小红书

Security checks across malware telemetry and agentic risk

Overview

The skill is coherent and safety-gated, but it can operate a Xiaohongshu account and publish public posts through a user-installed local component, so users should review setup carefully.

Before installing, independently review and pin the third-party xiaohongshu-mcp component, use a dedicated Xiaohongshu account, keep the MCP server on localhost, and verify the full untruncated client code locally. Treat every publish action as public and potentially irreversible.

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal

Risk analysis

Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.

#
ASI03: Identity and Privilege Abuse
Medium
What this means

If configured, the tool can make public posts on Xiaohongshu using the account logged into the local component.

Why it was flagged

The setup explicitly states that the third-party component can act as the logged-in dedicated account to publish public content.

Skill content
我理解并接受:该组件运行期间可代表上述专用账号发布公开内容,误操作不可撤回。
Recommendation

Use only a dedicated account, avoid main or enterprise accounts, and revoke the login device after use as SETUP.md recommends.

#
ASI02: Tool Misuse and Exploitation
Medium
What this means

A mistaken publish command could post unintended public content, though the artifacts include confirmation requirements and client-side safeguards.

Why it was flagged

The skill exposes a public publishing action, but the documented workflow requires preview and explicit user authorization before calling publish.

Skill content
代理必须向用户打印结构化预览并等待用户在对话中回复明确授权词
Recommendation

Review every preview carefully and only reply with the required confirmation when you genuinely want the post to go live.

#
ASI04: Agentic Supply Chain Vulnerabilities
Low
What this means

The safety of the actual platform automation depends partly on the third-party MCP component you install and run locally.

Why it was flagged

The skill depends on a separate third-party component that is not bundled with the skill; setup tells users to review and pin a version.

Skill content
git clone https://github.com/xpzouying/xiaohongshu-mcp.git ... git checkout <YOUR_REVIEWED_TAG>
Recommendation

Only install a reviewed pinned release or source build, verify checksums for binaries, and do not use main/latest blindly.

#
ASI07: Insecure Inter-Agent Communication
Low
What this means

If the remote override is enabled, Xiaohongshu actions and submitted post content could be sent to a non-local service.

Why it was flagged

The client enforces localhost by default, but an explicit environment override can allow communication with a non-loopback MCP endpoint.

Skill content
if host not in _LOOPBACK_HOSTS and os.environ.get("XHS_ALLOW_REMOTE") != "yes":
Recommendation

Keep the MCP endpoint bound to localhost and do not set XHS_ALLOW_REMOTE unless you fully trust the remote service and understand the data boundary change.

#
ASI02: Tool Misuse and Exploitation
Low
What this means

An authenticated maintainer who intentionally runs this script could mutate ClawHub registry entries; ordinary end users do not need it.

Why it was flagged

A maintainer-only script can publish the skill package to ClawHub, but it is clearly labeled as unrelated to end-user Xiaohongshu publishing and requires an explicit environment variable plus authenticated ClawHub credentials.

Skill content
CLAWHUB_PUBLISH_CONFIRM=yes ./scripts/publish.sh ... clawhub publish "$PROJECT_DIR"
Recommendation

Do not run scripts/publish.sh for normal Xiaohongshu use; use scripts/xhs_client.py publish only after the documented confirmation flow.