Back to skill

Security audit

varg-dropship-reel

Security checks for vulnerabilities and agentic risk

Overview

ClawScan marks this skill suspicious because it may send sensitive store or order screenshots to an external service without redaction guidance, and parts of its scripts were only partially inspected.

Consider avoiding installation until the screenshot workflow requires redaction or uses demonstration data, and until you are comfortable sending the workflow’s content to the named external services. The scan is incomplete: two scripts were only partially inspected, so their behavior is not fully assessed.

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
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (15)

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 100)May include surrounding context.

md
node scripts/captions.mjs words.json captions.ass --keywords "shopify,profit,demand,free,guide,china"

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 132)May include surrounding context.

md
p); voice on `eleven_v4` "Brian"; words on `groq_whisper_large_v3_turbo`; then `captions.mjs` + `assemble.mjs`. The hook read as phone footage with the product

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 101)May include surrounding context.

md
node scripts/assemble.mjs edit.json

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 132)May include surrounding context.

md
ven_v4` "Brian"; words on `groq_whisper_large_v3_turbo`; then `captions.mjs` + `assemble.mjs`. The hook read as phone footage with the product intact. In the la

Code capability 'env' detected in scripts/assemble.mjs but not covered by declared permissions.

High
Category
MCP Least Privilege
Confidence
75% confidence
Finding

The skill uses 'env' capability that is not covered by the declared 'allowed-tools' preapproval. Review the host runtime's actual enforcement policy; runtime permissions are unknown unless separately verified, and this declaration alone does not establish a permission ceiling.

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 45)May include surrounding context.

md
Captions: 1-2 words at a time, bold uppercase, white with a black outline, prices and keywords in lime, at 60 % of the height. Voice: one flat, confident male narrator, about 2.6 words per second, no music or a very quiet bed.

Generation runs on varg: the `varg-ai` skill (REST at `https://api.varg.ai/v2`) or the varg MCP server (`https://mcp.varg.ai/mcp`: `call_tool`, `get_job`, `estimate_cost`, `create_upload`). Every call returns a job; poll `GET /v2/jobs/{id}` until `completed`, the file is at `output.outputs[0].url`. Image-to-video jobs took 5-10 minutes in our test; one text-to-video job waited about an hour in the queue. Start them all at once and do the next step while they run.

## 1. Ask once, then go

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 87)May include surrounding context.

Word timings for the captions: transcribe the voiceover with groq_whisper_large_v3_turbo. It returns data.words with a start and end per word; whisper_large_v3 returns only sentence chunks.

bash
curl -s -X POST https://api.varg.ai/v2/tools/transcription/call -H "Authorization: Bearer $VARG_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "groq_whisper_large_v3_turbo", "audio_url": "VOICEOVER_URL", "language": "en"}'
# poll the job, then save its result (the object with "words") as words.json

The screenshot-generation workflow sends potentially sensitive store and order screenshots to an external AI service without a privacy warning.

Medium
Category
Output Handling
Confidence
94% confidence
Finding

The workflow requests real store and order-page screenshots and passes them to external image-generation services without requiring sanitization. Order screens can contain customer identities, addresses, or confidential business information, creating a credible unintended disclosure path even though external generation is an intended feature.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% 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 79)May include surrounding context.

Voiceover:

bash
curl -s -X POST https://api.varg.ai/v2/speech -H "Authorization: Bearer $VARG_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "eleven_v4", "voice": "Brian", "text": "<the script>"}'

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 79)May include surrounding context.

Voiceover:

bash
curl -s -X POST https://api.varg.ai/v2/speech -H "Authorization: Bearer $VARG_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "eleven_v4", "voice": "Brian", "text": "<the script>"}'

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · SKILL.md (reported line 122)May include surrounding context.

md
low 1 slows a shot, and a shot longer than its source holds the last frame. Shots should add up to the voiceover length. The script picks up Homebrew's `ffmpeg-full` by itself when the default `ffmpeg` can't burn subtitles.

**ffmpeg without captions support.** Homebrew's `ffmpeg` has no libass since 2025 (`ffmpeg -filters | grep subtitles` prints nothing). Install the keg-only `ffmpeg-full` next to it (`brew install ffmpeg-full`); it doesn't replace `ffmpeg`. Seen on macOS 15.1 (2026-10-27): the install upgraded shared libraries, after which the old `ffmpeg` failed to start (`Library not loaded: libx265…`) and `ffmpeg-full` failed on `freetype` (`Library not loaded: /usr/lib/libhvf.dylib`). Fix: `brew upgrade ffmpeg` and `brew reinstall --build-from-source freetype`. Don't run this on someone's machine without asking: it changes their system Homebrew.

Check the result: watch it once with sound and once muted; every price on screen must match the voice and the caption at that moment.

The ManyChat setup instructions omit a privacy warning about third-party processing of Instagram interactions.

Medium
Category
Output Handling
Confidence
90% confidence
Finding

The workflow instructs connecting an Instagram account to ManyChat and configuring automated comment replies and DMs. This involves third-party processing of commenters' account identifiers and interactions, but the file does not disclose that privacy impact or advise reviewing the integration's data access.

Content

No source excerpt is available for this finding.

The AI workflow sends real-store screenshots to a third-party service without a privacy warning or redaction guidance.

Medium
Category
Output Handling
Confidence
96% confidence
Finding

The source includes order and fulfillment lists among the required real screenshots, then instructs passing screenshots to an external image-generation service without a redaction check. This can expose customer information or internal account details to the provider and potentially preserve those details in media intended for publication.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% 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 · references/screen-demo.md (reported line 47)May include surrounding context.

Step A: first frame. Put the screenshot on a laptop with gpt_image_2_5 (best at keeping dense text). Pass the screenshot as the reference and ask for the exact screen:

bash
curl -s -X POST https://api.varg.ai/v2/image \
  -H "Authorization: Bearer $VARG_API_KEY" -H "Content-Type: application/json" \
  -d '{
    "model": "gpt_image_2_5",

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 · references/screen-demo.md (reported line 47)May include surrounding context.

Step A: first frame. Put the screenshot on a laptop with gpt_image_2_5 (best at keeping dense text). Pass the screenshot as the reference and ask for the exact screen:

bash
curl -s -X POST https://api.varg.ai/v2/image \
  -H "Authorization: Bearer $VARG_API_KEY" -H "Content-Type: application/json" \
  -d '{
    "model": "gpt_image_2_5",