tao-analyze-gaps-visual-changenet
Pass
Audited by Gen Agent Trust Hub on Sep 22, 2026
Risk Level: SAFEDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DYNAMIC_EXECUTION]: Local shell validation hooks (
rca-artifacts-check.sh,rca-label-coverage.sh,rca-script-check.sh, andrca-section-check.sh) execute fixed Python logic via heredoc structures (python3 << 'PYEOF'). This code executes locally to validate data schemas, confusion matrices, and report section completion. - [INDIRECT_PROMPT_INJECTION]: The skill exposes data parsing surfaces by ingesting machine learning output files and configurations.
- Ingestion points: Automated ingestion occurs when reading
inference.csvandtrain.yamlfrom the experiment results directory. - Boundary markers: Absent. No specific prompt delimiters are used because the files consist of structured data elements rather than free-form natural language prompts.
- Capability inventory: The skill invokes
docker runto execute data services tasks and runs local shell scripts for workflow automation. - Sanitization: Absent. The CSV columns are processed directly using structured data parsers without textual string sanitization filters.
Audit Metadata