The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted HTML data from external URLs or local files and uses it as the structural and visual reference for generating React Native application code. This creates a surface where malicious instructions embedded in the source markup could influence the AI agent to produce insecure, backdoored, or otherwise compromised code.
Ingestion points: The file temp/source.html, which is populated by the fetch-stitch.sh script or direct file copy, serves as the primary data source for the agent.
Boundary markers: The instructions lack explicit delimiters or "ignore instructions" warnings to prevent the agent from following natural language commands found within the source HTML.
Capability inventory: The agent has Write permissions to create and modify project files, Read permissions to access the ingested data, and Bash execution capabilities via the downloader script.
Sanitization: There is no evidence of sanitization or filtering of the source content before it is processed for code generation.
[EXTERNAL_DOWNLOADS]: The skill utilizes a custom shell script (scripts/fetch-stitch.sh) that wraps curl to download content from arbitrary URLs. While designed to fetch UI designs, this allows the agent to perform network requests to non-whitelisted or sensitive internal endpoints if directed to do so by a user or malicious design metadata.
[COMMAND_EXECUTION]: The skill relies on executing a local Bash script to perform its primary downloading task. This execution of local shell scripts is a high-privilege action within the agent environment.

stitch-react-native-components