juicebox-core-workflow-a
Pass
Audited by Gen Agent Trust Hub on Mar 12, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: Vulnerability surface for indirect prompt injection identified in the search query construction logic.
- Ingestion points: The
CandidateSearchinterface intypes/search.tsprocesses raw user-provided strings for fields likerole,skills, andlocation. - Boundary markers: The
buildSearchQueryfunction intypes/search.tsuses direct string interpolation. While some parameters are enclosed in double quotes, there is no escaping mechanism for these quotes or other special characters. - Capability inventory: According to the
SKILL.mdfrontmatter, the skill is granted extensive permissions includingBash(npm:*),Bash(pip:*),Write, andEdit. These could be leveraged if an attacker-controlled profile in the search results contains malicious instructions that the agent follows. - Sanitization: No input validation or output sanitization is implemented for the data exchanged with the external Juicebox API.
Audit Metadata