, to cut substantial repeated tokens.
Move the static permission table to a reference file (e.g., references/lark-task-permissions.md) and keep only the scope-lookup rule inline, since the body itself states that live CLI help/schema is authoritative and the table can drift.
Add a brief validation rule for destructive operations (tasks.delete, tasklists.delete), such as confirming the resolved task/list GUID with the user before executing the delete.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is operational throughout with no explanations of concepts Claude already knows, but the search-vs-list decision logic is restated three times in near-verbatim blockquotes ("任务搜索技巧", "任务清单搜索技巧", "意图区分补充") and a ~40-line static permission table is inlined in the main file despite the body itself declaring live CLI help authoritative. This fits 'mostly efficient but includes some unnecessary explanation or could be tightened' better than the minor-trims anchor at 4. | 3 / 5 |
Actionability | Concrete executable commands appear throughout ("lark-cli schema task.<resource>.<method>", "lark-cli task tasklists tasks --params '{\"tasklist_guid\":\"<tasklist_guid>\"}' --as user") alongside a discovery procedure tied to real CLI output. It falls short of the fully copy-paste-ready anchor at 5 because the most common cases (create, complete, assign) have no inline example and delegate entirely to reference files. | 4 / 5 |
Workflow Clarity | The command-discovery flow is a numbered sequence with an explicit error-recovery loop ("遇到 unknown_subcommand 时必须停止猜测或尝试变体,回到第 2 步重新发现能力") and pre-execution validation ("调用 API 前必须先查看参数结构"), and the locate-and-modify path adds post-execution verification via returned server fields. It is not 5 because destructive operations (tasks.delete, tasklists.delete, comments.delete) are listed with no confirmation or validation guidance. | 4 / 5 |
Progressive Disclosure | The body is a clear overview whose shortcut table maps each of 18 commands to its own dedicated reference file; all referenced files exist and link only back to SKILL.md and the shared skill (verified one level deep, no nested references). This matches the top anchor of a clear overview with well-signaled one-level-deep references and easy navigation. | 5 / 5 |
Total | 16 / 20 Passed |