The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exhibits a broad attack surface by ingesting and processing untrusted data from external sources to generate video content.
Ingestion points: The skill uses npx hyperframes capture <URL> to crawl websites, gh pr view to fetch GitHub Pull Request data, and parses Figma design files via URLs.
Capability inventory: The agent can write local files (BRIEF.md, project artifacts), execute shell commands (npx hyperframes), and perform network operations (npx hyperframes publish, heygen API calls).
Boundary markers: There are no explicit instructions to the LLM to ignore potentially malicious instructions embedded in the captured website text or GitHub PR descriptions.
Sanitization: While the skill mentions sanitizing SVGs during Figma imports, there is no evidence of sanitization for natural language text fetched from external URLs before it is used for script generation.
[COMMAND_EXECUTION]: The skill relies on executing several CLI tools to manage project state and perform rendering.
Evidence: Instructions in SKILL.md and references/skill-lifecycle.md direct the agent to run npx hyperframes@latest upgrade, npx hyperframes skills update, and npx skills add heygen-com/hyperframes.
Impact: These commands execute code from external packages and repositories to initialize the environment and install workflow-specific logic.
[EXTERNAL_DOWNLOADS]: The skill fetches code and configurations from remote sources during its lifecycle.
Evidence: The npx commands download packages from the NPM registry. The skills add command targets the heygen-com/hyperframes repository on GitHub to install lazily-loaded workflows.
[DATA_EXFILTRATION]: The skill contains functions that transmit project data to external services.
Evidence: The npx hyperframes publish command (documented in references/capability-menu.md) uploads the finished composition to a public URL. The heygen video create command sends data to a third-party AI video generation service.

hyperframes