Conciseness

The body is dense with non-obvious operational knowledge (e.g., '文字颜色必须用 color 属性而不是 fontColor', 'lineSpacing="multiple:xx"', '图片必须用 <img> 而不是 <image>', rgba gradient format with stop percentages) and explains nothing Claude already knows. However, the +replace-slide vs +update-slide routing guidance is repeated ~4 times (Quick Reference table, editing paragraph, Shortcuts table, 核心规则 7) and the read-planning-docs mandate appears in three CRITICAL blocks plus 执行前必做 plus Step 1 — minor trimming opportunities, matching anchor 4 rather than 3 since the repetition is deliberate emphasis on error-prone decisions, not padding.

Actionability

Everything is executable: copy-paste commands ('lark-cli wiki +node-get --node-token ... --as user --format json', 'lark-cli auth login --domain slides'), exact attribute syntax with a full gradient example, exact paths (.lark-slides/plan/<deck-or-task-id>/slide_plan.json), a machine-checkable lint gate ('summary.error_count 必须为 0'), and numeric limits (960x540, 20 MB, 10 pages per screenshot batch, 40px margins) covering all common cases.

Workflow Clarity

The four-step 生成流程 has explicit validation gates: mandatory xml_lint.py before submission, post-create read-back via slides +xml-get, verification per workflow/validation-xml.md, and an error-recovery feedback loop per workflow/error-handling.md ('失败或部分成功按 workflow/error-handling.md 处理；局部问题优先用 +replace-slide 修正'). Destructive deletes require read-back confirmation first, satisfying the feedback-loop requirement for destructive/batch operations — matching the top anchor.

Progressive Disclosure

The body acts as a hub with well-signaled, one-level-deep references (references/cli/*, references/workflow/*, references/xml/*, references/planning-layer.md), and all ~20 referenced paths were verified to exist; heavy assets (254 KB XML schema, 747 KB icon index) are correctly kept out of SKILL.md. Two minor gaps keep it below the top anchor: the CRITICAL-mandatory reference ../lark-shared/SKILL.md does not exist in the bundle (unverifiable/missing), and the ~65-line inline 豆包设计原则 design guidance could arguably live in a reference file like the other planning layers.

Specificity

Enumerates multiple concrete actions — "创建演示文稿、读取幻灯片内容、管理幻灯片页面（创建、删除、读取、局部替换）" — plus explicit routing and exclusion rules, giving comprehensive coverage of the skill's operations with no gaps, matching the top anchor rather than the 'minor gaps' of 4.

Completeness

Both questions are explicitly answered: the what ("创建和编辑幻灯片。创建演示文稿、读取幻灯片内容、管理幻灯片页面") and the when ("当用户需要创建或编辑幻灯片、读取或修改单个页面时使用"), plus an additional concrete trigger (doubao.com /slides/ URL/token). This matches the top anchor of explicit what AND when with concrete trigger phrases; score 4 would require the when to be less explicit.

Trigger Term Quality

Natural terms like 幻灯片、演示文稿、创建/编辑/读取/修改单个页面 and the /slides/ URL/token trigger are present, but common synonyms users actually say — "PPT" and ".pptx" (used throughout the skill body) — are missing from the description, matching 'good keyword coverage; a few natural terms missing' rather than comprehensive synonym coverage.

Distinctiveness Conflict Risk

The description explicitly fences sibling skills — "不负责：云文档内容编辑（走 lark-doc）、…画板对象（走 lark-whiteboard）、上传或下载普通文件（走 lark-drive）" — and adds a domain-based routing rule, giving a clear niche with minimal conflict risk.