Back to skill

Security audit

AI-Video Workshop

Security checks for vulnerabilities and agentic risk

Overview

ClawScan rates this skill suspicious with medium confidence: scanners report that it may change its persistent guidance after projects and send project content to external generation services without a clear disclosure or approval step.

Review the skill’s persistent-update and external-service practices before installing. Consider whether you are comfortable with project-derived guidance being added to the installed skill and project materials being sent to external generation services. The scan was partial: 11 of 12 components were scanned, and one static analysis was degraded, so the evidence is incomplete.

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
Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • 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 (6)

L071-L076 mandate modifying the skill's own persistent requirements and examples after project work, beyond the stated video-production purpose.

Medium
Category
Excessive Agency
Confidence
96% confidence
Finding

The skill mandates writing project-derived rules and examples into its own persistent reference files, rather than limiting changes to project outputs. This creates a cross-project instruction-poisoning path: mistaken or adversarial project material could become trusted guidance for subsequent sessions. Continuous improvement is a plausible benign goal, but automatic promotion into executable agent guidance crosses a trust boundary.

Content

No source excerpt is available for this finding.

Mandatory project wrap-up modifies the installed skill without warning the user about persistent changes.

Medium
Category
Output Handling
Confidence
96% confidence
Finding

Project wrap-up is required to modify the installed skill without a separate disclosure or approval step. Authorization to produce a video does not necessarily authorize persistent changes affecting future projects or other users of the installation. This is the consent and persistence aspect of the same self-modification mechanism identified above.

Content

No source excerpt is available for this finding.

Direct use of external generation services lacks a warning about transmitting project content.

Medium
Category
Output Handling
Confidence
90% confidence
Finding

The workflow prioritizes external MCP generation and permits direct invocation when connected, without establishing a disclosure or authorization gate for sending project inputs to that provider. Scripts, reference images, and unreleased commercial assets could therefore leave the local environment unexpectedly. External generation is legitimate for this skill, but tool connectivity alone does not establish consent to disclose every project's content.

Content

No source excerpt is available for this finding.

External video generation is invoked automatically without a privacy warning about transmitting project assets.

Medium
Category
Output Handling
Confidence
92% confidence
Finding

The workflow automatically prioritizes connected external video-generation services and supplies reference images and prompts without requiring disclosure or authorization for that data transfer. Unpublished scripts, proprietary product imagery, or other confidential assets could therefore leave the local environment unexpectedly; external generation is legitimate for this skill, but a configured connector does not necessarily establish project-specific consent to upload assets.

Content

No source excerpt is available for this finding.

Enhanced video prompts require Chinese-only headings without user opt-in.

Medium
Category
Output Handling
Confidence
98% confidence
Finding

The instruction explicitly requires all four fixed headings to be in Chinese and prohibits English labels. It provides neither a user language choice nor a documented technical or regional justification for this restriction.

Content

No source excerpt is available for this finding.

The video-prompt style module requires Chinese without offering a language choice or explaining why that restriction is necessary.

Low
Category
Output Handling
Confidence
91% confidence
Finding

The instruction “全中文” explicitly requires the style module to be entirely in Chinese. The file provides neither user opt-in nor a documented technical or locale-specific justification for this language restriction.

Content

No source excerpt is available for this finding.