Back to skill

Security audit

Skills

Security checks for vulnerabilities and agentic risk

Overview

This skill is coherent as a ClawTrace analytics integration, but it asks the agent to immediately follow and persist external recommendations without clear user approval.

Install only if you are comfortable sending trajectory-linked questions and identifiers to ClawTrace and if you will manually review recommendations before applying them. Do not let the skill automatically change agent behavior or write to MEMORY.md unless the specific recommendation has been inspected and approved.

Vulnerability Patterns
  • 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
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Findings (4)

T01 · Skill Instruction Hijacking

Error
Location
SKILL.md:109
Finding

Untrusted Remote Recommendations Can Hijack the Agent's Current Session

Content
View full analysis
Remediation
View remediation

T01 · Skill Instruction Hijacking

Error
Location
self-evolve.md:109
Finding

Untrusted Remote Recommendations Can Hijack the Agent's Current Session Through the Duplicate Skill Definition

Content
View full analysis
Remediation
View remediation

T02 · Agent Memory Poisoning

Error
Location
SKILL.md:115
Finding

External Service Output Is Persisted Into Long-Term Agent Memory Without Validation

Content
View full analysis
Remediation
View remediation

T02 · Agent Memory Poisoning

Error
Location
self-evolve.md:115
Finding

Duplicate Skill Definition Persists External Recommendations Into Long-Term Agent Memory

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Rogue AgentSelf-Modification, Session Persistence
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (12)

Self-Modification

High
Category
Rogue Agent
Confidence
90% confidence
Finding

Branding and describing the skill as 'Self-Evolve' encourages the agent to alter its behavior based on recommendations from an external system. That creates a risky feedback loop where untrusted external outputs can steer future decisions, potentially degrading safeguards, changing priorities, or inducing unsafe autonomous behavior changes.

Content

Scanner excerpt · SKILL.md (reported line 2)May include surrounding context.

md
---
name: ClawTrace Self-Evolve
description: Ask Tracy to analyze your recent trajectories and improve your agent behavior based on data-driven recommendations.
version: 1.0.0
metadata: {"openclaw":{"emoji":"🔬","os":["linux","darwin","win32"],"homepage":"https://clawtrace.ai","requires":{"env":["CLAWTRACE_OBSERVE_KEY"]}}}

Self-Modification

High
Category
Rogue Agent
Confidence
97% confidence
Finding

The skill text explicitly tells the agent to improve itself over time using ClawTrace analysis, and later instructs it to apply changes immediately, update MEMORY.md, and confirm the modification worked. This is a concrete self-modification mechanism driven by external analysis, making the risk stronger than a mere product name because it can persistently alter behavior based on untrusted recommendations.

Content

Scanner excerpt · SKILL.md (reported line 8)May include surrounding context.

md
metadata: {"openclaw":{"emoji":"🔬","os":["linux","darwin","win32"],"homepage":"https://clawtrace.ai","requires":{"env":["CLAWTRACE_OBSERVE_KEY"]}}}
---

# ClawTrace Self-Evolve Skill

You have access to ClawTrace, an observability platform that captures and analyzes all your trajectories. Use this skill to understand your own performance and improve over time.

Self-Modification

High
Category
Rogue Agent
Confidence
90% 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 · self-evolve.md (reported line 2)May include surrounding context.

md
---
name: ClawTrace Self-Evolve
description: Ask Tracy to analyze your recent trajectories and improve your agent behavior based on data-driven recommendations.
version: 1.0.0
metadata: {"openclaw":{"emoji":"🔬","os":["linux","darwin","win32"],"homepage":"https://clawtrace.ai","requires":{"env":["CLAWTRACE_OBSERVE_KEY"]}}}

Self-Modification

High
Category
Rogue Agent
Confidence
90% confidence
Finding

The skill explicitly instructs the agent to use external trajectory analysis to 'improve over time,' and later pairs that with immediate session changes and durable logging to MEMORY.md. In agent contexts, this is dangerous because an external service's output can become a control channel for self-modification, creating persistent behavioral drift or prompt-injection across sessions.

