skills/datadrivenconstruction/ddc_skills_for_ai_agents_in_construction/security-review-construction/Gen Agent Trust Hub
security-review-construction
Pass
Audited by Gen Agent Trust Hub on Sep 18, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted external data (CSV, Excel, JSON) provided by users, creating a surface for indirect prompt injection.\n
- Ingestion points: According to
instructions.md, the agent accepts project data, file paths, and parameters in CSV, Excel, and JSON formats.\n - Boundary markers: The instructions do not define clear delimiters or system-level directives to ignore instructions that may be embedded within the user-provided data files.\n
- Capability inventory: The skill manifest (
claw.json) grants the agentfilesystempermissions. Furthermore, the logic templates inSKILL.mdillustrate interactions with databases and cloud storage services like AWS S3.\n - Sanitization: While example code snippets in
SKILL.mddemonstrate validation checks (e.g., hash verification and GPS boundary checks), the core instructions ininstructions.mddo not explicitly mandate the sanitization of raw user data before it is parsed by the model.
Audit Metadata