web-typography
Pass
Audited by Gen Agent Trust Hub on Oct 22, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface because its primary function is to analyze untrusted user content (HTML, CSS, and Markdown files) to provide typographic recommendations.
- Ingestion points: The agent is instructed in
SKILL.mdunder the "Scripts" section to runscripts/check_text.pyandscripts/audit_css.pyon user-specified files or folders, bringing the content of those files into the agent's reasoning process. - Boundary markers:
SKILL.mdincludes explicit instructions stating that script outputs "produce candidates, not verdicts" and that the agent must "Confirm each one in context before changing anything." - Capability inventory: The agent has access to local analysis scripts, browser-based JavaScript execution for layout measurement (
scripts/inspect_lines.js), and the ability to modify project files based on the scan results. - Sanitization: There is no automated sanitization of instructions that might be hidden in the audited text; the skill relies on the agent following guidelines to manually verify findings.
Audit Metadata