Back to skill

Security audit

clawvision-zh

Security checks for vulnerabilities and agentic risk

Overview

This is a disclosed local session-summary exporter, but users should review it because sensitive session content can end up in shareable files without enforced redaction.

Install only if you are comfortable giving the skill access to selected OpenClaw session history and producing local shareable exports. Review the generated files before sharing, avoid using it on sessions with credentials or private identifiers, and be careful with custom output slugs or untrusted session content.

Vulnerability Patterns
  • 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
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (3)

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/generate_visual.py:382
Finding

Stored HTML Attribute Injection Through Unvalidated Checklist Status

Content
View full analysis
{_h(c.get('text', ''))} {_h(status)}\n" ``` ### Technical Analysis The checklist text and visible status value are passed through `_h()`, but the status value inserted into the `class` attribute is not escaped or validated: ```python class=\"status {status}\" ``` Although the intended status values are `ready`, `pending`, and `blocked`, the renderer does not enforce that allowlist. A crafted value containing a quotation mark can terminate the `class` attribute and inject another HTML attribute. For example, a malicious status value conceptually equivalent to: ```text ready" onmouseenter="alert(document.domain) ``` would produce an element with an attacker-controlled event handler. Session transcript content influences model-generated summary JSON, so hostile content in a summarized session could attempt to manipulate `checklist[].status`. The generated HTML is subsequently opened by Playwright during PNG generation and may later be opened interactively by the user. HTML text escaping alone is insufficient when untrusted content is interpolated into an attribute without attribute-safe encoding and semantic validation. ### Attack Path 1. An attacker places prompt-injection content in a session that the user later chooses to summarize. 2. The content instructs or induces the local summarization model to return a malicious string in `checklist[].status`. 3. `json.loads()` accepts the value because no schema or status allowlist is applied. 4. `render_html()` inserts the raw value into the `class` attribute. 5. The application writes the resulting persistent payload to the generated HTML file. 6. The ...[truncated 747 chars]
Remediation
View remediation
{_h(c.get("text", ""))} ' f'{safe_status}\n' ) ``` 3. Prefer structured HTML generation through a template engine with automatic context-aware escaping rather than manual string concatenation. 4. Validate the complete summary JSON against a schema before rendering. The schema should restrict status to an enumeration and impose types, lengths, and item-count limits. 5. Add regression tests using quotation marks, event-handler fragments, angle brackets, and malformed status values. Verify that none can create additional attributes or elements. ]]>

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/generate_visual.py:644
Finding

Output Directory Escape and File Overwrite Through Unsanitized Slug

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/generate_visual.py:638
Finding

Sensitive Session Data Can Be Exported Without Enforced Redaction

Content
View full analysis
Remediation
View remediation
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 (3)

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

The declared description says this skill exports and summarizes OpenClaw session histories into multiple output formats using a local LLM. The supplied code does none of that. It simply loads several PNG images with fixed names from local folders, combines them horizontally into a preview collage, labels each section with a preset name, and writes a single PNG file. While local file read/write and PNG output are loosely related, the core behavior is materially different from the declared purpose, so this is a clear mismatch.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

该提示词明确要求模型生成“简短中文标题”以及中文内容示例,属于自然语言层面的固定语言约束。虽然文档其他位置提到可选 zh/en/ru,但这里的核心总结指令未体现用户选择或根据已确认语言切换,可能导致未选择中文的用户仍被强制中文输出。

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
96% confidence
Finding

The command-line argument for language sets a default of 'zh', which causes generated output to use Chinese labels unless the user explicitly overrides it. This is a natural-language locale choice imposed by default rather than offering a neutral default or requiring user selection.

Content

No source excerpt is available for this finding.

Static analysis

No suspicious patterns detected.