The Agent Skills Directory

[METADATA_POISONING]: The README.md file includes installation instructions referencing the google-labs-code/stitch-skills namespace. This creates a misrepresentation of the skill's origin, as the identified author is abhicodebugger, potentially misleading users about the skill's authoritative source.
[INDIRECT_PROMPT_INJECTION]: The skill implements a baton-passing logic where the agent's next task is determined by reading and writing to local markdown files.
Ingestion points: The agent parses .stitch/next-prompt.md, .stitch/SITE.md, and .stitch/DESIGN.md to define its build logic and next steps.
Boundary markers: The instructions do not specify any delimiters or safety warnings to isolate these inputs from the agent's core instructions.
Capability inventory: The skill possesses Bash execution, file writing, and Stitch generation capabilities that can be steered by these ingested files.
Sanitization: There is no instruction to validate or sanitize the content of these files before they are used to influence the agent's behavior.
[COMMAND_EXECUTION]: The skill uses the Bash tool to execute npx serve for visual verification purposes. This allows the agent to run shell commands to manage a local web server.
[EXTERNAL_DOWNLOADS]: The skill is designed to download HTML and image assets from remote URLs provided by the Stitch tool output. These resources originate from infrastructure managed by Google.

stitch-loop