Back to skill

Security audit

Lossless Claw Skill

Security checks for vulnerabilities and agentic risk

Overview

This skill is purpose-aligned for Lossless Context Management, but it asks agents to install an unpinned third-party plugin, restart the OpenClaw gateway, and persist broad conversation history while under-disclosing those risks.

Review this before installing. The skill may be useful if you intentionally want Lossless Context Management, but installation should be explicit and should pin or verify the plugin version. Confirm you are comfortable with a plugin storing and searching conversation history locally, and expect an OpenClaw gateway restart during setup.

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 Third-Party Plugin Installation

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 18–26
Vulnerability Type: Unpinned and externally resolved third-party dependency
Risk Level: Medium

bash
Cuando un agente lee este skill, **debe verificar e instalar el plugin**:

```bash
# Verificar si el plugin está instalado
openclaw plugins list | grep lossless-claw

# Si NO está instalado → Instalarlo
openclaw plugins install @martian-engineering/lossless-claw
text

### Technical Analysis

The skill directs the agent to install the third-party package `@martian-engineering/lossless-claw` without specifying an exact version, cryptographic integrity hash, or trusted signed release. Package resolution can therefore retrieve a version that differs from the version originally reviewed.

The installation is subsequently followed by an OpenClaw gateway restart elsewhere in the documented workflow, which loads the installed plugin. Because the plugin's source code is not included in this project, the claims that it is audited, operates only locally, and performs no external communication cannot be independently verified from the audited artifact.

This constitutes an insecure dependency and supply-chain risk. Although no explicitly malicious payload is present in the reviewed files, a compromised publisher account, package registry, or future release could alter the effective code installed by this instruction.

### Attack Path

1. An attacker compromises the package publisher account, package distribution channel, or an unpinned future release.
2. The attacker publishes modified code under the same `@martian-engineering/lossless-claw` package name.
3. An agent loads this skill and follows its instruction to check for and install the plugin.
4. The package manager resolves the mutable package reference to the attacker-controlled release.
5. The documented gateway restart loads the modified plugin into OpenClaw.
6. The plugin can then act withi
...[truncated 803 chars]
Remediation
View remediation

Remediation Suggestions

  1. Pin the dependency to a specific, reviewed version rather than using an unconstrained package reference.
  2. Verify the package using a cryptographic integrity hash, signed release, or trusted lockfile before installation.
  3. Include the reviewed plugin source or a verifiable source commit in the project so its behavior can be audited alongside the skill.
  4. Require explicit, informed user approval before installing the plugin or restarting the gateway.
  5. Display the package source, exact version, requested permissions, and expected data access before installation.
  6. Run the plugin with least privilege and restrict its filesystem, network, process-execution, and conversation-data access to only what is required.
  7. Avoid asserting that the plugin is local-only, audited, or free of external calls unless those claims are supported by the pinned source and reproducible verification.
  8. Maintain an allowlist of approved package versions and reject installation if the resolved artifact does not match the approved digest.
Vulnerability Patterns
  • 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
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (5)

Intent-Code Divergence

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The security section claims the skill 'does not execute code,' but the same file explicitly instructs the agent to run installation and service-restart commands. This is dangerous because it misrepresents operational risk, potentially lowering user scrutiny around actions that modify the system and interrupt services.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The skill tells the agent to check, install a plugin, and restart the gateway, which are system-modifying and service-affecting actions, but it does not foreground the risks of package installation, supply-chain trust, or service interruption. In this context, the danger is elevated because the skill is trying to drive agent behavior directly, including automatic installation flow, which could normalize unauthorized environment changes.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill gives conflicting statements about plugin state: earlier it instructs the agent to verify and install the plugin if missing, while later it asserts the plugin is already installed and available. This inconsistency can cause unsafe assumptions about the environment and may lead an agent or user to trust unverified tool availability or skip necessary validation before using privileged functionality.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
96% confidence
Finding

The description, instructions, troubleshooting, and operational guidance are all presented in Spanish, which effectively imposes a language choice on users and agents reading the skill. There is no opt-in, alternative language option, or justification that the skill is intended only for a Spanish-speaking or region-specific environment.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
90% confidence
Finding

The description text is entirely in Spanish in a manifest intended to describe the skill, with no indication that the skill is region-specific or that users can choose their language. This can violate language/locale policy expectations when a skill implicitly forces a specific language without opt-in or documented justification.

Content

No source excerpt is available for this finding.