Meta Video Ad Deconstructor
ReviewAudited by ClawScan on May 1, 2026.
Overview
This appears to be a coherent Gemini-based video ad analysis skill, with main cautions around Google credentials, external AI processing of ad content, and a manual dependency install.
Before installing, confirm you are comfortable sending the analyzed ad content to Gemini, use a dedicated least-privileged Google service account, install Vertex AI dependencies from a trusted source, and review generated analysis before using it in downstream workflows.
Findings (4)
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
Installing or using the skill may require giving it access to a Google Cloud service account used for Gemini calls.
The skill setup requires a Google service-account credential for Gemini/Vertex AI access. This is purpose-aligned, but service-account credentials can carry broad cloud permissions and are not declared in the registry metadata.
GOOGLE_APPLICATION_CREDENTIALS=/path/to/service-account.json
Use a dedicated, least-privileged Google service account limited to the required Vertex AI/Gemini permissions, and avoid reusing broad administrative credentials.
Video ad transcripts, scene descriptions, and overlay text may be sent to Gemini for analysis.
The skill sends formatted transcript, scene, and text-overlay content to the configured Gemini model. This is core to the skill's function, but it means analyzed ad content leaves the local context for provider processing.
response = self.gemini_model.generate_content(full_prompt)
Do not process confidential or unreleased creatives unless your Gemini/Vertex AI data handling settings and policies are acceptable.
The installed dependency version may vary over time and depends on the user's package source.
The setup instructions require manually installing an unpinned package. This is expected for a Vertex AI integration, but version pinning and source verification are not provided.
pip install vertexai
Install dependencies from a trusted Python package index and consider pinning a reviewed version in your environment.
A crafted ad transcript or overlay could skew the analysis or produce unreliable JSON results.
Untrusted ad content is inserted directly into prompts for model analysis. This is necessary for the skill, but malicious or manipulative text in an ad could influence the model's output.
Transcript: {{transcript}}
Text overlays: {{text_timeline}}
Scenes: {{scenes}}Treat analysis outputs as advisory, review them before relying on them, and avoid automatically feeding results into high-impact workflows without validation.