Xiaohongshu Auto Posting

Xiaohongshu Auto-Posting (XHS Auto-Posting)

End-to-end Xiaohongshu content operation: topic discovery → style reference → writing → publishing → performance tracking.

Language

All process output to user (progress updates, questions, status notifications) follows the user's language.

Objective

Automate a complete Xiaohongshu content operation cycle for a configured product: discover high-engagement pain-point topics via search, collect writing style references, generate platform-native content aligned with the style fingerprint, publish with user approval, and track post-publish performance metrics.

Prerequisites

• Xiaohongshu creator center (creator.xiaohongshu.com) already logged in on a configured stealth browser (browser ID stored in session_state.json)
• workspaces/xhs-posting/ workspace (auto-created on first run)

Pre-execution Checks

1. Tool Readiness

If browser-act has been confirmed available in the current session → skip this step.

Invoke browser-act via Skill tool to load usage. If installation or configuration issues arise, follow its guidance to resolve then retry.

2. Login Verification

Covered in Phase 0.5 — login check is deferred until session state is loaded (Phase 0.2) and the browser is opened. If the user triggers Phase 6 directly (standalone tracking mode), still complete Phase 0.2 and Phase 0.5 first.

3. Session Lifecycle

Each execution generates a unique session name (xhs-<YYYYMMDD>-<HHMM>) to avoid cross-conversation conflicts. The session is opened once in Phase 0.5 and closed at the end of the workflow. All commands in this Skill reference the session via $SESSION — the actual name is determined at runtime.

Runtime Workspace

All runtime files (config, drafts, publish records, screenshots) are written to the user's working directory (CWD) under workspaces/xhs-posting/. This directory is auto-created on first run.

workspaces/xhs-posting/          ← relative to user CWD
├── session_state.json            # account & product config
├── config/
│   └── keywords.json             # keyword pool
├── tracking/
│   └── published.json            # published note records
├── reports/                      # performance reports
└── <YYYY-MM-DD>/                 # per-run date directory
    ├── topics/
    ├── selected_topics.json
    ├── style_fingerprint.json
    ├── drafts/
    └── replies/

Quick Start

/xiaohongshu-auto-posting

Standalone triggers (jump directly to the corresponding phase, skipping Phase 0–4):

• "track performance" / "see data" / "data collection" → Phase 6 performance tracking
• "settings" / "switch account" / "update keywords" → Phase 0 configuration wizard

Operational Rules

Three rules that apply to every browser-act session. Violating any one causes silent failures or long hangs. Full details in references/phase5-publish.md §0.

Rule 1 — Never pipe browser-act output directly

Redirect state output to a temp file, then grep the file. Piping may background the command when output contains multi-byte characters, making the result unavailable synchronously.

# WRONG
browser-act --session $SESSION state | grep "Publish"
# CORRECT
browser-act --session $SESSION state > /tmp/s.txt 2>&1 && grep "Publish" /tmp/s.txt

Rule 2 — Escape all non-ASCII content before passing to eval --stdin

Non-ASCII literals (Chinese, emoji) in JS source cause encoding errors in the subprocess pipeline. Escape to \uXXXX first:

def to_ascii_js(s):
    return ''.join(r'\u{:04x}'.format(ord(c)) if ord(c) > 127 else c for c in s)
js = to_ascii_js(open('tmp/script.js', encoding='utf-8').read())
subprocess.run([..., 'eval', '--stdin'], input=js.encode('ascii'), ...)

Rule 3 — Use click, not eval, for actions that trigger navigation

eval hangs (~30 s timeout) when the callback causes a page redirect. Always use browser-act --session $SESSION click <index> for the publish button, then verify the URL immediately after.

