The Agent Skills Directory

[PROMPT_INJECTION]: Indirect Prompt Injection Surface. * Ingestion points: Data entering through the ReviewBuilder class or the CLI in scripts/review_builder.py. * Boundary markers: None; the skill does not use specific delimiters to isolate untrusted data from instructions. * Capability inventory: The skill has the capability to write the generated HTML document to the local file system. * Sanitization: While data is escaped for JSON embedding, scripts/template.html uses .innerHTML to render the excerpt and source fields, creating a potential XSS vulnerability if input data is untrusted.

human-eval-review