The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes HTML markup from arbitrary external URLs or local files to generate React components. This creates a vulnerability where malicious instructions embedded in the source HTML could influence the agent's output or behavior.\n
Ingestion points: The file temp/source.html is populated with content from user-provided URLs or local paths as described in SKILL.md Step 1.\n
Boundary markers: Absent. The instructions do not provide specific delimiters or guidance to ignore natural language instructions that might be present within the design HTML.\n
Capability inventory: The skill uses Bash (to run fetch scripts and npm), Write (to create component and configuration files), and Read (to analyze source code).\n
Sanitization: No sanitization, filtering, or validation is performed on the downloaded HTML content before the agent interprets it for code generation.\n- [COMMAND_EXECUTION]: The skill uses shell scripts to perform file operations and manage the development environment.\n
Evidence: SKILL.md contains instructions to execute bash scripts/fetch-stitch.sh and npm run dev.\n- [EXTERNAL_DOWNLOADS]: The skill fetches content from remote servers using command-line tools.\n
Evidence: The scripts/fetch-stitch.sh script uses curl to download HTML from URLs provided by the user or external APIs.\n- [DYNAMIC_EXECUTION]: The skill automates the installation of dependencies, which involves executing code from third-party registries.\n
Evidence: Step 7 in SKILL.md directs the agent to run npm install if the node_modules directory is missing.

stitch-nextjs-components