The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill reads project configuration and task files to construct prompts for page generation, creating a surface for indirect instructions.
Ingestion points: The agent reads content from next-prompt.md, SITE.md, and DESIGN.md in SKILL.md (Steps 1 & 2).
Capability inventory: The skill utilizes Bash for command execution, Write for file modification, and chrome* tools for browser interaction as listed in the frontmatter of SKILL.md.
Boundary markers: Instructions do not specify the use of delimiters or explicit boundary markers when interpolating file content into tool prompts.
Sanitization: No sanitization or validation of the ingested content is defined before it is passed to the generation tools.
[COMMAND_EXECUTION]: The skill invokes the Bash tool to execute shell commands during the visual verification process.
Evidence: SKILL.md (Step 4.5) instructs the agent to run npx serve site/public to host the generated site locally.
[EXTERNAL_DOWNLOADS]: The skill downloads generated assets from remote sources as part of the page creation workflow.
Evidence: SKILL.md (Step 3.4) directs the agent to download content from htmlCode.downloadUrl and screenshot.downloadUrl provided by the Stitch MCP tools.

stitch-loop