The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to download and write component source code from an external repository into the local project. Specifically, it includes a command example that pipes remote JSON data into a Python interpreter using python3 -c to extract source code and save it to a .tsx file. While the Python command itself is a static one-liner for JSON parsing, this pattern creates a delivery mechanism for unverified remote code into the local environment where it is subsequently executed during project builds.\n- [EXTERNAL_DOWNLOADS]: The skill relies on fetching component registry items and source code from an external third-party domain. These downloads are used to populate the local project's codebase and define project dependencies, posing a risk of code injection if the external source is compromised.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes remote data which could potentially contain malicious instructions targeted at the agent.\n
Ingestion points: Remote registry JSON files are fetched from https://reactbits.dev via curl and CLI tools (defined in SKILL.md).\n
Boundary markers: The instructions in SKILL.md do not define any delimiters or validation steps for the content returned from the remote server.\n
Capability inventory: The skill possesses capabilities for writing files to the local file system (e.g., src/components/Magnet.tsx), executing shell commands, and installing software packages via npm or npx (described in SKILL.md).\n
Sanitization: There is no mentioned process for sanitizing or verifying the integrity of the downloaded content before it is processed or written to disk.\n- [COMMAND_EXECUTION]: The agent is directed to execute shell commands including npx shadcn, npx jsrepo, and python3. These commands utilize arguments, such as component names and URLs, which are derived from external sources or user requests.

react-bits