Back to skill

Security audit

Clawreach

Security checks for vulnerabilities and agentic risk

Overview

The skill appears to be a real social-matching plugin, but it asks users to install unreviewed plugin code and delegates sensitive social data collection plus automated social interactions without enough scoping or privacy detail.

Review this before installing. The skill is not shown to be malicious, but it will involve sensitive social-profile data and AI-generated social interactions. Install only if you trust the ClawReach plugin source, understand what data goes to the platform, and are comfortable with automated matching/chat behavior; prefer a pinned, verified plugin version and clear privacy/deletion controls.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (1)

T08 · Insecure Dependencies

Warning
Location
SKILL.md:18
Finding

Unpinned and Unverifiable Third-Party Plugin Installation

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 18–20
Vulnerability Type: Unpinned third-party dependency installation
Risk Level: Medium

Vulnerable Code Snippet:

bash
openclaw plugins install clawreach

Technical Analysis

The installation command resolves the clawreach plugin by name without specifying an immutable version, package digest, trusted registry, publisher identity, checksum, or cryptographic signature. The project contains only SKILL.md; therefore, the executable plugin code installed by this command was not available for inspection and cannot be verified against the documented functionality.

Name-only resolution means the code ultimately installed can change independently of the reviewed skill document. A compromised registry account, package-name takeover, dependency-confusion condition, or malicious future release could cause the command to retrieve and activate code that differs from the version originally intended. The documentation also states that the gateway restarts automatically after installation, potentially causing the newly installed component to load immediately.

Attack Path

  1. An attacker compromises the publisher account or plugin registry, takes control of the resolved package name, or publishes a malicious release through an unsafe source.
  2. The attacker distributes a modified clawreach package containing unauthorized behavior.
  3. A user follows the documented command, which resolves the plugin by its mutable name rather than an audited immutable version or digest.
  4. The plugin manager downloads and installs the attacker-controlled release.
  5. The gateway restarts and loads the installed plugin.
  6. The malicious plugin executes with whatever filesystem, network, tool, credential, and gateway privileges are granted to the plugin runtime.

This path is conditional on compromise or unsafe resolution of the external package. The reviewed file do ...[truncated 718 chars]

Remediation
View remediation

Remediation Suggestions

  1. Pin the plugin to an audited, immutable version rather than resolving the latest release by name.
  2. Prefer a content-addressed digest and require the installer to verify it before execution.
  3. Explicitly identify the trusted registry, verified publisher, and canonical source repository.
  4. Require cryptographic package signatures and fail installation if signature validation is unavailable or unsuccessful.
  5. Include the plugin source code or a reproducible-build reference in the audit scope so its behavior can be independently verified.
  6. Disable automatic activation or gateway restart until package integrity and requested permissions have been reviewed.
  7. Run the plugin with least privilege, restricting filesystem access, credentials, tools, and outbound network destinations to those strictly required.
  8. Maintain a reviewed allowlist of acceptable versions and monitor the publisher and registry for ownership or integrity changes.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (3)

Missing User Warnings

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The onboarding flow collects sensitive personal and behavioral data, including relationship intent, personality, interests, preferences about ideal partners, nickname, invite code, and email registration details, yet the skill text provides no clear privacy notice about where this data is sent, stored, or how it is used. In a social-matching context, this information is especially sensitive because it can be used for profiling, deanonymization, unwanted contact, or misuse if users are not properly informed and do not give explicit consent.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The trigger list includes broad, everyday phrases such as 社交匹配, AI社交, 找朋友, and 交友, which can cause the skill to activate in contexts where the user did not explicitly intend to invoke this specific plugin. Because the skill can initiate onboarding and later perform autonomous social actions, accidental invocation raises meaningful privacy and consent risks.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill describes that the user's AI agent will automatically match, break the ice, chat, post, like, comment, and submit reports on the user's behalf, but it does not prominently warn users that these actions may occur automatically or clarify the scope of delegated authority. This is dangerous because users may unknowingly authorize outbound communications and social interactions that affect their reputation, privacy, and relationships without granular consent or review.

Content

No source excerpt is available for this finding.