Xiaohongshu Skill

Use this skill as a file-backed operating system for studio and agency delivery. Keep all client/platform execution state in ~/.growth/vault/<profile>/xiaohongshu/ and use the bundled scripts to initialize workspaces, generate daily ops, and score account health.

Hard Dependency

This skill requires Python 3.10+, xiaohongshu-cli>=0.6.4, and its xhs binary for live Xiaohongshu work.

Install or upgrade with:

uv tool install xiaohongshu-cli
uv tool upgrade xiaohongshu-cli

Verify before any live Xiaohongshu operation:

python3 scripts/check_xhs_dependency.py
python3 scripts/check_xhs_dependency.py --research --auth

If --auth reports NEEDS_CONTEXT, run xhs login or xhs login --qrcode. Do not ask the user to paste raw cookies and do not print cookie values.

Operating Protocol

Role: act like a studio operator running repeatable Xiaohongshu delivery, not a one-off consultant.

Default execution style:

• Continue automatically through the next valid artifact instead of stopping after each file.
• Stop only when a required input, missing artifact, or unavailable capability blocks the next step.

Degradation protocol:

• If xhs is missing or outdated, stop live workflows with BLOCKED and give the exact install or upgrade command.
• If xhs is installed but unauthenticated, complete offline artifacts but mark live research or account actions as NEEDS_CONTEXT.
• If browser access is missing, prefer xhs live research. If xhs is unavailable, switch to exported URLs, screenshots, copied note metrics, or existing workspace artifacts.
• If metrics are missing, complete planning artifacts but mark health diagnosis as pending.
• If an artifact is stale or incomplete, repair it before generating downstream output.

Completion protocol:

• DONE: the requested workflow completed and the relevant artifact files are updated.
• DONE_WITH_CONCERNS: the workflow completed with explicit downgrade items or missing live evidence.
• BLOCKED: a required artifact, capability, or user input is missing.
• NEEDS_CONTEXT: the user must supply client specifics or metrics before the next step is valid.

Output Quality Contract

Apply this contract whenever producing Xiaohongshu copy, note drafts, rewrites, content calendars, daily ops, auto-reply candidates, or publish-ready bodies:

• Write in Xiaohongshu-native plain text, not dense essay prose.
• Keep Xiaohongshu titles at 20 characters or fewer, counting punctuation and symbols.
• Separate logical paragraphs, bullet groups, CTAs, and hashtag blocks with at most one blank line.
• Never emit two or more consecutive blank lines. Xiaohongshu only supports one blank spacer line.
• Keep visual groups to 2-3 consecutive non-empty lines. Use 4+ only when the user explicitly asks for a compact list.
• Use emoji or visible symbols as section openers and bullets, such as 🌿 1｜..., ✅ ..., 📌 ..., • ..., or —— dividers.
• Do not output bare Markdown bullets for publish-ready body copy. Convert - item into a native symbol line like • item.
• Do not leak visible Markdown syntax into publish-ready Xiaohongshu bodies: no **bold**, ## headings, Markdown tables, fenced code blocks, or link markup.

This contract does not apply to internal status reports, command output, JSON, or engineering/debug summaries.

Mode Router

Choose exactly one mode before doing the work:

1. launch-new-client Use when no client workspace exists yet, or only the client name / industry is known.
2. run-daily-ops Use when a client workspace already exists and the task is to continue planning or execution.
3. diagnose-underperforming-account Use when the user asks why traffic is weak, why notes are stuck, or whether the account is ready to monetize.

Capability Check

• For live Xiaohongshu research, run python3 scripts/check_xhs_dependency.py --research --auth first.
• If authenticated xhs is available, use it as the default live research path for search results, note reads, comments, account pages, own notes, and publishing preflight.
• If authenticated xhs is unavailable but browser access is available, inspect live Xiaohongshu search results, note pages, and account pages directly.
• If neither authenticated xhs nor browser access is available, require one of these before claiming live analysis:
  • exported note/account URLs
  • screenshots of note performance or account pages
  • copied note metrics
  • an existing metrics.csv
• Do not invent live research findings. When inputs are partial, complete the files you can and stop with the next missing artifact or input called out explicitly.

