xiaohongshu-ops

Openclaw Xiaohongshu Operations Skill (General Edition)

Goal: build a reusable "Xiaohongshu operations" process so that any account type can reuse the same action framework.

Scope of application (by default, the general process)

Account positioning and content direction
Topic generation and controversy point mining
Competitor/similar account benchmarking
Pre-publish rehearsal and content delivery for Xiaohongshu
Quick post-publish review (interaction structure, comment replies, trend tracking)
Viral Copy pipeline (input a URL, closely learn the cover/images, title, and body, and generate a near-identical publishable structure note)

Attach the industry details of each account type as a "case module" to the general process.

Common terms

选题: a content entry point that can be published, discussed, and shared
引流钩子: the title/opening sentence used to trigger lingering and clicks
结构化输出: the five-tuple of title, body, interactive question, topic, and tags
快照: a key evidence snapshot used to verify page status
回放: retry or reroute execution after a process failure

0) Startup and environment verification (followed by all tasks)

Before execution, first follow references/xhs-runtime-rules.md in "Run Rules"; prioritize the reusable failure order.

Fixed use of the built-in browser profile:openclaw, if a channel anomaly occurs, switch back first and then retry.
If the browser (openclaw-manager) capability is disabled/unavailable: first perform one lightweight retry (such as status/profiles); if still unavailable, enter troubleshooting guidance, clearly inform the user that "the current browser tool is not enabled," and guide the user to enable it according to the documentation before continuing (reference:https://docs.openclaw.ai/tools/browser）。
Prioritize evaluate , with a small amount of snapshotat key nodes, and retry a single-step action at most once.
After a failure, retain the results already obtained, switch to a robust path, and report.

1) Default skill behavior (followed by all tasks)

First read persona.mdunder this skill directory (Xiaohongshu platform-specific persona/tone/publishing and reply style). All external copy (posts/comment replies/private message scripts) must follow it.
Before starting a new task, first read knowledge-base/README.md , this overview entry, then follow references/xhs-knowledge-base.md rules to search for the most recent similar records; do not repeatedly explore patterns that can be reused.
Prioritize outputting executable SOPs rather than one-off content drafts
Prioritize language that "can converse" rather than "can write reports": short sentences, colloquial, clear stance, can guide comments
All outputs by default retain "follow-up points" for continued extension in the comment section

2) Account positioning (reusable)

For each account, first confirm 4 variables:

Target users: age/scenario/pain points (such as "fragmented time after work" "star-chasing discussion groups")
Content value proposition: what each post gives users (viewpoint, emotional value, practical advice)
Differentiation angle: what similar accounts do not do, and what you do
Style guidelines: tone, length, conflict boundaries (avoid over-aggression)

Output:

Persona keywords (3-5)
Content pillars (3)
Catchphrases/fixed sentence patterns (2-3)
List of bottom lines that must not be crossed (red lines) (spoilers, personal attacks, false promises)

2.5) Account analysis (new)

Follow references/xhs-account-analysis.md to execute.

By default, sample the most recent 9-15 pieces of content for a lightweight checkup
Judge from 5 dimensions: positioning, content structure, interaction conversion, distinctiveness, and sustainability
The output must include "biggest advantage, biggest shortcoming, next-step action"

3) General topic selection and benchmarking process

A. Platform-side signal capture (can be parallelized)

First capture high-interaction content on the same topic on Xiaohongshu (likes/saves/comments above the recent average)
Record reusable fields:title, hook, angle, 结构标签, 评论信号, 互动CTA, 标签组
Aggregate the top 10-20 items into a candidate pool

A.1 Homepage recommendation feed analysis (new)

Follow references/xhs-home-feed-analysis.md to execute.

First look at "why it was recommended to you" in the homepage recommendation feed
Then distill reusable propagation hooks, content structures, and topic directions
Results prioritize serving account positioning, topic inspiration, and subsequent content judgment

B. Demand-side supplementary signals (industry/scenario)

Go by theme to mainstream platforms/social media to capture "differing opinions in the comment section"
Extract one group each of supporting/opposing/neutral viewpoints
Output debatable points suitable for posting (controversial but controllable)

C. Form a topic list (at least 3 per round)

Each topic includes:

Topic title (optional within 20 characters)
Viewpoint label (support/oppose/neutral)
Expected interaction hook
Evidence source (which set of high-interaction data)
Risk note (whether it easily crosses the line)

3.2) Topic inspiration (new)

Follow references/xhs-topic-ideation.md to execute.

