The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches a registry manifest and component files from the HyperFrames GitHub repository (github.com/heygen-com/hyperframes) using curl and the hyperframes CLI tool.
[REMOTE_CODE_EXECUTION]: The hyperframes add command downloads HTML and JavaScript snippets from a remote registry and installs them into the local project for execution during rendering or previewing.
[COMMAND_EXECUTION]: The instructions guide the agent to execute various subcommands of the hyperframes CLI and other Node.js-based tools via npx, including project publishing, rendering, and running local scripts like scripts/upload-docs-images.sh.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection as it ingests and processes external JSON data from the registry manifest, which contains descriptive fields that could be manipulated.
Ingestion points: External JSON manifests (registry.json, registry-item.json) fetched from raw.githubusercontent.com (SKILL.md, references/discovery.md).
Boundary markers: The instructions do not define specific delimiters or instructions for the agent to ignore potentially malicious content within the manifest data.
Capability inventory: The agent is authorized to perform file system writes (hyperframes add), execute local CLI tools, and perform network uploads (npx hyperframes publish).
Sanitization: There is no evidence of content sanitization or validation for the metadata retrieved from the remote registry.
Mitigation: Implement strict schema validation for registry manifests and include clear boundary markers or warnings when the agent processes descriptive fields from the registry.

hyperframes-registry