The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from a user-provided URL in Phase 1 (Extract). The content is extracted using Playwright and then used to drive design decisions and variant generation.
Ingestion points: Phase 1 delegates to stardust:extract which crawls a remote URL and saves the content to local JSON files (pages/home.json, _brand-extraction.json).
Boundary markers: The skill does not define specific delimiters or instructions to ignore potential commands embedded in the scraped website content before it is processed by the agent.
Capability inventory: The skill performs file writing (uplift-improvements.md, uplift-questions.md, direction.md), file reading, and delegates to other skills that execute shell commands and render HTML prototypes.
Sanitization: There is no explicit mention of sanitizing or filtering the content extracted from the external URL before it influences the agent's logic in subsequent phases.
[COMMAND_EXECUTION]: The skill uses local shell commands for workflow management and output display.
Phase 5 (Prototype) describes using git worktree or directory copies to create isolated workspaces for parallel rendering of design variants.
Phase 6 (Open) explicitly requires using the open command to view the generated prototype files in a browser via the preview service worker.
Phase 5 also mentions the use of $impeccable craft for rendering.

uplift