mreq-close
Warn
Audited by Gen Agent Trust Hub on Nov 19, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill is designed to parse and execute arbitrary shell commands found within project files and task documents during its verification phase.
- Evidence: Step 3.2 ("Run") explicitly instructs the agent to run "test commands," "repro" scripts, and "Behavior Pins" exactly as they are written in the
.mreq/documents, the projectREADME.md, or other agent instruction files. - [REMOTE_CODE_EXECUTION]: Because the skill executes commands extracted from data files without a predefined whitelist or sanitization, it acts as a vector for arbitrary code execution if those files are malicious.
- Evidence: The instructions in Step 3.2 mandate running commands found in the input data "exactly as written" using
BashorPowerShelltools. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted documents and follows their internal instructions (goals and tasks), which creates an attack surface for indirect prompt injection.
- Ingestion points: Documents located in the
.mreq/directory and project-level files likeREADME.md. - Boundary markers: The skill does not implement specific boundary markers or instructions to ignore embedded malicious directives within the processed documents.
- Capability inventory: The skill has access to shell execution (
Bash/PowerShell), tool chaining (Skill(mreq-apply)), and file deletion capabilities. - Sanitization: There is no evidence of validation or escaping for the natural language instructions or shell commands extracted from the ingested documents.
Audit Metadata