hyperframes-cli
Pass
Audited by Gen Agent Trust Hub on Nov 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of untrusted external data via the
npx hyperframes capture <url>command. This command scrapes a live website to generate a video project, creating a surface for indirect prompt injection if the source content contains malicious instructions designed to influence the agent. - Ingestion points:
npx hyperframes capture(documented inSKILL.mdandreferences/init-and-scaffold.md) reads content from arbitrary URLs. - Boundary markers: The skill instructions explicitly warn agents to treat a
BLOCKED.mdfile or a non-zero exit code as a hard stop and to avoid building or inferring data from partial files. - Capability inventory: The skill possesses extensive capabilities across its scripts, including rendering media, executing cloud deployments (AWS SAM/Terraform), and uploading project assets to remote servers.
- Sanitization: The
capturetool provides anokflag in its JSON response to signal semantic success, which the agent is instructed to validate. - [EXTERNAL_DOWNLOADS]: The skill performs several external downloads necessary for its operation. These target well-known and reputable services:
- Downloads a pinned version of Chromium for consistent video rendering via
npx hyperframes browser ensure. - Fetches project templates, examples, and AI coding skills from GitHub.
- Downloads a quantized ONNX model (~33 MB) for on-device semantic catalog search, which requires explicit user consent.
- [DATA_EXFILTRATION]: The skill is designed to upload project data (HTML, assets, scripts) to HeyGen-managed cloud infrastructure or user-owned AWS/GCP buckets for rendering and publishing. While this involves sending local project files to remote servers, it is the primary intended functionality of the CLI and is clearly documented. The tool also collects anonymous telemetry (OS version, CPU/Memory stats, and runtime environment fingerprints) to distinguish managed-sandbox runs, which can be disabled via environment variables or subcommands.
- [DYNAMIC_EXECUTION]: The underlying tool functions by rendering HTML, CSS, and JavaScript within a headless browser to produce video frames. This involves the dynamic execution of project code, which is the core purpose of the HyperFrames CLI as a video engine.
Audit Metadata