agent-optimiser
Pass
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill relies on local Python scripts (
scan_agents.pyandbump_version.py) to perform its core auditing and versioning functions. These are executed via shell commands constructed by the AI agent using user-provided paths or directories as arguments. - [INDIRECT_PROMPT_INJECTION]: The skill ingests and parses external agent definition files which are untrusted. This represents an indirect injection surface, as the AI agent must reason over the contents of these files to suggest optimizations. The skill mitigates this by using a deterministic scanner script and enforcing a 'report-first' workflow that requires human review before any changes are applied.
- [DATA_EXPOSURE]: The skill reads subagent definitions from standard locations such as
~/.claude/agents/. While these files contain instructions, they do not typically house sensitive credentials, and the skill performs no network operations to exfiltrate the analyzed content.
Audit Metadata