Phase 0 — Initialization

0.1 Tool Check

Run browser-act browser list:

• Command not found → run uv tool upgrade browser-act-cli || uv tool install browser-act-cli --python 3.12, then continue
• Returns normally → continue

0.2 Session State Load

Read workspaces/xhs-posting/session_state.json:

• File does not exist, or browser_id is empty → proceed to 0.3 First-time Configuration Wizard
• File exists and browser_id is non-empty → load and display current config, ask user to confirm or modify

session_state.json structure:

{
  "browser_id": "",
  "account": {
    "nickname": "",
    "profile": ""
  },
  "product": {
    "name": "",
    "tagline": "",
    "url": "",
    "install_cmd": ""
  },
  "keywords_file": "workspaces/xhs-posting/config/keywords.json",
  "posting": {
    "daily_limit": 1,
    "min_interval_hours": 6,
    "last_posted_at": null
  }
}

0.3 First-time Configuration Wizard (when file does not exist or browser_id is empty)

Step 1: Browser Selection

Run browser-act browser list, then:

When existing browsers are present: display all browsers (id / name / type / desc / proxy), ask the user which one to use for posting:

Existing browsers:

#1  id=xxx  name="xxx"  type=stealth  desc="xxx"  proxy=XX
#2  id=xxx  name="xxx"  type=chrome   desc="xxx"

Select a browser number to use for posting, or enter "new" to create one:
>

When no browsers exist, or user enters "new":

1. Load advanced guide: browser-act get-skills advanced
2. Ask the user:
   • Browser name (custom, e.g. "xhs-poster")
   • Proxy requirements (needed? region or custom URL?)
3. Create the browser following the advanced guide's Confirmation Gate protocol
4. Record the new browser ID, continue to Step 2

Step 2: Product & Keyword Configuration

Ask the user:

1. Product name (brand keyword, used in tags and body)
2. Product tagline (one-sentence description)
3. Product URL (website / GitHub / App Store, etc.)
4. Install / usage command (optional; for SaaS products, enter the core operation path)
5. Initial keywords (1–5, comma-separated, used for topic search)

Write to workspaces/xhs-posting/session_state.json and workspaces/xhs-posting/config/keywords.json, then continue to Phase 0.4.

0.4 Account Safety Check

• Read posting.last_posted_at from session_state.json
• If less than min_interval_hours have passed since the last post, notify the user and ask whether to force continue
• If today's post count ≥ daily_limit, notify the user

0.5 Browser Launch & Login Verification

Generate a unique session name for this execution:

$SESSION = "xhs-<YYYYMMDD>-<HHMM>"   # e.g., xhs-20260611-1430

Open the browser using the stored browser_id:

browser-act --session $SESSION browser open <browser_id> https://creator.xiaohongshu.com/new/home
browser-act --session $SESSION wait stable

Verify login: check that the current URL stays on creator.xiaohongshu.com (not redirected to the login page) and that a user avatar / nickname element is present in state.

• Not logged in → instruct the user to scan the QR code in --headed mode:

  browser-act --session $SESSION browser open <browser_id> https://creator.xiaohongshu.com/new/home --headed

  Wait for the user to confirm login, then continue.

Phase 1 — Pain-Point Topic Collection

Full execution parameters: references/phase1-topic-collection.md.

Overview:

1. Keyword rotation: read the next keyword from config/keywords.json (sequential mode)
2. XHS search: search the keyword, sort by hottest, collect metadata for the top 10–15 notes
3. Listing Pass: extract shallow fields directly from search results (title, likes, collects, comments, snippet)
4. Engagement scoring: score = likes + collects × 2 + comments × 1.5
5. Top 5 Deep: click into the top 5 notes and extract deep fields (key quote, pain description, topic tags)
6. Generate report: output workspaces/xhs-posting/<date>/topics/TOPICS_<kw-slug>.md
7. Display Top 5 and wait for user selection

Phase 2 — Top Case Collection (Style Reference)

Full execution parameters: references/phase2-case-collection.md.

Overview:

1. Use the same search results as Phase 1; no additional request needed
2. Collect full text from the top 3 high-engagement notes (title + body + tags)
3. Analyze writing style: opening type, paragraph rhythm, emoji density, topic tag strategy
4. Generate Style Fingerprint
5. Write style fingerprint to workspaces/xhs-posting/<date>/style_fingerprint.json

Phase 2 runs in parallel with Phase 1 (shared search results, different analysis angle).

Phase 3 — Topic Selection

After displaying the Phase 1 Top 5, ask the user:

Here are the Top 5 topics. Which ones do you want to write?
(enter numbers like "1 3"; "all" for all; "skip" to pass today)
>

On receiving the reply:

1. Write the selected topic list to workspaces/xhs-posting/<date>/selected_topics.json
2. For each selected topic, confirm the publish strategy (publish now / save as draft)

If the user enters "skip" → end this run without advancing last_index.