Back to skill

Security audit

OpenClaw Skill Advisor

Security checks for vulnerabilities and agentic risk

Overview

This is a markdown-only skill recommender, but it encourages installing named third-party skills without enough verification guidance and may trigger on ordinary task requests.

Review any recommended skill before installing it: confirm the publisher, source repository, current registry listing, requested permissions, version, and recent changes. Treat the listed package names and ratings as unverified examples unless ClawHub independently confirms them, and install only in a low-risk environment first.

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:166
Finding

Unverified Installation of Third-Party Skills

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 166, 201, 276, 297, and 318
Vulnerability Type: Third-party supply-chain exposure through unverified installation guidance
Risk Level: Medium

Vulnerable Code Snippets

text
clawhub install email-auto-sorter
text
clawhub install code-assistant-pro
text
clawhub install ppt-generator-ai

Technical Analysis

The Skill instructs users to install third-party Skills directly by package name. It does not require validation of the package publisher, source repository, package version, cryptographic signature, checksum, requested permissions, or independent security review before installation.

Although the document describes rating, download-count, and update-frequency criteria, the audited project contains no implementation that retrieves or verifies this metadata. Consequently, the example ratings and popularity claims cannot be treated as security controls or evidence that the named packages are authentic.

Name-only installation creates supply-chain risk because package ownership and package contents can change independently of this reviewed document. It may also expose users to typosquatting, name takeover, publisher impersonation, or a malicious package using a recommended name. The audit did not establish that any named package is currently malicious; the vulnerability is the absence of verification before installation.

Attack Path

  1. An attacker publishes, impersonates, or takes control of a Skill using a recommended or confusingly similar package name.
  2. A user invokes this advisor and receives the documented name-only installation command.
  3. The user runs the command without verifying the publisher, source, version, integrity, or requested privileges.
  4. ClawHub downloads and installs the attacker-controlled package.
  5. When the installed Skill is loaded or invoked, its instructions or executable components ...[truncated 1004 chars]
Remediation
View remediation

Remediation Suggestions

  1. Query current registry metadata before making a recommendation rather than relying on static names, ratings, or download counts.
  2. Verify the publisher identity, ownership history, official source repository, maintenance status, and security-review status of every recommended Skill.
  3. Pin installation recommendations to a reviewed version and verify a trusted cryptographic signature or checksum where supported.
  4. Display the permissions, tools, filesystem paths, credentials, and network destinations requested by the Skill before installation.
  5. Clearly mark fictional or illustrative package names, ratings, and download counts as examples that must not be treated as verified registry data.
  6. Require explicit user confirmation after presenting provenance and permission information.
  7. Recommend initial execution in an isolated, low-privilege sandbox without production credentials or sensitive data.
  8. Prefer an allowlist of reviewed publishers and packages, and reject ambiguous, unowned, unsigned, or unexpectedly changed packages.
  9. Revalidate package integrity and publisher ownership on updates instead of implicitly trusting later releases.
Vulnerability Patterns
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (3)

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The trigger phrases are broad and include common-language requests such as asking for help finding or recommending a skill. This can cause the skill to activate in unintended contexts, increasing the chance it overrides user intent and steers users toward installation guidance or third-party skills without sufficiently explicit consent.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill provides direct installation commands immediately after recommendation without a clear safety warning, verification step, or trust guidance. Because this skill's core purpose is to steer users toward installing other skills, omission of a warning materially increases the risk of users executing unreviewed commands and installing potentially unsafe third-party components.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The examples normalize activation on very general prompts like '帮我整理邮件', '我要写代码', and '做PPT', which are ordinary task requests rather than explicit requests for a skill recommender. In practice, this could divert normal conversations into unsolicited recommendation and installation flows, broadening the attack surface for social engineering or unsafe tool adoption.

Content

No source excerpt is available for this finding.