The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user input for character descriptions and product contexts, which are then interpolated into prompts sent to the external Higgsfield.ai service.
Ingestion points: User-provided character descriptions, product context, and video style preferences are used to populate templates in Phase 2.
Boundary markers: No explicit boundary markers or "ignore instructions" wrappers are implemented to prevent user-supplied text from hijacking the prompt flow.
Capability inventory: The skill utilizes Playwright tools for full browser interaction, including typing, clicking, and DOM manipulation.
Sanitization: There is no evidence of sanitization or filtering of the user-provided strings before they are sent to the target website.
[DYNAMIC_EXECUTION]: The skill uses browser-side script execution to interact with the Lexical rich text editor on the target site.
Evidence: browser_evaluate is used in Step 2.6 to focus and click the editor element via JavaScript: () => { const el = document.querySelector('[data-lexical-editor]'); el.focus(); el.click(); }.
[COMMAND_EXECUTION]: The skill performs a high volume of automated browser commands through the Playwright MCP, including navigation, key presses, and element interactions, which are necessary for its stated purpose of pipeline automation.

ugc-video-auto