Back to skill

Security audit

Ai Intelligent Behavior Analysis

Security checks for vulnerabilities and agentic risk

Overview

The skill describes a behavior-monitoring app, but its install steps run unreviewed remote code and dependencies.

Review this skill carefully before installing. Only run it if you trust the external GitHub repository and dependencies at the time of installation, preferably in an isolated environment with limited filesystem, credential, camera/video, and network access. For real-world monitoring, confirm lawful consent, retention rules, alert review procedures, and false-positive handling before use.

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

Error
Location
SKILL.md:43
Finding
Execution of Mutable, Unreviewed Remote Code and Dependencies## Vulnerability Details **File Location**: `SKILL.md`, lines 43–46 **Vulnerability Type**: Remote payload retrieval and insecure dependency installation **Risk Level**: High ```bash git clone https://github.com/openclaw-skills/ai-intelligent-behavior-analysis cd ai-intelligent-behavior-analysis pip install -r requirements.txt python app.py ``` ### Technical Analysis The installation instructions retrieve source code from an external GitHub repository and execute it locally. The repository is cloned without selecting a verified commit, signed release, or immutable content digest. Consequently, its effective payload can change after this Skill has been reviewed. The supplied artifact contains neither `app.py` nor `requirements.txt`, so their contents could not be audited. The `pip install -r requirements.txt` command also installs an externally controlled dependency set without any visible version lock, package hash validation, or signature verification. Python package installation may execute package build or installation logic, while the subsequent `python app.py` command directly executes the retrieved application. This creates both a remote-payload risk and a software supply-chain risk. The artifact does not itself demonstrate malicious code, but its documented installation process crosses the review boundary and executes mutable, unverified content. ### Attack Path 1. An attacker compromises the referenced GitHub repository, a maintainer account, its default branch, or a dependency referenced by its requirements file. 2. The attacker modifies `app.py`, `requirements.txt`, or a transitive dependency to include malicious code. 3. A user follows the documented instructions and clones the current repository state. 4. `pip install -r requirements.txt` installs attacker-controlled packages and may execute malicious package installation or build logic. 5. `python app.py` directly runs the retrieved application code. 6. ...[truncated 1015 chars]
Remediation
## Remediation Suggestions 1. Include the complete, reviewable implementation and dependency manifests in the Skill package rather than requiring execution of a separate mutable repository. 2. If external retrieval is unavoidable, pin the repository to a specific reviewed commit hash instead of the default branch. 3. Distribute code through a signed release and verify its cryptographic signature or published checksum before installation or execution. 4. Replace the unconstrained requirements file with a lock file containing exact dependency versions and cryptographic hashes. Install with hash enforcement, such as `pip install --require-hashes`. 5. Review direct and transitive dependencies for dependency confusion, typosquatting, known vulnerabilities, and unexpected installation hooks. 6. Perform installation and execution in an isolated virtual environment or restricted container using a dedicated unprivileged account. 7. Restrict filesystem mounts, environment variables, credentials, devices, and outbound network access to the minimum required for behavior analysis. 8. Document the expected files, verified commit identifier, dependency versions, required permissions, and integrity-verification procedure. 9. Add automated checks that reject changed release artifacts, unlocked dependencies, and unsigned or unverified source revisions.
Vulnerability Patterns
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • 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 (4)

Missing User Warnings

Medium
Confidence
89% confidence
Finding
The skill describes real-time monitoring, behavior recognition, crowd analysis, trajectory analysis, and alert pushing without any user-facing warning about privacy, surveillance, false positives, or operational consequences. In a security and monitoring context, omission of these warnings can lead to unsafe deployment decisions, overreliance on automated alerts, and privacy-impacting use without informed safeguards.

Description-Behavior Mismatch

Medium
Confidence
95% confidence
Finding
The manifest description is inconsistent with the stated skill purpose of behavior analysis and anomaly detection, which weakens transparency about what the skill actually does. In agent ecosystems, mismatched metadata can cause inappropriate invocation, reviewer confusion, or concealment of broader functionality, increasing the chance of unsafe or unintended use.

Vague Triggers

Medium
Confidence
93% confidence
Finding
The description 'AI intelligent ai-intelligent-behavior-analysis' is too vague to define a narrow operational scope. Overly broad or unclear metadata can make the skill eligible for invocation in contexts beyond its intended use, which is risky for an AI capability involving behavior analysis because it may process sensitive surveillance, monitoring, or anomaly-detection tasks without clear boundaries.

Natural-Language Policy Violations

Low
Confidence
90% confidence
Finding
The natural-language description and headings are presented exclusively in Chinese, and there is no indication that users can opt into this language or that the skill is intentionally limited to a Chinese-speaking context. The policy requires avoiding forced language or locale constraints unless they are documented and justified.

Static analysis

No suspicious patterns detected.