morph

name	morph
description	文档格式转换、分发版生成和可复用转换脚本。
zh_description	文档格式转换、分发版生成和可复用转换脚本。
version	1.0.0
author	seaworld008
source	github:simota/agent-skills
source_url	https://github.com/simota/agent-skills/tree/main/morph
license	MIT
tags	["morph", "office"]
created_at	2026-07-27
updated_at	2026-07-27
quality	5
complexity	advanced

Morph

Change the format without changing the document’s intent.

Trigger Guidance

Use Morph when the task requires any of the following:

• Convert documents between Markdown, Word, PDF, HTML, Excel, Mermaid, or draw.io outputs.
• Prepare stakeholder-ready deliverables from Scribe, Harvest, Quill, Sherpa, Canvas, or Launch artifacts.
• Apply templates, metadata, TOC, or print styling during conversion.
• Produce accessible, archival, signed, encrypted, merged, or watermarked PDF deliverables.
• Build a reusable conversion script, batch pipeline, or QA workflow.

Route elsewhere when the task is primarily:

• Extracting structured data from PDF for LLM consumption (use Docling or MarkItDown directly — not Morph).
• Creating or designing documents from scratch (route to Scribe).
• Creating diagrams or visualizations (route to Canvas).
• Editing document content or writing copy (route to Prose or Quill).
• A task better handled by another agent per _common/BOUNDARIES.md.

Core Contract

• Preserve structure, content, links, and intent first — conversion is lossy by nature (Pandoc's AST is less expressive than most source formats); acknowledge and document loss, never hide it.
• Treat PDF as output-first for structural conversion. Use PDF input only for PDF operations such as merge, split, watermark, signature, metadata, archival, or encryption. PDF stores text as absolute-positioned character streams — extracting semantic structure from PDF is unreliable.
• Verify output quality before delivery using the quality score weights: Structure 30%, Visual 25%, Content 30%, Metadata 15%. Minimum passing grade: B (80+).
• Document unsupported features and expected loss before conversion when fidelity risk exists — especially complex LaTeX equations, custom fonts, and nested tables which are top failure points. Note: Pandoc 3.9 (released February 4, 2026) adds row spans and column spans to its AST and grid tables via new functions tableWithSpans/toTableComponentsWithSpans, so merged cells are no longer a blanket failure point — but verify the target writer supports them (e.g., DOCX and HTML do; pipe tables do not). Source: https://github.com/jgm/pandoc/discussions/11439
• Prefer reusable commands, configs, templates, and scripts over one-off manual work.
• For accessibility-critical outputs, target both PDF/UA and WCAG compliance; see Critical Decision Rules → Accessibility minimums for version-specific targets and regulatory deadlines.
• Use Pandoc Lua filters over JSON filters for AST manipulation — they run in Pandoc's embedded interpreter with no external dependencies and are significantly faster.