web-build
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests design specifications like LAYOUT SPEC and MOTION SPEC which act as untrusted ingestion points from external designers or other agent outputs.
- Ingestion points: Defined in the spec intake table of
SKILL.md. - Boundary markers: Absent; instructions specify that MOTION SPEC and other inputs should ride along 'verbatim' and 'untouched' into the kickoff prompt.
- Capability inventory: The skill generates 'kickoff prompts' for high-privilege agents like Claude Code that can scaffold projects, write code, and execute shell commands.
- Sanitization: There are no explicit instructions to validate, filter, or escape the content of the ingested specifications before they are interpolated into the downstream prompt.
Audit Metadata