product-manager-toolkit
Pass
Audited by Gen Agent Trust Hub on Sep 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (interview transcripts and feature lists) which are then summarized and synthesized by the agent. This creates a surface for indirect prompt injection if the source files contain malicious instructions disguised as user feedback or feature descriptions.
- Ingestion points: The scripts
scripts/customer_interview_analyzer.pyandscripts/rice_prioritizer.pyingest and process external.txtand.csvfiles. - Boundary markers: Absent. The instructions do not define clear boundaries or 'ignore' commands for the agent when interpreting extracted content from these files.
- Capability inventory: The skill allows the agent to read files, write summaries, and execute provided Python scripts. These scripts do not contain dangerous sinks (like
eval()or network calls) but their output is presented directly to the agent's context. - Sanitization: Absent. The scripts extract text matching specific keywords or regex patterns verbatim into analysis reports, which could carry malicious payloads into the agent's primary reasoning loop.
Audit Metadata