Skill v1.0.1

version: "1.0.1" name: Annotation Pass description: Prepare PDF annotations and request approval before writing them. allowed-tools:

• get_outline
• get_paper_section
• get_pages
• search_paper_content
• get_annotations
• inspect_pdf_page
• propose_annotations
• propose_highlights
• commit_annotations
• update_annotation
• delete_annotation
• open_item

triggers:

• annotate
• 标注

The host freezes each PDF's pre-existing colors when it joins this task and remaps conflicts automatically. Use the final colors returned in previews and receipts in the report legend. The task chat has one persistent annotation batch across follow-ups, retries and Agent changes; new and branched tasks have new batches. Only verified Confucius Agent annotations may be changed with update_annotation (comment and/or a same-PDF text anchor), update_annotation_comment, or delete_annotation, across tasks and Agents. Keep the original batch and creation source; unknown old marks remain existing annotations. Do not add batch tags to native annotations. Existing tags stay unchanged. Labels, author names and colors never authorize changes. After interruption, use receipts and reconcile unknown writes; never recreate an already saved or deleted annotation.

Follow an explicit output-language request first, otherwise the current substantive user request, then the most recent effective conversation language, with the interface language only as a fallback. A bare skill command, retry or continuation inherits the conversation language. This applies to any language, including progress, annotation comments, artifact titles, report prose and revisions. Preserve source quotations and annotation anchors in their original language.

Prepare annotations for the user's reading goal. Respect the requested count and scope. If optional preferences are absent, use the settings below without asking follow-up questions.

Use this default legend unless the user overrides it for this task:

• yellow highlight (#ffd400) = key point
• blue underline (#2ea8e5) = supporting detail
• purple image-region note (#a28ae5) = visual evidence with an explanation

Read physical pages with get_pages and existing marks with get_annotations. Text contains [anchor:ID] before selectable passages. Copy the ID into annotation.anchor, and write a useful comment in the user's response language. Omit page and quote; the tool resolves the exact native text and positions. Preserve the claim's conditions and denominators and distinguish source findings from your interpretation. Image-region notes need a comment and coordinates from inspect_pdf_page. Inspect at most one visual page per model round. If no page image is returned, omit the region instead of guessing its coordinates. Follow any colors, legend, method-summary format, note voice, or focus set by the user.

Call commit_annotations directly with annotations:[{anchor,comment}], one entry or a batch. Type defaults to highlight; underline and color are optional. Use propose_annotations only for a saved candidate draft the user requests. The commit tool approval dialog handles consent, so do not request it in chat. After partial success, repair only unresolved entries; preserve existing marks and user edits. Never retry a denied write or an unchanged non-retryable failure. Create an annotation_set artifact with its legend and the actual saved/skipped outcomes. Use only zoteroUri values returned by tools. Use the user's response language for progress, comments, artifacts and the final answer; keep source quotations in their original language.