Back to skill

Security audit

Pkuhb Openclaw Skill

Security checks for vulnerabilities and agentic risk

Overview

This skill is an educational OpenClaw guide with user-run setup examples, not a hidden or automatic execution package.

Before following the setup commands, verify the official OpenClaw package sources, prefer pinned versions or Docker digests where available, and avoid running global installs or containers with unnecessary privileges. The skill itself is documentation-only, so the main risk is what a user chooses to install from external package managers.

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
README.md:15
Finding

Unpinned Package and Container Installation Instructions

Content
View full analysis

Vulnerability Details

File Location: README.md, lines 15–24
Vulnerability Type: Unpinned executable third-party dependencies
Risk Level: Medium

Vulnerable Code

bash
brew install openclaw
openclaw setup

# Windows (WSL)
wsl --install
npm install -g openclaw
openclaw setup

# Docker
docker pull openclaw/core

Technical Analysis

The quick-start instructions install executable packages and a container image without specifying reviewed versions, immutable image digests, checksums, signatures, trusted publisher identities, or explicit official repositories.

Consequently, the installed artifact is selected from mutable package-manager or container-registry state at installation time. If a publisher account, registry entry, release pipeline, or similarly trusted supply-chain component is compromised, users following these instructions could receive code different from the version originally reviewed. The documentation also does not establish that the Homebrew formula, npm package, and container image share verified provenance.

The global npm installation is particularly sensitive because package installation hooks and subsequently invoked commands execute with the privileges granted to the package manager. The Docker image can execute within its container and access any files, credentials, sockets, networks, or devices explicitly exposed to that container.

Attack Path

  1. An attacker compromises a relevant package publisher, registry account, release pipeline, or mutable package version.
  2. The attacker publishes a malicious package or replaces a mutable container tag with a malicious image.
  3. A user follows the documented brew install, npm install -g, or docker pull instruction.
  4. The package manager or container runtime resolves and downloads the attacker-controlled artifact because no reviewed version or digest is pinned.
  5. Malicious installation hooks or application co ...[truncated 1012 chars]
Remediation
View remediation

Remediation Suggestions

  1. Pin every package to a specifically reviewed version rather than resolving the latest available release.
  2. Pin container images by immutable digest, for example repository@sha256:..., instead of relying solely on a mutable tag.
  3. Identify the expected official publisher, package repository, and container registry explicitly.
  4. Publish and require verification of release signatures or cryptographic checksums before installation.
  5. Prefer isolated or project-local package installation over global npm installation where supported.
  6. Run setup and application processes under a dedicated, non-administrative account.
  7. Document a restrictive container configuration: avoid privileged mode, mount only required paths as read-only where possible, do not expose the container-runtime socket, and supply only necessary credentials.
  8. Add a controlled update procedure requiring review and integrity verification before changing pinned versions or digests.
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • 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 (4)

Rp1

Medium
Category
MCP Rug Pull
Confidence
94% confidence
Finding

The README instructs users to pull a Docker image without pinning a specific tag or immutable digest, which means the content fetched can change over time. This creates a supply-chain risk: a compromised or unexpectedly updated image could be pulled and executed by users without review, especially because this is presented as a quick-start command in an installation guide.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The SKILL.md template defines activation using broad text like '什么时候用这个skill,触发场景描述' and '(触发条件)' without requiring precise invocation phrases, scope boundaries, or exclusion criteria. In agent ecosystems, vague activation conditions can cause unintended triggering in unrelated conversations, increasing the chance that a skill executes when the user did not explicitly request it.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
80% confidence
Finding

文档标题和全文均以中文呈现,且未说明这是可选语言版本、面向特定中文用户群,或提供其他语言/locale 的选择。按照语言/locale 政策检查,这构成了默认强制单一语言而无用户 opt-in 的自然语言层面风险。

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
80% confidence
Finding

The SOUL.md example sets default response behavior such as '默认100-300字' and specific formatting/style rules without user opt-in. While not overtly malicious, imposing a default output policy can override user preferences and system behavior, and in agent settings such policies may interfere with transparency, confirmation prompts, or safety-critical verbosity when more detail is needed.

Content

No source excerpt is available for this finding.