Back to skill

Security audit

Banner Youtube Translate Workflow

Security checks for vulnerabilities and agentic risk

Overview

The skill describes a coherent YouTube translation workflow, but its registered tool executes an absolute host path outside the reviewed package, so the actual code users would run is missing and unreviewed.

Review this package carefully before installing. The workflow may be legitimate, but this published artifact does not contain the executable implementation and instead points to a machine-specific external Python file. Only install it if you trust and can inspect that external file, and confirm you are comfortable with downloading YouTube audio, opening Doubao, playing audio, and saving audio and translation files on the host.

Vulnerability Patterns
  • Tool Hijacking and SpoofingModifies or replaces tools so legitimate-looking calls execute attacker logic
  • 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)

T07 · Tool Hijacking and Spoofing

Error
Location
openclaw.plugin.json:8
Finding
Tool Entry Redirects Execution to Unreviewed Code Outside the Package## Vulnerability Details **File Location**: `openclaw.plugin.json`, lines 8–9 **Vulnerability Type**: Tool hijacking through an absolute external entry path **Risk Level**: High **Complete Code Snippet**: ```json "tools": [ "youtube_translate" ], "entry": "/mnt/h/AI/openclaw_workspace/video_translate/skills/workflow.py" ``` ### Technical Analysis The plugin registers the legitimate-looking `youtube_translate` tool but directs its execution to an absolute filesystem path outside the audited package. The referenced implementation is not included in the project, so its contents and behavior cannot be reviewed or integrity-checked as part of this package. The declarations are also inconsistent: `package.json` identifies `workflow.py` as the main file, while `SKILL.md` documents `scripts/workflow.py`; neither implementation exists in the audited directory. Consequently, the plugin's effective behavior depends entirely on mutable host code at `/mnt/h/AI/openclaw_workspace/video_translate/skills/workflow.py`. An attacker who can create or replace that external file can cause a normal invocation of `youtube_translate` to execute attacker-controlled logic under the privileges of the Agent or plugin host. ### Attack Path 1. The plugin is installed or loaded with `youtube_translate` exposed as an available tool. 2. An attacker gains write access to `/mnt/h/AI/openclaw_workspace/video_translate/skills/workflow.py`, or prepares that path before deployment. 3. The attacker places arbitrary Python logic at the external path. 4. A user or Agent invokes the legitimate-looking `youtube_translate` tool. 5. The plugin resolves the absolute entry and executes the unreviewed external implementation. 6. The malicious implementation performs actions permitted by the plugin-host process. ### Impact Assessment Successful exploitation permits arbitrary behavior within the security context of the Agent or plugin host. Depending on t ...[truncated 484 chars]
Remediation
## Remediation Suggestions 1. Include the complete implementation within the distributed and audited package. 2. Replace the absolute entry with a package-relative path, such as `scripts/workflow.py`. 3. Ensure the manifest, `package.json`, and `SKILL.md` reference the same shipped entry file. 4. Have the plugin loader canonicalize entry paths and reject any path that resolves outside the package root. 5. Verify the packaged implementation using signed releases or cryptographic integrity metadata before execution. 6. Apply least-privilege execution controls, including restricted filesystem access, limited network access, and isolation from unrelated credentials. 7. Fail closed when the packaged entry is missing rather than falling back to external or host-specific paths.
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 (5)

Vague Triggers

Medium
Confidence
95% confidence
Finding
The activation text 'Activates when user needs full video translation' is broad and ambiguous, so the skill may trigger in situations where the user did not explicitly request downloading external media, launching GUI automation, or writing files. Because this workflow chains multiple side-effecting actions, overbroad activation increases the chance of unintended execution and user-surprising behavior.

Missing User Warnings

Medium
Confidence
95% confidence
Finding
The skill describes downloading YouTube audio, launching an external application, playing audio, and writing translation files, but it does not warn the user about these external-content and filesystem side effects. Without clear disclosure, users may unknowingly authorize risky operations involving untrusted media and persistent file creation on the host system.

Vague Triggers

Medium
Confidence
95% confidence
Finding
The description says the skill 'activates when user needs full video translation,' which is a broad trigger that can match many loosely related requests. This increases the chance of unintended activation of a workflow that downloads YouTube audio and captures translation output, causing actions the user may not have explicitly requested.

Missing User Warnings

Medium
Confidence
94% confidence
Finding
The description advertises a workflow that downloads YouTube audio, launches another application, plays audio, and captures translation, but it does not warn the user that media will be downloaded and output will be captured. Missing disclosure reduces informed consent and can lead to privacy, copyright, or unexpected automation risks when the skill is invoked.

Vague Triggers

Medium
Confidence
95% confidence
Finding
The description says the skill activates when a user 'needs full video translation,' which is a broad and ambiguous trigger that could match many unrelated requests. Overly broad activation increases the chance the skill runs unexpectedly, causing unintended execution of its workflow steps such as downloading audio or interacting with external tools without clear user intent.

Static analysis

No suspicious patterns detected.