ClawHub
Back to skill

Security audit

sjht-cam-anno

Security checks for vulnerabilities and agentic risk

Overview

This skill is a disclosed local workflow for turning security-camera videos into annotated training data, with no evidence of hidden execution, exfiltration, persistence, or destructive behavior.

Install only if you intend to process the specified security-camera video dataset. Review the input and output directories before running it, because it creates frame images and training records that may contain sensitive information about people, children, elderly persons, homes, or workplaces.

A.I.G

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

SkillSpector

Vulnerability Patterns
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • 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
Findings (11)

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The skill claims it can analyze video content, generate structured annotations, and output dataset.jsonl training data, but the described/observed implementation does not actually provide those capabilities. This mismatch is dangerous because operators may trust the skill to perform higher-level analysis or data generation it cannot do, causing unsafe automation decisions, incorrect training data, or unintended handling of sensitive surveillance footage.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
84% confidence
Finding

The skill instructs use of shell execution plus file read/write behavior, but it does not declare any explicit tool scope or permissions boundaries. That makes the runtime authority ambiguous and increases the chance the agent will execute filesystem or shell actions more broadly than intended, especially given references to processing arbitrary video directories.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The invocation description uses broad trigger phrases such as generic requests for data annotation, video annotation, or VL fine-tuning, which can cause the skill to activate for requests outside its narrowly intended context. Over-broad matching is risky because it may route unrelated tasks involving local files, videos, or shell-capable processing into a skill that reads/writes data and invokes scripts.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The document is written entirely in Chinese and provides no indication that users may choose another language or locale. Under the stated policy, forcing a specific language without user opt-in is a natural-language policy violation unless the locale restriction is explicitly documented and justified.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The prompt is written entirely in Chinese and requires output constraints such as '不超过20个汉字', which effectively enforces a specific language/locale. The file does not offer user opt-in or an alternative language choice, and no region-specific justification is documented.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
85% confidence
Finding

The module docstring describes this script as converting existing video annotation results into dataset.jsonl records. However, build_jsonl_entry constructs training examples containing a system prompt and assistant output only, plus an image path, without any user-side instruction or explicit aggregation logic beyond direct wrapping. This is a meaningful intent/documentation mismatch about what format is actually being produced.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The embedded natural-language instructions require the model to produce output in Chinese, including Chinese field content and a limit measured in Chinese characters, but provide no user opt-in or configurable language choice. This creates a language/locale policy issue because the skill forces a specific language regardless of user preference.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The module docstring presents the skill entirely in Chinese, and the rest of the user-facing strings in the file are also Chinese-only. For a general-purpose extraction tool, this imposes a specific language/locale without user opt-in or any documented reason that the tool is region-specific.

Content

No source excerpt is available for this finding.

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · scripts/extract_frames.py (reported line 31)May include surrounding context.

python
"ffmpeg", "-y", "-ss", str(timestamp), "-i", video_path,
            "-frames:v", "1", "-q:v", "2", out_path
        ]
        result = subprocess.run(cmd, capture_output=True, text=True)
        if result.returncode == 0 and os.path.exists(out_path):
            frame_paths.append(out_path)
        else:

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · scripts/extract_frames.py (reported line 48)May include surrounding context.

python
"ffmpeg", "-y", "-ss", str(timestamp), "-i", video_path,
            "-frames:v", "1", "-q:v", "2", out_path
        ]
        result = subprocess.run(cmd, capture_output=True, text=True)
        if result.returncode == 0 and os.path.exists(out_path):
            frame_paths.append(out_path)
        else:

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
67% confidence
Finding

The natural-language description is written to activate on Chinese phrases and presents the skill behavior in Chinese only, with no indication that users may choose another language. This can be interpreted as a language/locale constraint without explicit opt-in.

Content

No source excerpt is available for this finding.

Static analysis

No suspicious patterns detected.