Back to skill

Security audit

OpenClaw NovelAI

Security checks for vulnerabilities and agentic risk

Overview

This is a disclosed NovelAI workflow helper that can use a configured NovelAI token and spend NovelAI resources when requested, but I found no hidden or deceptive behavior.

Install only if you are comfortable configuring a host-managed NOVELAI_TOKEN and using the pinned third-party NovelAI Image MCP server. Do not paste the token into chat or files. Start with tool discovery and a cost estimate, and be aware that image generation and the optional live API test can consume NovelAI resources.

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
Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • System Prompt LeakageDirect Leakage, Indirect Extraction, Tool-Based Exfiltration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (15)

Tp4

High
Category
MCP Tool Poisoning
Confidence
83% confidence
Finding

The description emphasizes OpenClaw creative workflows spanning fiction/chapter-planning and cost-aware execution with billing-profile reuse and bounded retries. The supplied code instead implements a sequential live API capability test harness. It calls NovelAI account endpoints, enumerates text models, tests text completion/chat, exercises multiple image endpoints, saves outputs locally, and emits a diagnostic report. Some declared image-related capabilities do match (V4.5/V5 generation, img2img, inpainting, Vibe/Director-like tools, local cost estimate, token redaction), but the code materially differs in purpose and includes undeclared diagnostic/account-testing behaviors while omitting several declared workflow features.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The code chunk is narrowly focused on local filesystem project management and metadata handling. It creates a project layout, reads context markdown files, builds a bounded prompt string, and stores generation metadata with basic redaction of sensitive-looking keys. While this partially aligns with 'fiction context' and 'secret-safe asset records,' the declared description substantially overstates the implemented behavior. There is no network/API interaction, no NovelAI request execution, no chapter-planning logic, no image prompt/generation/edit operations, no V5/V4.5 model handling beyond free-form metadata fields, no billing-profile reuse, and no retry mechanism. Therefore the description does not accurately represent what this code actually does.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The declared description presents an operational NovelAI creative workflow skill with generation, editing, planning, billing reuse, retries, and secret-safe asset behavior. The actual code chunk does not implement any of those runtime capabilities. It is only a test file that inspects SKILL.md text for required metadata and policy language, especially around a free/zero-cost variant and blocked paid operations. That is a materially different primary purpose from the declared skill behavior, so this is a mismatch.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The declared description says the skill performs NovelAI creative workflows and operational behaviors such as generation, editing, cost-aware execution, retry handling, and session billing-profile reuse. However, the provided code chunk does not implement any of those runtime capabilities. It is only a test suite that inspects documentation files for required metadata and wording. This is a materially different primary purpose: documentation/contract validation rather than NovelAI workflow execution. Therefore the description does not accurately represent what this code chunk actually does.

Content

No source excerpt is available for this finding.

Direct Prompt Extraction

High
Category
System Prompt Leakage
Confidence
85% confidence
Finding

Skill contains instructions that could directly expose system prompts, internal rules, or hidden instructions to users or external parties.

Content

Scanner excerpt · scripts/live_api_test.py (reported line 697)May include surrounding context.

python
def director(self, tool: str) -> None:
        if not self.source_image or not self.no_auth_suite_guard():
            self.add(f"director_{tool}", "skip", details={"reason": "source image/auth unavailable"})
            return
        prompt = ""
        if tool == "colorize":
            prompt = "blue hair, warm evening light"

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
89% confidence
Finding

The skill advertises shell, file, network, and environment-dependent behavior but does not declare an explicit tool scope such as allowed-tools or permissions. That makes the operational boundary unclear and can let the host grant broader capabilities than users or reviewers expect, increasing the blast radius if the skill is invoked in a permissive environment.

Content

No source excerpt is available for this finding.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
80% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · SKILL.md (reported line 20)May include surrounding context.

md
## Security rules

- Never ask the user to paste a NovelAI token, password, cookie, or bearer header into chat.
- Never place a credential in a prompt, command argument, URL, generated file, log, or generation metadata.
- Treat every third-party MCP server, CLI, package, and skill as untrusted until the operator has reviewed and approved it.
- Do not auto-install packages. Explain the prerequisite and wait for operator approval.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
80% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · tests/test_skill_contract.py (reported line 60)May include surrounding context.

python
## Security rules

- Never ask the user to paste a NovelAI token, password, cookie, or bearer header into chat.
- Never place a credential in a prompt, command argument, URL, generated file, log, or generation metadata.
- Treat every third-party MCP server, CLI, package, and skill as untrusted until the operator has reviewed and approved it.
- Do not auto-install packages. Explain the prerequisite and wait for operator approval.

Unbounded Resource Access

Medium
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill allows unbounded resource consumption (API calls, storage, compute). Without rate limits or quotas, a compromised or misbehaving agent can cause denial-of-service or cost overruns.

Content

Scanner excerpt · docs/FULL-USER-MANUAL.md (reported line 474)May include surrounding context.

md
### Dedicated upscaling returns 404

The current `/ai/upscale` route may be unavailable. Do not retry indefinitely; use larger-resolution img2img or a local upscaler and describe the difference.

## 10. Recommended first-run instruction

Rp1

Medium
Category
MCP Rug Pull
Confidence
65% confidence
Finding

uvx/uv tool run commands without ==version create a rug-pull risk.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
87% confidence
Finding

The language selector line presents 'English README | 简体中文', making this document exclusively Chinese-language content for the current file. Under the policy rule, forcing a specific language without user opt-in can be a locale-policy issue unless the file explicitly offers language choice within the skill experience or documents a justified locale constraint.

Content

No source excerpt is available for this finding.

Vague Triggers

Low
Category
Not specified by scanner
Confidence
79% confidence
Finding

The phrase 'Use this skill when the user wants a NovelAI-assisted story workflow ... long-form writing, chapter context, character consistency, advanced image generation, image editing, or asset tracking' covers very broad user intents and does not provide explicit exclusion conditions. Because many of these intents are common across general creative tools, the activation scope could overlap with ordinary requests that are not specifically meant for this skill.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
86% confidence
Finding

The example user intents are presented only in Chinese, which can imply a language preference or default without any user opt-in. The document does not state that language should follow user preference or that the skill is region-specific.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
97% confidence
Finding

This markdown file contains user-facing instructions only in Chinese, including all setup, safety, and usage guidance. Under the language/locale policy rule, forcing a specific language without user opt-in can be a natural-language policy violation unless the locale constraint is documented and justified.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
89% confidence
Finding

该说明以自然语言规定用户应使用中文下指令,属于语言/locale 约束。文档中未提供其他语言选项、未说明只是示例,也未给出必须限定中文的合规性理由,因此构成轻微的语言政策风险。

Content

No source excerpt is available for this finding.