web-video-presentation
Pass
Audited by Gen Agent Trust Hub on Nov 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONCOMMAND_EXECUTIONDATA_EXFILTRATIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from external articles, creating a surface where malicious instructions in the data could influence agent behavior during code generation.
- Ingestion points: The agent reads the user-provided
article.mdand generatedscript.mdto design the presentation outline and React components. - Boundary markers: While the skill suggests using
---to separate beats, it lacks formal isolation or system-level delimiters to prevent data-to-instruction confusion. - Capability inventory: The agent has extensive capabilities including writing React components and CSS, executing shell commands for build processes, and performing network requests for audio synthesis.
- Sanitization: No explicit sanitization or filtering is applied to the input article content before it is processed by the agent.
- [DYNAMIC_EXECUTION]: The skill uses dynamic module loading and shell sourcing to handle its pluggable architecture for chapters and TTS providers.
- Evidence:
templates/scripts/extract-narrations.tsusesawait import(url)to load chapter narration files from computed paths. Additionally,templates/scripts/synthesize-audio.shuses thesourcecommand to load provider-specific shell scripts based on a variable name. - [COMMAND_EXECUTION]: The scaffolding process involves the execution of shell commands to set up the development environment and install dependencies.
- Evidence:
scripts/scaffold.shexecutesnpm create vite@latest,npm install, andnpm install --save-dev tsxto initialize the project structure. - [DATA_EXFILTRATION]: The skill is designed to send narration text to external cloud-based text-to-speech services, which is a required step for its primary functionality.
- Evidence: Integration scripts for OpenAI, MiniMax, ElevenLabs, Google Cloud, and Azure Speech utilize
curlto transmit text segments to remote APIs. - [EXTERNAL_DOWNLOADS]: The skill fetches assets and dependencies from established third-party services during the setup and execution phases.
- Evidence: The project installs Node.js packages via npm, imports typography from Google Fonts, and references preview assets hosted on a GitHub-backed CDN.
Audit Metadata