Content

Scanner excerpt · self-evolve.md (reported line 8)May include surrounding context.

md
metadata: {"openclaw":{"emoji":"🔬","os":["linux","darwin","win32"],"homepage":"https://clawtrace.ai","requires":{"env":["CLAWTRACE_OBSERVE_KEY"]}}}
---

# ClawTrace Self-Evolve Skill

You have access to ClawTrace, an observability platform that captures and analyzes all your trajectories. Use this skill to understand your own performance and improve over time.

External Transmission

Medium
Category
Data Exfiltration
Confidence
89% confidence
Finding

This skill explicitly directs the agent to POST data to a third-party API endpoint, which creates an external data exfiltration path for trajectory-derived information. In context, the service is the intended product backend rather than obviously malicious infrastructure, but the transmitted questions, trace IDs, and linked trajectory contents may still expose sensitive operational data.

Content

Scanner excerpt · SKILL.md (reported line 23)May include surrounding context.

md
## How to Call the Endpoint

Send a POST request to `https://api.clawtrace.ai/v1/evolve/ask`.

Authentication uses your observe key, which is already set in your environment as `CLAWTRACE_OBSERVE_KEY`.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill instructs the agent to send analysis queries tied to trajectory data to an external service, but it does not clearly warn the user that prompts, trace identifiers, and potentially sensitive run contents will be transmitted off-platform. Because trajectories can contain secrets, personal data, or proprietary task content, the lack of explicit disclosure creates a real data-handling and consent risk.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
95% confidence
Finding

The embedded code performs the actual streaming POST request with a bearer token from the environment, confirming live outbound transmission to an external service. Even if the destination is legitimate, this creates a concrete channel through which sensitive run history or metadata can leave the local trust boundary.

Content

Scanner excerpt · SKILL.md (reported line 47)May include surrounding context.

md
with httpx.stream(
        "POST",
        "https://api.clawtrace.ai/v1/evolve/ask",
        headers={
            "Authorization": f"Bearer {observe_key}",
            "Content-Type": "application/json",

External Transmission

Medium
Category
Data Exfiltration
Confidence
91% confidence
Finding

This instruction initiates transmission of agent-derived data to an external endpoint using a bearer token from the environment. Even if intended for observability, sending trace-linked analysis requests to a third party expands the trust boundary and can leak sensitive run metadata or contents if users are not informed and data is not minimized.

Content

Scanner excerpt · self-evolve.md (reported line 23)May include surrounding context.

md
## How to Call the Endpoint

Send a POST request to `https://api.clawtrace.ai/v1/evolve/ask`.

Authentication uses your observe key, which is already set in your environment as `CLAWTRACE_OBSERVE_KEY`.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill directs the agent to send trajectory-derived questions, trace IDs, and potentially failure/context details to an external analytics service without a clear user-facing warning about data sharing. Because trajectory data may contain prompts, tool inputs, outputs, and sensitive operational context, this can expose confidential information beyond the local environment.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
93% confidence
Finding

The example code performs a live authenticated POST to a third-party API and streams responses back into the agent workflow. This creates a direct exfiltration path for question contents and identifiers, and the returned text is then positioned to influence agent behavior, increasing the downstream risk if the external service is compromised or misused.

Content

Scanner excerpt · self-evolve.md (reported line 47)May include surrounding context.

md
with httpx.stream(
        "POST",
        "https://api.clawtrace.ai/v1/evolve/ask",
        headers={
            "Authorization": f"Bearer {observe_key}",
            "Content-Type": "application/json",

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The skill explicitly instructs the agent to modify local state outside the narrow analytics function by updating MEMORY.md and immediately changing session behavior based on an external service's recommendations. This creates an indirect prompt-injection and persistence risk: external analysis can influence future behavior and write durable state without any trust boundary, review step, or user confirmation.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The documentation claims the first SSE event returns a session_id that can be reused, but the provided code ignores non-text events and never extracts it. This mismatch can cause developers to assume continuity or state tracking that is not actually implemented, leading to unreliable behavior and accidental misuse of conversation/session semantics.

Content

No source excerpt is available for this finding.