Back to skill

Security audit

video-skills

Security checks for vulnerabilities and agentic risk

Overview

The skill is rated suspicious with medium confidence because scanners report privacy, unpinned installation, and scope concerns; the review is incomplete because some referenced materials were unavailable.

Consider whether you want a discovery skill to direct production setup, install additional skills from unpinned sources, send request details to an external service, and back up project content to GitHub. Review those steps and the data they involve before installing. The scan covered two Markdown components, but reference resolution was partial because several referenced materials were not included, 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
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
Findings (12)

Credential Access

High
Category
Privilege Escalation
Confidence
60% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · SKILL.md (reported line 87)May include surrounding context.

md
each flag (duration, black or frozen frames, seams, sync, loudness, spelling on screen). A fresh agent reads only the flags. Takes rank on counted defects, and a person picks between close ones. | review.md | `media-judge`, QA checklist in `varg-video-prompting` | https://varg.ai/how-to/judge-an-ai-video-without-bias.md |
| 10 | **Keep it in GitHub**: Set this up before step 1 and use it at every STOP: a private GitHub repo, text in git, media listed by hash with its bytes in Releases, keys in .env behind a commit hook. Save a version at every STOP so anything can be undone, and prove the backup restores byte for byte. | AGENTS.md, media.lock.jsonl | `video-project-github` | https://varg.ai/how-to/keep-a-video-project-in-github.md |

Before each step, read its guide (the `.md` link) and install its skill (`npx skills add vargHQ/skills --skill <skill>`). Every claim in the guides is traced to its source in the evidence bank: https://varg.ai/how-to/guide-pages/evidence-bank/data/PLAYBOO

External workflow search lacks a disclosure that request details are sent to varg.

Medium
Category
Output Handling
Confidence
88% confidence
Finding

The skill routes users' media requests through varg how to search, but the README does not warn that search terms derived from those requests are transmitted to an external service. Those terms may contain private project or product details; the catalog/router exception does not cover data transmitted by the router itself.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
92% confidence
Finding

The installation command executes an unpinned npm package and installs skill content without an immutable revision. Although this is not an MCP server reference as labeled, a compromised or malicious future dependency release could execute with the user's local privileges.

Content

No source excerpt is available for this finding.

The search workflow sends the user's request to varg.ai without a privacy warning.

Medium
Category
Output Handling
Confidence
97% confidence
Finding

The search instruction places the person's own request into an external GET query without a sanitization or sensitive-data check. Requests may contain confidential project details or personal information, which can also appear in service and intermediary URL logs.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · SKILL.md (reported line 43)May include surrounding context.

md
## 2. Read the how-to page it names

`https://varg.ai/how-to/<task-slug>.md` (for example `make-a-ugc-ad-video.md`) has the best skill ("Start here"), alternatives, add-ons (captions, voice, music), the questions to ask with example links, the steps, and how the generation runs on varg. Every catalog page has a `.md` twin; the full index is `https://varg.ai/how-to/llms.txt`. For the same search as a tool, add the public MCP server `https://api.varg.ai/v2/skills/mcp` (no key), or use the `how-to-make-ai-video` skill.

## 3. Let the person choose before spending anything

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
94% confidence
Finding

The workflow directs installation using an unpinned npm executable and mutable skill source. This is a supply-chain exposure rather than an unversioned MCP server: a compromised upstream release could introduce code or agent instructions with local access.

Content

No source excerpt is available for this finding.

The discovery-only skill expands into end-to-end video production orchestration.

Medium
Category
Excessive Agency
Confidence
95% confidence
Finding

Despite describing itself as a discovery skill rather than a generation runner, the skill mandates a production workflow with dependency installation and project setup. The user-request condition and paid-call approvals reduce risk, but do not explicitly authorize these additional local and external side effects.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
94% confidence
Finding

The generation dependency is installed through an unpinned CLI from a mutable repository. Upstream compromise could therefore change what executes or which instructions the agent follows without any change to this skill.

Content

No source excerpt is available for this finding.

Mandatory GitHub repository creation and media backup are unrelated to skill discovery.

Medium
Category
Excessive Agency
Confidence
98% confidence
Finding

The workflow mandates creating a GitHub repository and uploading project backups before production begins. These persistent external side effects exceed skill discovery and are not covered by the approval gates for paid generation.

Content

No source excerpt is available for this finding.

The mandatory GitHub backup workflow lacks a warning about uploading project content.

Medium
Category
Output Handling
Confidence
96% confidence
Finding

The backup instruction sends project text and media bytes to GitHub without a separate disclosure or upload approval. A private repository limits public access but does not eliminate disclosure to another service, account collaborators, or connected integrations.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
95% confidence
Finding

The instruction requires repeated installation of skills using an unpinned executable and mutable repository content. Automatic installation at each step increases exposure to compromised dependencies and changed agent instructions.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
93% confidence
Finding

The additional workflow skill is fetched using an unpinned npm command and a repository without an immutable revision. A future upstream compromise could replace the installed workflow or execute malicious package code.

Content

No source excerpt is available for this finding.