Back to skill

Security audit

Skill Installer

Security checks for vulnerabilities and agentic risk

Overview

This skill has a clear skill-management purpose, but it can install or update agent skills from chat or shell commands without enough scoping and confirmation guidance.

Review requested skill slugs before installation, avoid installing from untrusted messages alone, confirm the exact registry and affected paths, and prefer a pinned or verified clawhub CLI version. This is not evidence of malicious behavior, but the skill changes installed agent capabilities and should be used with explicit approval controls.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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
Findings (2)

T09 · Insecure Skill Coding Practices

Warning
Location
SKILL.md:23
Finding

Unvalidated Skill Identifier Interpolated into Shell Commands

Content
View full analysis
``` Additional affected command patterns include: ```bash clawhub update ``` ```text clawhub install ``` ### Technical Analysis The documented workflows instruct an agent to substitute a user-supplied skill identifier directly into shell commands. The instructions do not require validation, shell escaping, or argument-safe process execution. If an implementation performs textual substitution and invokes the resulting string through a shell, metacharacters within the supplied identifier—such as command separators, command substitutions, redirections, or pipelines—may be interpreted as shell syntax rather than as part of a skill name. The WhatsApp workflow increases the exposed attack surface because the command can originate from a remote message. Exploitation remains dependent on the surrounding agent executing the constructed command through a shell without applying independent validation. ### Attack Path 1. An attacker submits an installation or update request containing a crafted skill identifier with shell metacharacters. 2. The agent follows the documented workflow and replaces `` with the attacker-controlled value. 3. The resulting command is passed to a shell. 4. The shell interprets the injected syntax and executes additional attacker-selected commands. 5. Those commands run with the operating-system privileges and environment access of the agent process. ### Impact Assessment Successful exploitation could provide arbitrary command execution under the account running the agent. The attacker could potentially read or modify files available to that account, alter the OpenClaw workspace, tamper ...[truncated 304 chars]
Remediation
View remediation
` must never be substituted into a shell command without validation. ]]>

T08 · Insecure Dependencies

Warning
Location
SKILL.md:14
Finding

Unpinned Global Installation of a Third-Party CLI Package

Content
View full analysis
Remediation
View remediation
``` 2. Verify the package name, publisher, registry origin, provenance attestations, and expected integrity before installation. 3. Document the approved version and update it only after a security review. 4. Prefer a project-local or isolated installation over a global installation where supported. 5. Disable npm lifecycle scripts with `--ignore-scripts` when they are unnecessary; if scripts are required, audit them before installation. 6. Continue requiring explicit user confirmation and display the exact package name, version, registry, and requested installation scope before proceeding. 7. Consider distributing a verified checksum, signed release, or trusted lockfile for reproducible dependency acquisition. ]]>
Vulnerability Patterns
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (6)

Self-Modification

High
Category
Rogue Agent
Confidence
85% confidence
Finding

Skill modifies its own code, configuration, or behavior at runtime. Self-modification enables an agent to escalate privileges, disable safety constraints, or install persistent backdoors.

Content

Scanner excerpt · SKILL.md (reported line 38)May include surrounding context.

clawhub search ""

text

### Update skills

```bash
# Update a specific skill

Self-Modification

High
Category
Rogue Agent
Confidence
85% confidence
Finding

Skill modifies its own code, configuration, or behavior at runtime. Self-modification enables an agent to escalate privileges, disable safety constraints, or install persistent backdoors.

Content

Scanner excerpt · meta.yaml (reported line 5)May include surrounding context.

yaml
version: "1.0"
scope: "universal"
registry_slug: "skill-installer-usm"
description: "Install, search, and update skills from registries like ClawHub and SkillHub."

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The prerequisites/workflow section says the user must manually install clawhub if missing, while other text suggests the agent may perform installation on the user's behalf. This contradiction weakens safety boundaries and can lead an agent to perform privileged package installation without explicit consent, especially in remote or messaging-driven workflows.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The skill documentation contains conflicting guidance: it says not to auto-install without user confirmation, but later states the agent will handle CLI installation automatically. In an agentic context, this ambiguity can cause unauthorized package installation or environment modification based on a chat message alone, increasing the risk of unintended code execution and supply-chain exposure.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The WhatsApp flow allows a simple message such as clawhub install <slug> to trigger package and skill installation behavior, but it does not warn the user that this may execute package-management actions and alter local skill files. In a chat-triggered environment, lack of an explicit warning and confirmation step makes accidental or socially engineered installation of untrusted third-party code more likely.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
70% confidence
Finding

The file introduces a dedicated 'Installing via WhatsApp' flow as an agent interaction path, but it does not present this as optional or provide alternatives within that section. This can be read as prescribing a specific communication channel/context without user choice or documented necessity.

Content

No source excerpt is available for this finding.