ln-811-algorithm-optimizer

Warn

Audited by Gen Agent Trust Hub on Mar 10, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill invokes language-specific build and test utilities to perform benchmarks on both modified existing code and newly generated benchmark scripts.
  • Evidence: Found in references/benchmark_generation.md which references tools like go test, pytest, vitest, cargo bench, and dotnet run.
  • [REMOTE_CODE_EXECUTION]: The workflow retrieves optimization strategies from external web searches and integrates them into the codebase, followed by immediate execution for performance verification.
  • Evidence: SKILL.md Phase 2 (Research Strategy) uses WebSearch to gather optimization techniques applied in Phase 4 (Optimize Loop).
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it processes untrusted research data and local source code to generate executable logic.
  • Ingestion points: target_file (local source) and WebSearch results (Phase 2).
  • Boundary markers: Absent. No specific instructions are provided to the agent to disregard instructions embedded within the researched data.
  • Capability inventory: The agent can modify local files, execute shell commands, and perform Git operations (git add, git commit).
  • Sanitization: Verification is limited to functional testing; it does not check for malicious side effects or unauthorized network activity in the suggested code changes.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 10, 2026, 03:47 PM