Vault Content Model

The vault is split into a reusable platform library and profile-specific distilled workspaces. Do not mix user personas, benchmarks, or evidence from different apps into one generic folder.

~/.growth/vault/
├── _library/
│   ├── xiaohongshu/
│   │   ├── raw/
│   │   ├── evidence/
│   │   ├── personas/
│   │   ├── content-patterns/
│   │   ├── platform-rules/
│   │   └── benchmarks/
│   └── _shared/
│       ├── offers/
│       ├── brand-assets/
│       └── cross-platform-insights/
└── <profile>/
    └── xiaohongshu/

Use ~/.growth/vault/_library/xiaohongshu/ for reusable Xiaohongshu corpus: generic platform research, reusable benchmark patterns, app-specific persona archetypes, platform rules, and cross-client evidence. A future Facebook skill must use ~/.growth/vault/_library/facebook/ for Facebook-specific personas and benchmarks instead of writing them into the Xiaohongshu library.

Use ~/.growth/vault/_library/_shared/ only for genuinely cross-platform inputs such as offers, brand assets, and cross-platform observations. Do not put platform-specific user personas there. If an audience is shared at the brand level, keep the abstract audience in _shared/ and write the app-specific projection under each platform library.

Use ~/.growth/vault/<profile>/xiaohongshu/ for distilled client delivery state: brief, competitor analysis, account strategy, content calendar, daily ops, metrics, playbook, action log, profile-specific evidence, and lessons. This folder is the source of truth for execution, cron, diagnosis, publishing preflight, and client delivery.

For live research:

• If the research was collected for one profile, write evidence to ~/.growth/vault/<profile>/xiaohongshu/xhs-evidence/ and distill it into that profile's 02-competitor-analysis.md.
• If the research is a reusable market corpus, write raw or semi-raw evidence to ~/.growth/vault/_library/xiaohongshu/evidence/<date>-<topic>/, then distill only the relevant takeaways into profile workspaces.
• Never let _library/ directly drive execution. Execution must go through a profile/platform artifact first.

XHS Action Boundary

Default automation is read-only: xhs search, xhs read, xhs comments, xhs user, xhs user-posts, xhs my-notes, xhs topics, and xhs hot.

Write operations require an explicit user request for the specific action: xhs post, xhs delete, xhs like, xhs favorite, xhs comment, xhs reply, xhs follow, or xhs unfollow.

Before any write operation:

1. Run xhs whoami --json to confirm the current account.
2. Execute only the requested single action; do not batch or infer adjacent actions.
3. Append the command result or structured error to ~/.growth/vault/<profile>/xiaohongshu/xhs-action-log.md.
4. For approved note publishing, do not manually assemble an xhs post command from a Markdown draft. Use scripts/publish_note.py as the publishing boundary. Draft files may remain Markdown for review; the script extracts title/body/hashtags, converts the body to Xiaohongshu-native plain text, rejects leaked Markdown syntax, posts only when --post is passed, logs the full JSON response, verifies via xhs my-notes --json, and appends initial metrics.csv.
5. Always dry-run the exact draft before posting:

python3 scripts/publish_note.py \
  --client-dir ~/.growth/vault/<profile>/xiaohongshu/ \
  --draft ~/.growth/vault/<profile>/xiaohongshu/drafts/<draft>.md \
  --images ~/.growth/vault/<profile>/xiaohongshu/assets/<cover>.png \
  --content-type "<bucket>" \
  --keyword "<keyword>" \
  --body-output /tmp/xhs-prepared-body.txt

