Xiaohongshu

Xiaohongshu

Search notes, read note details, view comments, browse user profiles, and get the home feed from Xiaohongshu (小红书).

Skill Directory

<SKILL_DIR> is the directory containing this SKILL.md file. Determine it ONCE at the start and reuse it.

Setup (run FIRST — every time, before any operation)

You MUST complete this setup before running any script. Do NOT skip.

sig status xiaohongshu 2>&1

Check the JSON output fields configured and valid:

• configured: false → run Provider Setup below.
• valid: false (but configured: true) → run sig login xiaohongshu --mode visible, complete the captcha, then re-check.
• valid: true → run Vendor Setup (one-time per machine), then execute the user's request.

Provider Setup

1. Read <SKILL_DIR>/references/provider-config.yaml
2. Append the provider block to ~/.sig/config.yaml under providers:
3. Run sig login xiaohongshu --mode visible — a browser opens; scan the QR code with the Xiaohongshu app and complete the slider/image captcha if prompted
4. Verify: sig status xiaohongshu should show valid: true

Why the validateRule is so picky — three response shapes

/api/sns/web/unread_count is a permissive endpoint that hides a "partially expired cookie" state behind what looks like a successful response. The provider's validateRule is built specifically to reject it:

If sig login exits without prompting you, but later API calls fail with empty data — the partially-expired-cookie state was bypassed by a stale rule. Re-pull references/provider-config.yaml and refresh the rule in ~/.sig/config.yaml, then re-login.

What this rule does NOT catch. unread_count is more permissive than search/feed APIs. A cookie that passes the rule (state #1) may still fail on search if the risk-control / session-token portion has aged out. If sig status shows valid:true but search keeps returning empty data, you're in this gray zone — the only fix is sig logout xiaohongshu && sig login xiaohongshu --mode visible. See Diagnosis below for how to confirm.

Vendor Setup (one-time per machine)

Unlike most skills, xiaohongshu signs every request via JS files at <SKILL_DIR>/vendor/static/, evaluated through PyExecJS. The skill ships those source files but not their npm/pip deps — install them once:

node --version                          # must be >= 18; install from https://nodejs.org if missing
cd <SKILL_DIR>/vendor && npm install    # installs crypto-js, jsdom
pip install --user -r <SKILL_DIR>/requirements.txt   # PyExecJS, requests, loguru, retry

If pip complains about externally-managed Python (PEP 668 on macOS/Debian), use a venv or pipx.

Smoke-test signing (no network, no cookie needed):

cd <SKILL_DIR>/vendor && python3 -c "
import sys; sys.path.insert(0, '.')
from xhs_utils.xhs_util import generate_x_rap_param, generate_xs_xs_common
xs, xt, xsc = generate_xs_xs_common('a1=test', '/api/sns/web/v1/feed', '', 'POST')
rap = generate_x_rap_param('/api/sns/web/v1/user_posted', '')
print('OK' if xs and rap else 'FAIL')
"

Expect OK. If it fails, see Error Handling.

Running Scripts

All scripts output JSON to stdout. Read operations need a cookie (the public site does not allow anonymous access to most endpoints).

Recommended — sig run injects the cookie as SIG_XIAOHONGSHU_COOKIE:

sig run xiaohongshu -- bash -c 'python3 <SKILL_DIR>/scripts/xiaohongshu_search_note.py --keyword "AI" --page 1'

Alternative — pass --cookie explicitly:

COOKIE=$(sig get xiaohongshu --no-redaction --format value)
python3 <SKILL_DIR>/scripts/xiaohongshu_search_note.py --keyword "AI" --cookie "$COOKIE"

Scripts Reference

All scripts are in this skill's scripts/ directory. Output is JSON to stdout.

Script Arguments

xiaohongshu_search_note.py

• --keyword (required) — search keyword
• --page — page number, default 1
• --sort — 0 general (default), 1 newest, 2 most-liked, 3 most-commented, 4 most-collected
• --note-type — 0 any (default), 1 video, 2 image
• --cookie — override env var

xiaohongshu_get_note_info.py

• --url (required) — full note URL like https://www.xiaohongshu.com/explore/<id>?xsec_token=...&xsec_source=pc_search
• --cookie

xiaohongshu_get_note_comments.py

• --note-id (required)
• --xsec-token (required) — comes from a search/feed result
• --cursor — pagination, default empty
• --cookie

xiaohongshu_get_user_info.py

• --user-id (required)
• --cookie

xiaohongshu_get_user_notes.py

• --user-id (required)
• --cursor — pagination, default empty
• --xsec-token, --xsec-source — pass through if available
• --cookie

xiaohongshu_get_homefeed.py

• --category — channel id, default homefeed_recommend
• --cursor-score — pagination
• --refresh-type, --note-index — feed iteration state
• --cookie

Key Concepts

• xsec_token / xsec_source: per-note security tokens issued by the search/feed endpoints. To call get_note_info or get_note_comments, you must first call search_note to obtain these tokens (they are embedded in the note URL returned).
• Why Node.js? XHS signs every request with x-s, x-t, x-s-common, x-rap-param, and x-xray-traceid derived from JSVMP-obfuscated JavaScript. The skill ships those JS files (<SKILL_DIR>/vendor/static/) and runs them via PyExecJS, which shells out to node. Pure-Python signing libraries (e.g. xhshow) currently miss x-rap-param and fail on data APIs with HTTP 406.

