The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function involves crawling external websites provided by the user to capture their content and structure for redesigning. This introduces a vulnerability to indirect prompt injection, where malicious instructions hidden on a crawled website could potentially influence or override the agent's behavior during the redesign pipeline.
Ingestion points: Untrusted website content is ingested via the extract phase and stored in stardust/current/pages/ for processing by subsequent phases.
Boundary markers: The skill instructions do not explicitly mandate the use of delimiters or 'ignore' instructions when the agent processes the crawled data.
Capability inventory: The skill has significant capabilities, including executing shell scripts (ledger.mjs, state.mjs), writing to the project root, and performing network operations.
Sanitization: The skill includes a deploy/sanitise.js script and a recursive validation loop that renders content in a headless browser to check for layout and console errors.
[EXTERNAL_DOWNLOADS]: The skill fetches version metadata from GitHub (raw.githubusercontent.com) to check for updates for its peer dependency. Since GitHub is a well-known and established service, this reference is documented as a safe functional requirement for dependency management.
[DATA_EXPOSURE]: The skill accesses local configuration paths (~/.claude/ and ~/.copilot/) to locate and verify the version of the required impeccable skill. While this involves reading metadata outside the immediate project workspace, it is performed to ensure dependency compatibility within the agent's operating environment.

stardust