Back to skill

Security audit

Her Agent Dashboard

Security checks for vulnerabilities and agentic risk

Overview

This is a local dashboard skill with some disclosure and accuracy issues, but I found no evidence of exfiltration, persistence, privilege escalation, or destructive behavior.

Before installing, understand that this dashboard is Chinese-only, mostly a static/local snapshot rather than true live agent telemetry, and its updater reads OpenClaw diary and learning-note markdown paths to produce a local HTML dashboard. For stronger safety, vendor or integrity-pin D3 and disclose the local files the updater reads.

Vulnerability Patterns
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • 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
  • 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 (1)

T03 · Remote Payload Retrieval and Execution

Warning
Location
index.html:7
Finding

Unpinned Remote JavaScript Is Retrieved and Executed at Runtime

Content
View full analysis
<\/script> ``` The dashboard generator embeds the same remote dependency into every regenerated dashboard: ```python html = f''' 🧠 Her-Agent 发展进度看板 `. In ordinary HTML source, the escaped form can cause incorrect parsing. This parsing defect does not eliminate the underlying security concern that the page requests an external executable resource. ### Attack Path 1. The user opens `index.html`, or opens a dashboard generated by `scripts/update_dashboard.py`. 2. The browser requests `https://d3js.org/d3 ...[truncated 1321 chars]
Remediation
View remediation
``` 2. Record and verify the vendored file's cryptographic checksum during releases. 3. If remote delivery is unavoidable, use an immutable versioned URL and Subresource Integrity: ```html ``` 4. Add a restrictive Content Security Policy that permits scripts only from explicitly approved sources and avoids `unsafe-inline` where practical. 5. Update both `index.html` and the template in `scripts/update_dashboard.py`; otherwise, running the updater will restore the vulnerable reference. 6. Replace `<\/script>` with the valid `` closing tag in ordinary HTML output and test the generated document with an HTML validator. ]]>
Vulnerability Patterns
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • 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
Findings (7)

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
70% confidence
Finding

Without declared permissions the skill's intent is opaque and cannot be validated.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The manifest description and the entire markdown content are written in Chinese, and there is no indication that users can choose another language or that the skill is intended only for a Chinese-language or region-specific context. This creates a natural-language locale policy concern because the skill effectively imposes a language preference without explicit opt-in.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The document declares lang="zh-CN", and the UI text and date formatting are consistently fixed to Simplified Chinese. Under the policy rule, forcing a specific language/locale without user opt-in is a natural-language policy violation unless the file clearly documents a justified region-specific scope, which it does not.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The manifest promises a real-time dashboard for self-awareness, emotions, knowledge graph, learning progress, and evolution status. In code, only simple counts from diary/notes are computed, while the knowledge graph, emotion system, evolution level, reflections, and timeline are hard-coded placeholders rather than live state derived from actual agent data.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

This code sets the HTML document language to zh-CN, and elsewhere uses Chinese-only UI text and toLocaleString('zh-CN'), which forces a specific language/locale experience. The policy allows locale constraints only when the user can opt in or when the restriction is clearly documented and justified, neither of which appears in this file.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The UI text says it '实时追踪自我意识、情感、知识图谱和进化状态', implying live tracking. However, the JavaScript refresh logic only re-renders the same embedded knowledge data, cycles through a fixed emotions array, and updates the displayed current time; it never loads new backend data or agent state.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
91% confidence
Finding

The code formats timestamps with toLocaleString('zh-CN'), which forces Chinese locale formatting for all users. This is a locale policy issue because no alternative or consent mechanism is provided, and the file does not state that the dashboard is intentionally limited to that locale.

Content

No source excerpt is available for this finding.