The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill performs forensic audits of external websites provided by the user. The instructions emphasize an 'Anti-Flattening Doctrine' and 'verbatim' transplantation of site artifacts into the final replication prompt. This creates a surface for indirect prompt injection, as malicious instructions embedded in a target website's text or HTML could be carried over into the final output prompt.
Ingestion points: Target website content analyzed in references/phase-1-forensic-audit.md using DOM queries and visual inspection.
Boundary markers: The output templates lack explicit delimiters or instructions to warn downstream agents about potentially malicious content from the reference site.
Capability inventory: The skill executes JavaScript within a browser context for auditing and writes markdown files to a local plans/ directory.
Sanitization: There is no evidence of filtering or sanitization performed on the ingested content to remove potential prompt injection vectors.
[COMMAND_EXECUTION]: The skill provides various JavaScript snippets in references/phase-1-forensic-audit.md (Steps 1.1, 1.2, 1.3, 1.8). These are designed to be executed within a browser's developer console or by a browser-capable agent tool to extract design tokens, DOM structure, and technology stack information. These snippets are functional for the skill's stated purpose and do not target local system resources.

ui-cloner