The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses a vulnerability surface for indirect prompt injection as it ingests and processes web content and text annotations from the browser during design sessions.\n
Ingestion points: The server script live-server.mjs accepts element context including outerHTML and textContent, as well as user-authored notes from the browser picking interface.\n
Boundary markers: The skill maintains structural integrity by using specific delimiters like impeccable-variants-start and impeccable-variants-end to isolate AI-generated code within the project's source files.\n
Capability inventory: The skill is capable of modifying the local project source through live-accept.mjs and live-wrap.mjs, which perform file system write operations based on user interaction.\n
Sanitization: Event data received from the browser is validated for correct structure in live-server.mjs before being passed to the agent context.\n- [COMMAND_EXECUTION]: The skill executes local management scripts and system commands to facilitate its design operations.\n
Evidence: The polling client live-poll.mjs executes internal scripts like live-accept.mjs using execFileSync to automate code updates.\n
Evidence: The utility is-generated.mjs performs shell calls to git check-ignore to identify build artifacts and avoid writing persistent changes to temporary files.

impeccable