The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses ingestion surfaces that process untrusted data, which could contain malicious instructions.
Ingestion points: The capture command (references/init-and-scaffold.md) fetches content from external URLs to scaffold projects. The preview --context command (references/preview-render.md) retrieves state and metadata from a running Studio session.
Boundary markers: The skill documentation does not mention the use of delimiters or 'ignore' instructions when processing captured HTML or Studio context.
Capability inventory: The skill has extensive capabilities, including shell command execution (render, lambda deploy, cloudrun deploy), network operations (cloud render), and local file system modifications.
Sanitization: There is no explicit mention of sanitization or validation of the content retrieved from external URLs before it is used to scaffold a project.
[EXTERNAL_DOWNLOADS]: The skill performs several legitimate external operations associated with its functionality.
The browser ensure command (references/doctor-browser.md) downloads a specific version of the Chromium browser.
The init command (references/init-and-scaffold.md) downloads Whisper models for audio transcription.
The skills command (references/init-and-scaffold.md) fetches AI coding skill definitions from GitHub to update the local agent environment.
Standard npx invocations throughout the skill download and execute the hyperframes package from the npm registry.
[DATA_EXFILTRATION]: The skill includes features that share project data externally.
The publish command (references/preview-render.md) uploads project source code to a public URL for sharing.
The feedback --file-issue command (references/preview-render.md) uploads reproduction packets, including project files and logs, to public URLs for debugging. These operations are described as opt-in and consent-gated.
Anonymous telemetry (references/upgrade-info-misc.md) transmits usage counters and environment fingerprinting data (e.g., detecting the specific AI agent runtime via environment variables).
[COMMAND_EXECUTION]: The skill relies heavily on executing shell commands via npx for core functions such as rendering video, deploying infrastructure to AWS Lambda or Google Cloud Run, and managing project dependencies.

hyperframes-cli