oma-dev-workflow
Fail
Audited by Gen Agent Trust Hub on Oct 6, 2026
Risk Level: HIGHREMOTE_CODE_EXECUTIONPERSISTENCEDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: Downloads and executes the installation script for the mise tool using
curl https://mise.run | sh. While this is an official installation method, it involves executing remote code in a shell. - [PERSISTENCE]: Configures persistent command execution by appending
evalstatements to shell profile files such as~/.zshrcto activate environment tools on startup. - [PERSISTENCE]: Automatically creates and writes executable scripts to the
.git/hooksdirectory, which triggers automated task execution during local Git operations (e.g.,commit-msg,pre-commit). - [DYNAMIC_EXECUTION]: Employs the
evalcommand to dynamically execute shell environments and activation scripts generated by the mise tool at runtime. - [COMMAND_EXECUTION]: Invokes multiple CLI utilities including
docker compose,git,gh(GitHub CLI), and package managers likeuvandbunto perform automated tasks. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from configuration files and system logs to determine automated actions.
- Ingestion points:
mise.tomltask definitions, CI/CD pipeline logs, and build artifacts found inSKILL.mdandresources/troubleshooting.md. - Boundary markers: Absent; instructions do not provide delimiters or safety warnings for the agent when reading potentially attacker-controlled task definitions.
- Capability inventory: Includes shell command execution via
mise run, file system modification, and interaction with GitHub repositories via CLI tools. - Sanitization: No input validation or sanitization is performed on external task definitions or log data before they are used to drive agent actions.
Recommendations
- HIGH: Downloads and executes remote code from: https://mise.run - DO NOT USE without thorough review
Audit Metadata