The dry run must report "markdown_leaks": []. If it does not, stop and repair the script or draft before posting. Also inspect the generated --body-output for vertical rhythm: logical paragraphs must be separated by blank lines, and the body should not contain more than 3 consecutive non-empty lines before a blank line unless the user explicitly approves a dense list. 6. Only after the user explicitly authorizes the specific write action, run the same command with --post. Do not bypass the script with direct xhs post unless the script itself is broken and the user accepts the risk. 7. Xiaohongshu does not render Markdown. The exact body sent to xhs post must not contain visible Markdown syntax such as **bold**, ## headings, fenced code blocks, checklist markers, Markdown tables, or link markup. Use short lines, blank lines, emoji section markers, Chinese punctuation, and —— dividers instead. 8. If the command returns verification_required, ip_blocked, not_authenticated, or another upstream error, stop and report DONE_WITH_CONCERNS or NEEDS_CONTEXT.

Side Workflows

• check-xhs-dependency Use before any live research or action. Run:

python3 scripts/check_xhs_dependency.py --research --auth

• collect-live-research Use after 02-competitor-analysis.md exists and needs real Xiaohongshu evidence. Run:

python3 scripts/collect_xhs_research.py \
  --brief ~/.growth/vault/<profile>/xiaohongshu/01-client-brief.md \
  --output ~/.growth/vault/<profile>/xiaohongshu/02-competitor-analysis.md

The live collector now defaults to safe merge mode: it appends/refreshes a ## Live Research Evidence section instead of overwriting existing manual analysis. Use --overwrite only when replacing the whole analysis is intentional; a .bak is created first. Low-sample, partially failed search, or incomplete account enrichment is marked Research Status: PARTIAL and returns exit code 1 unless --allow-partial is explicitly set. Treat PARTIAL as usable evidence, not a completed research gate. Use --account-limit to bound account page/user-post sampling, --retries --delay-min --delay-max for transient retry pacing, and --command-delay-min --command-delay-max for conservative global pacing between live xhs commands.

• check-client-workspace Use when the user asks what is missing, what is stale, or where a client is currently blocked. Run:

python3 scripts/diagnose_workspace.py --client-dir ~/.growth/vault/<profile>/xiaohongshu/

• review-studio-queue Use when the user wants a multi-client status sweep across the whole studio workspace. Run:

python3 scripts/diagnose_workspace.py --all

• learn-client-edits Use when the user revised titles, posting cadence, or diagnosis recommendations and wants the system to adapt. Read learn-client-edits.md, then run:

python3 scripts/learn_client_edits.py \
  --client-dir ~/.growth/vault/<profile>/xiaohongshu/ \
  --draft <path-to-previous-artifact> \
  --final <path-to-client-edited-artifact>

Workspace Contract

Distilled client state lives under one client/platform folder in the system user's home directory, never inside the skill package or repository:

~/.growth/vault/<profile>/xiaohongshu/
├── 01-client-brief.md
├── 02-competitor-analysis.md
├── 03-account-strategy.md
├── 04-content-calendar.md
├── 05-daily-ops.md
├── 06-health-report.md
├── metrics.csv
├── playbook.md
├── xhs-action-log.md
├── xhs-evidence/
└── lessons/

Initialize a new workspace with:

python3 scripts/init_client_workspace.py --client "<client-name>" --profile "<profile>" --industry "<industry>" --root .

Treat an artifact as incomplete if it still contains TODO, {{...}}, or empty required sections. Do not skip ahead unless all lower-numbered artifacts are complete, except in diagnosis mode.

