sales-engineer
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources (such as customer RFP requirements) which is then interpolated into the agent's context for evaluation. This creates a surface for indirect prompt injection if the source files contain malicious instructions.
- Ingestion points: Data enters the agent context via the output of scripts/rfp_response_analyzer.py, scripts/competitive_matrix_builder.py, and scripts/poc_planner.py reading from assets/sample_rfp_data.json or user-provided files.
- Boundary markers: The workflow phases in SKILL.md do not include delimiters or specific instructions for the agent to ignore potential embedded prompts within the analyzed content.
- Capability inventory: The skill utilizes local Python execution and performs file system operations (read/write).
- Sanitization: The Python scripts use standard JSON parsing which validates structure but does not filter or sanitize the textual content for natural language injection patterns.
Audit Metadata