The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructions and reference documents do not contain any language attempting to override agent behavior, bypass safety guardrails, or extract system prompts. All content is strictly relevant to user interface design and implementation.- [DATA_EXPOSURE_AND_EXFILTRATION]: No sensitive file paths, hardcoded secrets, or network exfiltration patterns were found. The skill operates purely on visual geometry and standard browser APIs like getBoundingClientRect.- [OBFUSCATION]: There is no evidence of hidden instructions, Base64-encoded strings, or characters intended to deceive static analysis tools. All text is provided in clear markdown and documented pseudocode.- [UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The skill does not instruct the agent to install third-party packages or download remote scripts. It relies on standard web platform APIs and documented mathematical algorithms.- [INDIRECT_PROMPT_INJECTION]: The skill's ingestion points are limited to reading DOM geometry. It employs a strong architectural boundary (two-layer model) that ensures untrusted visual data does not influence the agent's logic or the underlying semantic controls.

liquid-interface