Standard Workflow

launch-new-client

1. Collect intake using intake-and-positioning.md.
2. Run init_client_workspace.py if ~/.growth/vault/<profile>/xiaohongshu/ does not exist.
3. Fill 01-client-brief.md before doing research.
4. Prepare 02-competitor-analysis.md with:

python3 scripts/prepare_competitor_analysis.py \
  --brief ~/.growth/vault/<profile>/xiaohongshu/01-client-brief.md \
  --output ~/.growth/vault/<profile>/xiaohongshu/02-competitor-analysis.md

Then fill it with xhs live research, browser findings, or fallback artifacts using research-rubric.md. If playbook.md exists, treat its preferences as research bias, not just downstream copy bias. 5. Immediately collect live Xiaohongshu evidence if xhs is authenticated:

python3 scripts/check_xhs_dependency.py --research --auth
python3 scripts/collect_xhs_research.py \
  --brief ~/.growth/vault/<profile>/xiaohongshu/01-client-brief.md \
  --output ~/.growth/vault/<profile>/xiaohongshu/02-competitor-analysis.md

If authentication is missing, keep 02-competitor-analysis.md as a research brief and mark the live evidence gap explicitly. 6. Build 03-account-strategy.md with:

python3 scripts/generate_account_strategy.py \
  --brief ~/.growth/vault/<profile>/xiaohongshu/01-client-brief.md \
  --analysis ~/.growth/vault/<profile>/xiaohongshu/02-competitor-analysis.md \
  --output ~/.growth/vault/<profile>/xiaohongshu/03-account-strategy.md

Use intake-and-positioning.md to review the generated persona and niche choices before accepting them. If playbook.md exists, the strategy must carry those constraints into naming, topic architecture, and content boundaries. 7. Build 04-content-calendar.md with:

python3 scripts/generate_content_calendar.py \
  --brief ~/.growth/vault/<profile>/xiaohongshu/01-client-brief.md \
  --strategy ~/.growth/vault/<profile>/xiaohongshu/03-account-strategy.md \
  --analysis ~/.growth/vault/<profile>/xiaohongshu/02-competitor-analysis.md \
  --output ~/.growth/vault/<profile>/xiaohongshu/04-content-calendar.md

Use content-and-compliance.md and copywriting-style.md to improve the generated calendar before finalizing it. The generated calendar must incorporate not only 03-account-strategy.md, but also the keyword map, repeatable patterns, and research summary from 02-competitor-analysis.md. If playbook.md has rules, the script must apply them to title shape, hook style, emoji usage, and posting volume. 8. Regenerate 05-daily-ops.md with:

python3 scripts/build_daily_ops.py \
  --brief ~/.growth/vault/<profile>/xiaohongshu/01-client-brief.md \
  --calendar ~/.growth/vault/<profile>/xiaohongshu/04-content-calendar.md \
  --output ~/.growth/vault/<profile>/xiaohongshu/05-daily-ops.md

9. Leave 06-health-report.md as a pending template until metrics exist.
10. Leave playbook.md untouched until there is at least one real client edit to learn from.

run-daily-ops

1. Run diagnose_workspace.py first and use its first incomplete artifact as the starting point.
2. Continue from that file instead of rewriting completed work.
3. If 02-competitor-analysis.md is incomplete or stale, rerun prepare_competitor_analysis.py if needed, then run check_xhs_dependency.py --research --auth and collect_xhs_research.py when live evidence is available.
4. If 02-competitor-analysis.md changes materially, rerun generate_account_strategy.py. If 03-account-strategy.md changes, rerun generate_content_calendar.py. If 04-content-calendar.md changes, rerun build_daily_ops.py so 05-daily-ops.md stays in sync.
5. Append new note performance data to metrics.csv whenever the user provides it.
6. If at least 5 rows of metrics exist, refresh 06-health-report.md with score_health.py.

diagnose-underperforming-account

1. Require recent note metrics before giving prescriptive advice.
2. If the user gives free-form metrics, normalize them into metrics.csv using the header from metrics-template.csv.
3. If the user explicitly authorizes using the logged-in account, run check_xhs_dependency.py --auth and xhs my-notes --json to help fill missing own-note identifiers or visible live data; otherwise keep metrics.csv as the source of truth.
4. Run:

python3 scripts/score_health.py \
  --metrics ~/.growth/vault/<profile>/xiaohongshu/metrics.csv \
  --output ~/.growth/vault/<profile>/xiaohongshu/06-health-report.md

5. Use diagnosis-rubric.md and content-and-compliance.md to explain the bottleneck and propose the next actions. If playbook.md exists, the health report must reflect the client's learned preferences.
6. Do not recommend monetization until the exit criteria in the health report pass.
7. If the user later rewrites the diagnosis recommendations, capture that learning via learn_client_edits.py so future health reports match the client's decision style.

