ln-814-optimization-executor

Pass

Audited by Gen Agent Trust Hub on Mar 15, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill retrieves and executes arbitrary shell commands from the test_command and e2e_test_command fields within the .optimization/{slug}/context.md configuration file.
  • Evidence: Phase 1 (Establish Baseline) and Phase 2 (Strike-First Execution) utilize these dynamically loaded strings as safety gates and benchmarking tools.
  • [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it performs high-impact actions—such as modifying source code and executing shell commands—based on instructions ingested from an external configuration file without validation.
  • Ingestion points: .optimization/{slug}/context.md (specifically the hypotheses, test_command, and e2e_test_command fields).
  • Boundary markers: No markers or 'ignore' instructions are used to separate the configuration data from the agent's logic.
  • Capability inventory: The skill has the ability to modify multiple files across the repository (APPLY all uncontested hypotheses) and execute subprocesses (test_command).
  • Sanitization: There is no evidence of command sanitization, path validation, or instruction filtering for the content loaded from the context file.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 15, 2026, 09:02 PM