The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external web pages during the audit process via scripts/collect-evidence.mjs. There are no explicit boundary markers or instructions within the skill to disregard potential malicious instructions embedded in the audited surface's text or metadata. The capability inventory includes browser navigation and script execution (via page.evaluate) and file-system modification in REPAIR mode. No sanitization is performed on extracted DOM or CSS data before it enters the agent's context.
[DYNAMIC_EXECUTION]: The scripts/collect-evidence.mjs utility performs dynamic execution by running the inspectElement function within the browser's context using page.evaluate(). It also uses createRequire to dynamically load playwright dependencies at runtime.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of playwright or playwright-core for its evidence collection script. These are well-known, legitimate packages for browser automation and are correctly documented in the skill's compatibility metadata.

interface-forensics