subagent
Pass
Audited by Gen Agent Trust Hub on May 19, 2026
Risk Level: SAFECOMMAND_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The protocol describes the use of
evo bashto execute shell commands within an experimental worktree or remote container, facilitating environment setup and direct code interaction.- [COMMAND_EXECUTION]: Subagents can register shell commands as validation gates usingevo gate add, which are executed during the benchmarking process to ensure stability.- [PROMPT_INJECTION]: The skill architecture handles untrusted data from task logs and source files, representing an indirect injection surface. - Ingestion points: Task traces from
evo tracesand code content from the worktree. - Boundary markers: The subagent brief includes specific 'Boundaries / anti-patterns' to restrict agent behavior.
- Capability inventory: Includes shell command execution (
evo bash), file write/edit operations, and benchmark execution (evo run). - Sanitization: The protocol relies on instructional constraints, specifically forbidding the hard-coding of secrets or the copying of .env files into source code.
Audit Metadata