local-delegation
Pass
Audited by Gen Agent Trust Hub on Apr 23, 2026
Risk Level: SAFE
Full Analysis
- [COMMAND_EXECUTION]: The skill provides Python scripts (scripts/agent_lm.py and scripts/query_lm.py) designed to interface with LM Studio's local API. These scripts are invoked via standard command-line calls and do not involve unauthorized or hidden execution of commands.
- [DATA_EXFILTRATION]: While the skill reads local files, it only transmits the data to a local API endpoint (defaulting to localhost:1234). The implementation includes strong directory traversal protections by resolving and validating paths against the intended working directory.
- [PROMPT_INJECTION]: The skill handles codebase data that could contain malicious instructions (Indirect Prompt Injection). This surface is mitigated by limiting the local model to read-only tools and explicitly instructing the main agent to maintain oversight and verify any findings.
- [SAFE]: The skill demonstrates high security standards for local tooling. It uses only Python standard libraries, prevents access to binary files, skips sensitive metadata folders like .git, and enforces budget limits on tool usage to prevent runaway processes.
Audit Metadata