hyperframes-cli

Pass

Audited by Gen Agent Trust Hub on Nov 21, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The capture command ingests untrusted data from external URLs (found in references/init-and-scaffold.md).
  • Ingestion points: The npx hyperframes capture command processes external website content.
  • Boundary markers: The documentation instructs agents to check for ok: false in JSON output or the presence of a BLOCKED.md file as a security signal.
  • Capability inventory: The skill can execute subprocesses via ffmpeg, docker, and terraform, and perform file-write operations across the project directory.
  • Sanitization: Instructions explicitly state that agents must not build or infer design data from partial/blocked files, establishing a clear security gate.
  • [DATA_EXFILTRATION]: The skill includes telemetry and a feedback reporting mechanism (documented in references/feedback.md and references/upgrade-info-misc.md). Telemetry includes environment metadata such as sandbox and agent runtime types. Feedback reports are sent to a public channel, but the skill provides strict guidelines on anonymizing data and redacting sensitive information like absolute paths, credentials, and machine identifiers.
  • [COMMAND_EXECUTION]: The skill executes local scripts and leverages external CLI tools like ffmpeg, docker, and terraform for its core functionality. It also includes instructions for managing infrastructure on AWS (Lambda) and GCP (Cloud Run), requiring the use of standard environment-based or configuration-file credentials.
  • [EXTERNAL_DOWNLOADS]: The skill manages its own browser environment, downloading a pinned version of Chrome (documented in references/doctor-browser.md) from a trusted source to ensure reproducible rendering across different host environments.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 21, 2026, 10:39 AM