xhs-native-ops

A shared OpenClaw/Codex skill for turning Xiaohongshu topics into review-ready, platform-native publish packages.

It handles the execution layer: research synthesis, topic and angle design, page-level card planning, visual packaging, pre-publish checks, and review-ready handoff.

Why this repo exists

Most Xiaohongshu automation projects solve one narrow problem:

• they can publish
• or they can scrape
• or they can generate images

xhs-native-ops is opinionated about the whole package:

• the content should feel Xiaohongshu-native
• the output should be page-based, not article-dumped
• the operator should review before publish
• logged-in actions should stay on a controlled Chrome attach path

This repo is for teams that want a shared skill, not a one-off script.

What you get

• a standalone skill folder that can be copied into an OpenClaw workspace
• Python scripts for title checks, card generation, scoring, and checklist generation
• SVG-based cover and card templates
• a sample content package with generated outputs for smoke testing
• a structure that is easy to export, version, and adapt for GitHub

Workflow

flowchart LR
    A["Research signals"] --> B["Topic and angle design"]
    B --> C["Page-level card outline"]
    C --> D["Visual packaging"]
    D --> E["Score and checklist"]
    E --> F["Human review"]
    F --> G["Manual publish in Chrome attach"]

Loading

3-minute quick start

1. Install the skill

bash install-into-openclaw-workspace.sh

Default install target:

• ${OPENCLAW_HOME:-$HOME/.openclaw}/workspace/skills

Custom install target:

bash install-into-openclaw-workspace.sh /path/to/skills-root

2. Run the sample package

From the repo root:

python3 xhs-native-ops/scripts/title_check.py \
  --title "做二次元出海后我最想交给AI的五件事" \
  > xhs-native-ops/examples/sample-package/titles.json

python3 xhs-native-ops/scripts/md_to_xhs_cards.py \
  --input xhs-native-ops/examples/sample-package/package.md \
  --output-dir xhs-native-ops/examples/sample-package/cards \
  --layout six

python3 xhs-native-ops/scripts/post_score.py \
  --input xhs-native-ops/examples/sample-package/package.md \
  > xhs-native-ops/examples/sample-package/score.json

python3 xhs-native-ops/scripts/publish_checklist.py \
  --input xhs-native-ops/examples/sample-package/package.md \
  --cards-dir xhs-native-ops/examples/sample-package/cards \
  > xhs-native-ops/examples/sample-package/publish-checklist.md

3. Check the outputs

Expected outputs:

• xhs-native-ops/examples/sample-package/titles.json
• xhs-native-ops/examples/sample-package/score.json
• xhs-native-ops/examples/sample-package/cards/
• xhs-native-ops/examples/sample-package/publish-checklist.md

Who this is for

This repo is a good fit if you want:

• a shared skill for OpenClaw or Codex
• review-ready Xiaohongshu packaging, not just raw automation
• a page-level content workflow
• a controlled publishing boundary for logged-in browser work

This repo is not a good fit if you want:

• a fully autonomous growth agent
• one-click no-review auto publishing
• account-specific persona logic baked into the skill
• a generic social scheduler for every platform

Hard boundaries

• shared skill only; no account-specific persona
• compatible with upstream research outputs, but does not require them
• final publish click remains human-owned by default
• logged-in publish work must use Google Chrome + Chrome DevTools MCP attach
• PinchTab must not be used for logged-in publish actions
• relay-style browser paths are out of scope

Repo layout

xhs-native-ops/
├── README.md
├── .gitignore
├── install-into-openclaw-workspace.sh
└── xhs-native-ops/
    ├── SKILL.md
    ├── _meta.json
    ├── agents/openai.yaml
    ├── references/
    ├── scripts/
    ├── assets/
    └── examples/sample-package/

Key files

• skill entry: xhs-native-ops/SKILL.md
• sample run contract: xhs-native-ops/references/sample-run.md
• artifact formats: xhs-native-ops/references/artifact-formats.md
• implementation pitfalls: xhs-native-ops/references/implementation-pitfalls.md

What the skill actually produces

The working package contract is:

• package.md
• titles.json
• score.json
• cards/
• publish-checklist.md

The generated cards/ directory is page-level output, not article-style output.

Publishing model

xhs-native-ops is review-first by default:

• the agent can build the package
• the agent can score and checklist the package
• the agent can prepare the Chrome attach workflow
• the final publish confirmation stays human-owned unless explicitly overridden

Status

Current repo state is a GitHub-ready project skeleton for the xhs-native-ops skill, with a runnable v0.1 sample package.