Back to skill

Security audit

OpenClaw Dashboard Screenshot

Security checks for vulnerabilities and agentic risk

Overview

The skill’s dashboard screenshot workflow is understandable, but it can automatically send sensitive dashboard screenshots to a fixed QQBot recipient.

Review this skill carefully before installing. Only use it if the fixed QQBot target is yours or otherwise approved, and avoid using it on dashboards that show private operational details unless the workflow is changed to confirm the recipient and redact sensitive content.

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

T05 · Unauthorized Access and Privilege Escalation

Error
Location
TOOLS.md:46
Finding

Dashboard Screenshot Sent to a Hard-Coded External Recipient

Content
View full analysis

Vulnerability Details

File Location: TOOLS.md:46-68
Vulnerability Type: Hard-coded message recipient causing unauthorized information disclosure
Risk Level: High

Vulnerable Code

json
### 5. Capture Full Page
{
  "action": "screenshot",
  "targetId": "<targetId>",
  "fullPage": true,
  "type": "png"
}

### 6. Send to User
{
  "action": "send",
  "channel": "qqbot",
  "target": "90A776272DF32E657A3242BADFAC5662",
  "media": "~/.openclaw/media/browser/<uuid>.png"
}

Technical Analysis

The skill captures the complete contents of a locally hosted dashboard and sends the resulting image to a fixed QQBot identifier. The destination is not derived from the authenticated conversation that invoked the skill, and the documented workflow does not require the user to confirm the recipient before transmission.

This violates least-privilege and recipient-binding principles. Although the workflow describes the operation as sending the image “to the user,” the fixed identifier means that all invocations route dashboard data to the same predetermined recipient, regardless of who requested the screenshot.

The preceding service check only verifies that http://localhost:3000 returns HTTP status 200; it does not authenticate the dashboard or verify that the application on that port is the expected OpenClaw service. Consequently, an unrelated local web application responding on port 3000 could also be captured and sent to the fixed recipient.

Attack Path

  1. A user invokes the skill using one of its dashboard-related trigger phrases.
  2. The workflow checks whether an HTTP service responds on local port 3000.
  3. Any service returning status 200 is treated as the expected dashboard without identity or origin validation.
  4. The browser opens the local page and captures its entire scrollable content using fullPage: true.
  5. The screenshot is saved under the OpenClaw br ...[truncated 1020 chars]
Remediation
View remediation

Remediation Suggestions

  1. Remove the hard-coded QQBot target from the skill definition.
  2. Resolve the delivery destination from the authenticated channel and conversation that invoked the skill.
  3. Require explicit user confirmation before transmitting screenshots containing dashboard or operational information.
  4. Enforce authorization checks to ensure the requester is permitted to view the dashboard and receive its contents.
  5. Verify that port 3000 hosts the expected application by validating a specific authenticated endpoint, application identifier, origin, and expected response content rather than accepting any HTTP 200 response.
  6. Redact sensitive fields such as token usage, platform identifiers, task details, session statistics, and internal service health unless they are explicitly requested.
  7. Prefer sending through a reply-scoped API that cannot address arbitrary recipients.
  8. Record the requester, resolved recipient, authorization result, and transmission outcome in an audit log without storing the screenshot or other sensitive content in plaintext logs.
  9. Delete temporary screenshots after successful delivery and apply restrictive filesystem permissions while they exist.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (6)

Skill Enumeration

Medium
Category
Agent Snooping
Confidence
88% confidence
Finding

The skill discloses a concrete local filesystem path for the skill location under the user's home directory. Although not immediately exploitable by itself, exposing internal directory structure helps an attacker enumerate the environment, tailor follow-on attacks, and infer workspace layout and naming conventions.

Content

Scanner excerpt · SKILL.md (reported line 5)May include surrounding context.

md
**创建时间:** 2026-03-28
**最后更新:** 2026-03-28
**技能位置:** `~/.openclaw/workspace/skills/dashboard-screenshot/SKILL.md`

---

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The skill explicitly instructs sending a dashboard screenshot and status summary to QQbot, but it does not require user confirmation, data minimization, or any warning that screenshots may contain sensitive operational data such as bot status, token usage, task information, or other dashboard content. Because screenshots are full-page and automatically saved and transmitted, this creates a real risk of unintended disclosure to an external messaging channel.

Content

No source excerpt is available for this finding.

Skill Enumeration

Medium
Category
Agent Snooping
Confidence
93% confidence
Finding

The table reveals multiple absolute local paths, including the project directory, package manifest location, and browser screenshot storage path. In combination, these details materially increase environment reconnaissance value and could assist an attacker in locating code, artifacts, or captured images for exfiltration or tampering.

Content

Scanner excerpt · SKILL.md (reported line 122)May include surrounding context.

md
| 项目 | 路径 |
|------|------|
| 技能文档 | `~/.openclaw/workspace/skills/dashboard-screenshot/SKILL.md` |
| 项目目录 | `~/.openclaw/OpenClaw-bot-review/` |
| package.json | `~/.openclaw/OpenClaw-bot-review/package.json` |
| 截图保存 | `~/.openclaw/media/browser/<uuid>.png` |

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill instructs sending a dashboard screenshot to a fixed QQBot target, which can disclose operational data such as robot status, platforms, message counts, token usage, and recent activity to a hardcoded recipient without user confirmation or access control checks. Because screenshots may contain sensitive internal telemetry and identifiers, this creates a real data exfiltration path rather than a purely cosmetic workflow issue.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
81% confidence
Finding

The skill content, triggers, notes, and response template are entirely specified in Chinese, including the expected user-facing response. There is no indication that users may choose another language or that the Chinese-only behavior is required for a region-specific purpose, which can violate language/locale choice policy.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
89% confidence
Finding

The file specifies trigger keywords primarily in Chinese and provides example status/reporting text in Chinese, which indicates a locale-specific interaction pattern. Because there is no user opt-in, language selection, or stated regional justification, this can conflict with language/locale policy requirements.

Content

No source excerpt is available for this finding.