XHS-Ops: Xiaohongshu Operations Toolkit

Xiaohongshu (小红书) end-to-end operations skill: hot topic research, post writing with built-in audit, automated commenting with rate limiting, and cover image...

Security Scan

The package is internally consistent with its stated Xiaohongshu operations purpose: it uses a local MCP service, Playwright for cover screenshots, and local scripts for search, write, audit and comment workflows — nothing requests unrelated credentials or hidden endpoints.

Purpose & Capability

Name/description (Xiaohongshu ops: trending search, write/audit, comment, cover) align with code and SKILL.md. Scripts call a local MCP service and use Playwright for screenshots, which are reasonable for the declared features.

Instruction Scope

SKILL.md and scripts limit actions to searching feeds, generating prompts/audits, posting comments via the local MCP, and rendering covers. Data written is mostly /tmp/xhs-search-results.json and optional ~/covers; comment posting requires per-feed xsec_token obtained from search results. The skill enforces a sensitive-word blacklist and comment rate limits in code.

Install Mechanism

No formal install spec in registry (instruction-only), but scripts/setup.sh instructs installing Python deps and Playwright and suggests downloading xiaohongshu-mcp from a GitHub releases URL. Downloading/executing a third‑party binary is expected for this skill but carries normal risks—verify the release and platform binary before running.

Credentials

The skill requests no environment variables, credentials, or config paths in the registry metadata. Runtime requires a locally running MCP service and the user's login for that binary (external to the skill). There are no unexpected credential requests in code or SKILL.md.

Persistence & Privilege

always is false and the skill does not request persistent platform privileges or modify other skills. It runs as user-invoked tools and relies on a local service; autonomous invocation is allowed by default but not exceptional here.

Assessment

This skill appears coherent and implements the features it advertises, but a few practical cautions: 1) setup.sh suggests downloading and running a platform-specific xiaohongshu-mcp binary from GitHub releases — treat any third‑party binary as sensitive: verify the upstream repo, checksum, and platform build before executing, or run it in an isolated VM/container. 2) Playwright will download browser binaries during install; that is expected for cover generation. 3) Automated commenting interacts with your Xiaohongshu account via the MCP service — ensure you control the account/login used, understand posting limits and platform rules (automated comments can risk rate limits or account action). 4) The scripts save search results to /tmp/xhs-search-results.json and read config.json in the skill folder; review and edit config.json before running. If you want higher assurance, inspect the included Python files locally and run the MCP binary in an isolated environment before connecting it to any real account.

4 tools for Xiaohongshu content operations, all backed by xiaohongshu-mcp.

Prerequisites

1. xiaohongshu-mcp binary running at http://localhost:18060/mcp
2. Playwright Chromium installed (for cover generation)
3. Run bash scripts/setup.sh for one-click dependency install

# Quick start
bash scripts/setup.sh
# Verify MCP is running
curl -s http://localhost:18060/mcp

Tools

🔥 search-hot — Find trending topics

python3 scripts/search_hot.py                        # Use default keywords from config
python3 scripts/search_hot.py "Claude Code" "AI写代码" # Custom keywords

Returns Top 10 posts sorted by likes: title, author, likes/collects/comments, source keyword. Saves results to /tmp/xhs-search-results.json for other tools to consume.

📝 write-post — Generate post content

Agent workflow — not a standalone script. Use the prompt and audit functions:

from scripts.write_post import get_write_prompt, audit_post, format_audit_report

# 1. Generate writing prompt
prompt = get_write_prompt("3个AI Agent每晚自动开会", reference="optional context")

# 2. Send prompt to LLM, get back title + body + tags

# 3. Audit before publishing
result = audit_post(title="标题", body="正文", tags="#tag1 #tag2")
print(format_audit_report(result))

Audit checks: sensitive words, title length (≤20), body length (≤500), trailing question, numeric specificity, tag count (5-8).

💬 comment — Automated commenting

python3 scripts/comment.py --dry-run --count 5   # Preview only
python3 scripts/comment.py --count 10             # Post comments

Requires search-hot results. Random interval 3-8 min between comments. Daily limit configurable.

For single comment via Agent:

from scripts.comment import post_single_comment
result = post_single_comment(feed_id, xsec_token, "评论内容")
# Returns {"status": "ok"} or {"status": "blocked", "reason": "敏感词: ..."}

🎨 cover — Generate cover images

python3 scripts/cover.py "主标题" "副标题" -o ~/covers/output.png

Uses hand-drawn notebook style template. Playwright screenshots at 2x scale → 1200×1600 PNG. Replace assets/cover_template.html to change visual style.

Configuration

Edit scripts/config.json:

{
  "keywords": ["Claude Code 实战", "AI写代码 程序员"],
  "mcp_url": "http://localhost:18060/mcp",
  "comment_interval_min": 180,
  "comment_interval_max": 480,
  "daily_comment_limit": 15
}

Content Rules

See references/content_rules.md for:

• Writing style guide (接地气、短句、移动端友好)
• Sensitive word blacklist (platform names, absolute claims, solicitation)
• Post structure template (hook → points → details → CTA)
• Comment generation guidelines

Oral Script Template

See references/koubo.md for converting posts to video narration scripts.

MCP Session Notes

• Each tool call requires a fresh initialize to get a Session ID
• xsec_token from search results cannot be reused across sessions
• Cookie expires ~30 days; re-login: ./xiaohongshu-login-darwin-arm64