model-evaluation-benchmark
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from benchmark results and trace logs which are subsequently analyzed by a reviewer subagent.
- Ingestion points: Reads result.json files and trace logs from the ~/.amplihack/.claude/runtime/benchmarks/suite_v3/ directory.
- Boundary markers: The instructions lack explicit delimiters or warnings to ignore instructions embedded within the logs or result files.
- Capability inventory: The skill has the capability to execute local Python scripts (run_benchmarks.py), interact with GitHub via the gh CLI (create/close issues, create releases), and perform git operations.
- Sanitization: There is no evidence of sanitization or validation of the ingested log content before it is processed by the subagent.
- [COMMAND_EXECUTION]: The skill executes shell commands and local scripts to perform benchmarks and cleanup.
- Evidence: Execution of python run_benchmarks.py, gh pr close, gh issue close, and git worktree remove commands.
Audit Metadata