ln-34-benchmark-comparator
Pass
Audited by Gen Agent Trust Hub on Jul 20, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands and use runner tools to measure the performance and correctness of various implementations.
- Evidence: The 'Tool Routing' section defines using 'The same shell runner and wrapper for every candidate' to capture 'commands, exit status, stdout, stderr, timing, and artifacts'.
- Evidence: The checklist (Step 2 and 3) requires building a harness to 'Run candidates in a balanced or randomized order' and capturing execution state.
- [PROMPT_INJECTION]: The skill processes external repository data and implementation outputs, which creates a surface for indirect prompt injection if those sources contain adversarial instructions.
- Ingestion points: 'SKILL.md' (Section 1 and 3) directs the agent to 'Read repository instructions' and analyze 'candidate-specific artifacts', 'logs', and 'outputs'.
- Boundary markers: The skill does not explicitly define delimiters to isolate external text data from its operational instructions.
- Capability inventory: The agent has the capability to run shell commands, create/delete Git worktrees, manage system processes via IDs, and write files.
- Sanitization: The skill emphasizes environment isolation (clean-room setup) but does not specifically address the sanitization of incoming text-based instructions.
Audit Metadata