The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The README.md file contains installation instructions that use a curl | sh pattern: curl -sL skil.sh | sh -s -- voxel51/fiftyone-skills. This method is inherently insecure as it executes a remote script directly in the shell without prior inspection, allowing for arbitrary code execution.
[EXTERNAL_DOWNLOADS]: The skill initiates network requests to external domains to function and install:
It downloads a shell script from skil.sh during the installation process.
It uses the WebFetch tool to retrieve a Markdown reference file from voodo.dev.fiftyone.ai during runtime to inform code generation.
[COMMAND_EXECUTION]: The skill provides instructions that lead to the execution of shell commands through the installation script, which can potentially perform system-level changes.
[INDIRECT_PROMPT_INJECTION]: The skill relies on fetching an external documentation file (voodo-llm-reference.md) to guide the AI's generation of React code. This creates a surface where the agent's behavior could be influenced by malicious instructions embedded in the external content.
Ingestion points: WebFetch call in SKILL.md for https://voodo.dev.fiftyone.ai/voodo-llm-reference.md.
Boundary markers: Absent; there are no instructions to the agent to ignore or delimit embedded instructions within the fetched data.
Capability inventory: The skill is capable of generating React components, defining application layout patterns, and specifying library dependencies.
Sanitization: None; the skill does not mention any validation or filtering of the fetched content before processing it.

fiftyone-voodo-design