# Xhs Claude Opus Refine

> Refine Xiaohongshu card markdown with local Claude Opus using red/blue/layout/cover passes, then hand off to direct renderer. Use when users ask for 狠改文案、去AI味、提高点击、避免重叠和短句可读.

- Skill: `yanghu819/xhs-claude-opus-refine` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add yanghu819/xhs-claude-opus-refine`
- Raw SKILL.md: https://api.skillmd.com/api/skills/yanghu819/xhs-claude-opus-refine/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Docs & Writing
- Author: yanghu819 (https://skillmd.com/u/yanghu819)
- Updated: 2026-09-10
- Page: https://skillmd.com/skills/yanghu819/xhs-claude-opus-refine

---


# xhs-claude-opus-refine

Use this skill to run deterministic copy refinement through local Claude CLI (Opus) before rendering.

## Inputs
- Source markdown path (10-card XHS markdown)
- Output root directory
- Pass sequence (default: red -> blue -> layout)
- Max line chars (default: 14-16)
- Claude proxy (default local 7897)

## Workflow
1. Red pass (`--pass red`): amplify hook and urgency.
2. Blue pass (`--pass blue`): improve actionability and compliance.
3. Layout pass (`--pass layout`): shorten lines and reduce overlap risk.
4. Optional cover pass (`--pass cover`) for first-card only punch.
5. Validate output:
   - frontmatter present
   - card count = 10
6. Render with `xhs-direct-render` skill.

## Primary script
- `python3 scripts/refine_markdown_with_claude.py --input <in.md> --output <out.md> --pass <red|blue|layout|cover> --model opus --effort high --max-line-chars 14`

## Quality gates
- Card1 must contain sharp pain hook.
- Card9 must be checklist (`- [ ] ...`).
- Card10 must include quote/closing CTA and no truncation.
- No forbidden absolute terms.

## Mandatory memory update
- Append each run summary to `../state/YYYY-MM-DD-<topic>.md`.
- Promote stable defaults into `../LEARNINGS.md`.
- Commit + push to `main`.