Error Handling

Workflow Examples

Find a topic and read the top result with comments

# 1. Search — output is JSON with a list of notes
sig run xiaohongshu -- bash -c \
  'python3 <SKILL_DIR>/scripts/xiaohongshu_search_note.py --keyword "城市探索" --sort 2' \
  > /tmp/xhs_search.json

# 2. Extract note_id + xsec_token from the first result.
#    Path: items[0].id  and  items[0].xsec_token
NOTE_ID=$(jq -r '.items[0].id' /tmp/xhs_search.json)
XSEC=$(jq -r '.items[0].xsec_token' /tmp/xhs_search.json)
URL="https://www.xiaohongshu.com/explore/${NOTE_ID}?xsec_token=${XSEC}&xsec_source=pc_search"

# 3. Get full detail
sig run xiaohongshu -- bash -c \
  "python3 <SKILL_DIR>/scripts/xiaohongshu_get_note_info.py --url '$URL'"

# 4. Get comments
sig run xiaohongshu -- bash -c \
  "python3 <SKILL_DIR>/scripts/xiaohongshu_get_note_comments.py --note-id '$NOTE_ID' --xsec-token '$XSEC'"

Diagnosis: when search fails, do NOT jump to conclusions

sig status xiaohongshu showing valid: true is necessary but not sufficient for search to work. The validateUrl is permissive on purpose (so it can run without XHS request signing), which means a cookie that has aged out enough to fail on search may still pass validation. Before re-logging in or "fixing" anything, triangulate with three probes:

# Probe 1 — what sig itself thinks
sig status xiaohongshu

# Probe 2 — what the validateUrl says (permissive endpoint)
sig run xiaohongshu -- bash -c \
  'curl -sL -H "Cookie: $SIG_XIAOHONGSHU_COOKIE" -H "User-Agent: Mozilla/5.0" \
   https://edith.xiaohongshu.com/api/sns/web/unread_count'

# Probe 3 — what a stricter authenticated endpoint says
sig run xiaohongshu -- bash -c \
  'curl -sL -H "Cookie: $SIG_XIAOHONGSHU_COOKIE" -H "User-Agent: Mozilla/5.0" \
   https://edith.xiaohongshu.com/api/sns/web/v2/user/me'

Read the three answers together:

The mistake the previous version of this skill encouraged: see KeyError: 'msg' → assume captcha → tell the user to re-login + solve captcha. Most of the time captcha never appears, the real cause is cookie aging, and re-login does fix it — but for the wrong reason. Document the right reason so the next debugger doesn't get misled.

Two-tier verification. Run tier 1 unconditionally; tier 2 only after tier 1 passes.

Tier 1: prerequisite checks (read-only, safe)

# 1. Provider authenticated and validated against the strict rule
sig status xiaohongshu
# Expect: configured: true, valid: true

# 2. Probe the validateUrl directly — must return non-empty data object
sig run xiaohongshu -- bash -c 'curl -sL --max-time 10 \
  -H "Cookie: $SIG_XIAOHONGSHU_COOKIE" \
  -H "User-Agent: Mozilla/5.0" \
  https://edith.xiaohongshu.com/api/sns/web/unread_count'
# Expect: {"code":0,"success":true,"msg":"成功","data":{"unread_count":N,...}}
# If `data` is `{}` → captcha not solved. Re-login with --mode visible and complete the captcha.
# If code is -101 → session gone. Re-login.

# 3. Vendor signing JS deps installed
test -d <SKILL_DIR>/vendor/node_modules && echo "OK: node_modules present" || echo "FAIL: cd vendor && npm install"

# 4. Smoke-test signature generation (no network, no cookie)
cd <SKILL_DIR>/vendor && python3 -c "
import sys; sys.path.insert(0, '.')
from xhs_utils.xhs_util import generate_x_rap_param, generate_xs_xs_common
xs, xt, xsc = generate_xs_xs_common('a1=test', '/api/sns/web/v1/feed', '', 'POST')
rap = generate_x_rap_param('/api/sns/web/v1/user_posted', '')
print('OK' if xs and rap else 'FAIL')
"
# Expect: OK

Tier 2: live execution test

Run a tiny search and assert that items is a non-empty list. This is the canary for "everything actually works":

sig run xiaohongshu -- bash -c \
  'python3 <SKILL_DIR>/scripts/xiaohongshu_search_note.py --keyword "AI"' \
  | python3 -c "
import json, sys
d = json.load(sys.stdin)
items = d.get('items', [])
assert isinstance(items, list) and len(items) > 0, f'expected items list, got: {d}'
print(f'OK: {len(items)} items')
"
# Expect: OK: N items
# If you see {'error': 'API_ERROR', 'message': \"code=0, msg=\\\"'msg'\\\"\"} → captcha-not-solved
# soft-reject. The cookie is half-cooked even though sig status shows valid:true.
# Refresh references/provider-config.yaml into ~/.sig/config.yaml, then re-login with captcha.

This skill bundles a minimal slice of cv-cat/Spider_XHS (MIT) at <SKILL_DIR>/vendor/. To bump the vendored version:

<SKILL_DIR>/scripts/sync-vendor.sh           # latest master
<SKILL_DIR>/scripts/sync-vendor.sh <ref>     # specific commit/tag

License and upstream metadata are in <SKILL_DIR>/vendor/LICENSE and <SKILL_DIR>/vendor/UPSTREAM.md.

Source: sigcli/sigcli — distributed by TomeVault.