Document design skill for zylos agents. Turns structured content (reports, proposals, digests, decision memos) into beautiful, human-friendly standalone HTML pages.
After registry publication:
zylos add doccraftPre-registry (from GitHub release):
zylos add zylos-ai/zylos-doccraft- Methodology (
references/methodology.md): Design principles — conclusion-first information architecture, progressive disclosure, dual light/dark theme, mobile responsiveness. - Template (
assets/template.html): Production-ready HTML template with CSS for status boards, collapsible sections, number cards, and responsive grids. - Skill (
SKILL.md): Trigger rules and execution workflow for the agent.
The skill is triggered when the agent needs to produce a document for human consumption. Trigger words include "可视化版", "HTML 版", "做成网页", "readable version".
The agent reads the methodology, uses the template as a starting point, and generates a self-contained HTML file. Delivery is pluggable: if the pages component is installed, the file is published there for a share link; if not, the skill hands over the file path directly. pages is a recommended companion, not a requirement — the artifact has no external dependencies of its own.
This skill is runtime-neutral — it works on both Claude Code and Codex. The execution model declares claude-sonnet-5 as a quality preference with a three-tier fallback: declared model, strongest available model, or inline execution. The output quality standard is the same regardless of execution path.
- Conclusion first — TL;DR + 3 number cards at the top
- Progressive disclosure — overview → status board → collapsible details
- Dual theme — automatic light/dark via
prefers-color-scheme - Mobile responsive — flexbox/grid, overflow-x containers
- Sensitive info scan — no internal IDs or credentials in published output