merge
Pass
Audited by Gen Agent Trust Hub on Sep 3, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes multiple shell commands including
git checkout,git merge,git tag, andgit branch -Dusing interpolated variables. The use of the--no-ffflag for merging is a noted safety practice for maintaining clean history. - [INDIRECT_PROMPT_INJECTION]: The skill contains an attack surface for command injection because it interpolates data from session states and command arguments into shell scripts.
- Ingestion points: Data enters the process via command line arguments (e.g.,
session-id,agent-id) and from the output of previous/hub:evalsessions (e.g., winner identification). - Boundary markers: Absent. The skill does not use delimiters or instructions to prevent the agent from obeying embedded instructions within the task description or session metadata.
- Capability inventory: The skill has capabilities to perform branch management, tag creation, branch deletion, file writes to the
.agenthubdirectory, and execution of Python scripts. - Sanitization: No explicit sanitization or validation of the
{winner},{session-id}, or{base_branch}variables is visible before they are passed to the shell. - [DYNAMIC_EXECUTION]: The skill executes a local Python script
session_manager.pylocated within the skill's own directory structure to perform session cleanup and state updates.
Audit Metadata