Merge platform signals, demand signals, and account positioning into publishable topics
By default, output 3-5 items, each with an interaction hook and a three-part structure
The output can be used directly as the前置 input for content generation or Viral Copy

3.5) Search and browse (new operation type)

Follow references/xhs-runtime-rules.md 's search and comment entry chapter to execute.

Only entering posts from the search results page is allowed;
In notification/reply scenarios, perform alignment verification first.
For the fallback strategy after consecutive failures, see the referenced file.

3.6) Viral Copy (URL → new note)

Follow references/xhs-viral-copy-flow.md to execute.

Input: target viral note URL(s) (multiple allowed).
Output: 1 set of publishable materials (cover/image plan + title + body + topics).
Replication principle: closely match the theme and structure (title sentence pattern, cover information hierarchy, body rhythm, interaction mechanism), while avoiding verbatim copying and asset infringement.

4) General content template (Xiaohongshu)

Produce at least 2 alternatives each time:

Title (controversial/stance/rhetorical question, preferably ≤20 characters)
Opening hook (1-2 sentences)
Body (3 paragraphs: viewpoint → evidence → opposing side)
Interactive question (1 sentence)
Topics (5-8)
Risk annotation (whether spoilers / provocation boundaries / copyright risk)

5) General publishing pipeline (do not publish)

For the detailed publishing execution path, directly follow references/xhs-publish-flows.md to execute, avoiding duplicate maintenance.

Core points that must be satisfied before publishing:

First log in to the creator backend for the account, and confirm the page is in openclaw profile and operable.
Clarify the publishing type (video / image-text / long-form article), with three elements: cover, title, and body.
Stop when the "Publish" button is visible, and by default do not click publish directly.
If screenshot confirmation is involved, prefer sending it as an attachment to Feishu, and publish only after the user confirms.

6) Comments and replies (lightweight)

Comment checking and replies uniformly follow references/xhs-comment-ops.md, and combine with examples/reply-examples.md for the copywriting style.

By default, prioritize the notifications page: locate first, then input, then send.
By default, one-send-per-turn (do not send consecutively unless explicitly required).
For risk control items such as length, implicit promises, and risk-control stop-loss points, refer to the cited files.

6.5) Knowledge base accumulation (new)

Follow references/xhs-knowledge-base.md to execute.

The overview entry is fixed as knowledge-base/README.md
Detailed records are written by type into knowledge-base/accounts/、knowledge-base/topics/、knowledge-base/patterns/、knowledge-base/actions/、knowledge-base/reviews/
Analysis is preferentially accumulated into pattern / topic / review
Execution actions are preferentially accumulated into action
At the end of the task, at least leave searchable conclusions, evidence, risks, and next steps

7) Failures and fixes (must be followed)

If automation fails, first retry once (same strategy)
If it still fails, switch routes: change to a "more stable synonymous path"
Do not perform ineffective repeated actions; keep the current progress reusable, and report once the single action the user needs to do manually
If the knowledge base is temporarily unwritable, first return a structured summary, add the record after the task ends, and do not block the main process

8) General extraction example (Evaluate)

For a general field extraction script example, see references/xhs-eval-patterns.md。

9) Specific case: Watch Drama With You (retained as a special case)

How to use

This skill's main file retains the general framework; vertical industry experience is placed in the examples/ directory, selected by content type:

First run through it according to the "General Process"
Then load the corresponding case file to fill in industry-specific actions

Currently available cases:

examples/drama-watch/case.md(Watch Drama With You account)

Each content type is organized by directory, and files may be named:

examples/<vertical>/<vertical>.md(recommended)
or examples/<vertical>/README.md
examples/lifestyle/(to be supplemented)
examples/cosmetics/(to be supplemented)
examples/fitness/(to be supplemented)

Practical experience (continuously valid)

Unified rule: all browser operations always use the built-in browser profile=openclaw(unless the user explicitly requests use of the Chrome extension Relay).
Text with images is a stable writing entry point; typed text directly becomes the cover copy
For publishing topics, prefer UI topic selection; it is not recommended to paste large amounts of plain text #话题
evaluate When batch rewriting rich text, try to change the layout as little as possible to avoid losing topic entity
Keep a snapshot before key steps, which can be used for review and problem localization
发布 The button being visible ≠ successful publishing; it must be explicitly marked as "stop at the publish page"
If a new type of comment pacing issue arises, prioritize reducing the reply density per hour rather than increasing frequency

Mature operational path (optional)

Title pool: keep 10 reusable templates each for "taking sides/rhetorical questions/conflict"
Topic pool: build a list of commonly used keywords and synonymous substitutions based on the account's tone
Reuse mechanism: after each review, sync reusable expressions into the case file