Back to skill

Security audit

Ai Intelligent Call Center

Security checks for vulnerabilities and agentic risk

Overview

The skill describes a call-center app, but its install instructions run unreviewed remote code and the documentation underspecifies privacy controls for recordings and AI quality review.

Review this skill carefully before installing. Only run the GitHub install commands in an isolated environment after verifying the repository owner, pinning a reviewed commit, and reviewing dependencies. For any production call-center use, define recording consent, retention, deletion, access controls, and audit practices before processing real calls.

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:42
Finding

Unpinned Remote Payload Retrieval and Execution

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 42-45
Vulnerability Type: Remote payload retrieval followed by local execution
Risk Level: High

Vulnerable Code:

bash
git clone https://github.com/openclaw-skills/ai-intelligent-call-center
cd ai-intelligent-call-center
pip install -r requirements.txt
python app.py

Technical Analysis

The documented installation procedure retrieves a mutable Git repository without pinning or verifying a commit, release artifact, checksum, or signature. It then installs dependencies specified by the retrieved repository and executes its application.

The reviewed skill package does not contain the referenced application source or requirements.txt. Consequently, the effective code and dependency set cannot be audited from this artifact and may change after this skill has been reviewed. Running pip install can itself execute package build or installation logic, while python app.py directly executes the remotely retrieved application.

The publisher metadata is also inconsistent: SKILL.md identifies the author as “OpenClaw Skills Team,” while skill.json identifies "yang1002378395-cmyk". This inconsistency does not independently establish malicious behavior, but it reduces confidence in the provenance of the remote repository.

Attack Path

  1. An attacker compromises the referenced repository, an authorized publisher account, or an unpinned dependency referenced by its requirements.txt.
  2. The attacker adds malicious application, package-build, or dependency code.
  3. A user follows the installation instructions and clones the repository's current state.
  4. pip install -r requirements.txt executes applicable package installation or build logic.
  5. python app.py executes the downloaded application.
  6. The payload operates with the privileges of the user running these commands.

Impact Assessment

Successful exploitation could pr ...[truncated 497 chars]

Remediation
View remediation

Remediation Suggestions

  1. Include the complete application source and dependency manifest in the reviewed skill package so that the executed implementation matches the audited artifact.
  2. If remote retrieval is necessary, pin the repository to a specific reviewed commit rather than cloning a mutable default branch.
  3. Distribute signed release artifacts and verify their cryptographic signatures or SHA-256 checksums before installation or execution.
  4. Lock every Python dependency to an exact version and require hashes, for example through a generated lock file and pip install --require-hashes.
  5. Review transitive dependencies and use a trusted package index with controls against dependency confusion and typosquatting.
  6. Run installation and application startup in an isolated, least-privileged environment without unnecessary secrets, host filesystem access, or administrative permissions.
  7. Reconcile the author information in SKILL.md and skill.json, and document the verified ownership relationship between the skill publisher and the referenced repository.
Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • 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 (3)

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill explicitly advertises call recording and AI quality inspection, both of which involve handling highly privacy-sensitive communications data, but it provides no warning about consent, retention, access control, or regulatory obligations. In a call-center context this omission can lead operators to deploy the skill without safeguards, creating risk of unlawful recording, excessive data collection, or exposure of transcripts and evaluations.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
92% confidence
Finding

The natural-language content of the skill, including the description and usage sections, is fully in Chinese. Because no opt-in, language selection, or justification for a Chinese-only experience is provided, this may violate a language/locale policy requiring user choice or explicit documentation of locale constraints.

Content

No source excerpt is available for this finding.

Vague Triggers

Low
Category
Not specified by scanner
Confidence
85% confidence
Finding

This manifest description is extremely broad and repetitive, and it does not indicate when the skill should or should not be invoked. Without domain-specific trigger constraints or exclusion examples, the activation scope is ambiguous and could overlap with common call-center or automation requests.

Content

No source excerpt is available for this finding.