The Agent Skills Directory

[PROMPT_INJECTION]: The skill implements a 'baton-passing' loop where the agent reads its next task from a local file (next-prompt.md). This creates an indirect prompt injection surface.
Ingestion points: The next-prompt.md file is parsed at the start of each iteration to determine the next page to build.
Boundary markers: The file structure uses YAML frontmatter but lacks delimiters or instructions to the LLM to ignore potentially malicious instructions embedded in the task description.
Capability inventory: The skill possesses Bash access, file Write capabilities for project integration, and access to the Stitch generation tool.
Sanitization: There is no evidence of sanitization or validation of the task content before it is processed.
[COMMAND_EXECUTION]: The skill utilizes the Bash tool to start a local web server using npx serve. This involves executing arbitrary commands and dynamically loading packages from the npm registry.
[EXTERNAL_DOWNLOADS]: The agent is instructed to download HTML and screenshot assets from URLs provided by the Stitch tool's output (htmlCode.downloadUrl and screenshot.downloadUrl) and save them to the local queue/ directory.

stitch-loop