References

• intake-and-positioning.md: intake fields, persona selection, niche rules
• research-rubric.md: competitor capture schema, keyword harvesting, benchmark criteria
• content-and-compliance.md: title patterns, cover guidance, cadence, compliance checks
• copywriting-style.md: platform-native voice, emoji rules, sentence structure, power words, body templates, hashtag conventions
• diagnosis-rubric.md: traffic tiers, engagement thresholds, escalation rules
• learn-client-edits.md: how to capture client edits and update playbook.md
• xhs-native-publish-boundary.md: why drafts stay Markdown while the publish script converts to Xiaohongshu-native plain text and blocks Markdown leakage
• paragraph-rhythm-and-repost.md: blank-line rhythm gate and delete/repost recovery when a live note reads as a dense block

Scripts

• scripts/init_client_workspace.py: create a standard client folder from templates
• scripts/check_xhs_dependency.py: verify xiaohongshu-cli>=0.6.4, read-only research commands via --research, write commands for publishing preflight, and optional authentication
• scripts/xhs_cli_utils.py: invoke xhs --json and validate the structured ok/schema_version/data/error envelope
• scripts/publish_note.py: publish an approved Markdown draft safely; converts draft Markdown to Xiaohongshu-native plain text, dry-runs by default, requires --post for the actual write action, logs the result, verifies via my-notes, and appends initial metrics. Always inspect --body-output for paragraph rhythm before posting; see references/paragraph-rhythm-and-repost.md.
• scripts/collect_xhs_research.py: collect live Xiaohongshu search/read/comment evidence into 02-competitor-analysis.md
• scripts/build_daily_ops.py: turn a brief plus content calendar into D1-D7 or D1-D10 checklists
• scripts/prepare_competitor_analysis.py: generate a playbook-aware research brief for 02-competitor-analysis.md
• scripts/generate_account_strategy.py: generate 03-account-strategy.md from the client brief, competitor analysis, and playbook rules
• scripts/generate_content_calendar.py: generate 04-content-calendar.md from the client brief, account strategy, and playbook rules
• scripts/score_health.py: score recent note metrics and write a health summary
• scripts/diagnose_workspace.py: inspect required artifacts, stale health reports, and client readiness
• scripts/learn_client_edits.py: capture recurring client edits and rebuild a client-specific playbook

Operating Rules

• Prefer file-backed continuity over ad hoc chat summaries.
• Keep user/customer workspace data out of the skill package, repo, and dist/openclaw; distilled workspaces belong under ~/.growth/vault/<profile>/xiaohongshu/, while reusable platform corpus belongs under ~/.growth/vault/_library/xiaohongshu/. When syncing or packaging this skill, preserve/migrate user data there and exclude repo-local workspace folders.
• Keep app-specific user personas app-specific. Xiaohongshu personas belong in _library/xiaohongshu/personas/; future Facebook personas belong in _library/facebook/personas/; _library/_shared/ is only for true cross-platform inputs.
• Prefer concrete artifacts over generic strategy prose.
• Prefer xhs live evidence over browser/manual evidence when authenticated.
• Prefer capability-aware fallbacks over pretending unavailable tools exist.
• Never auto-publish, auto-like, auto-comment, auto-follow, or delete content from a calendar or plan; write operations require an explicit user instruction and an action log entry. Dashboard-managed auto-reply defaults to send-reply mode for new or missing profile settings, while profiles explicitly saved as draft-only keep that setting until changed.
• Keep recommendations consistent with the studio workflow in this skill, not a solo creator workflow.