mdtask-do
Pass
Audited by Gen Agent Trust Hub on Aug 23, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The workflow requires the agent to define and execute "behavior checks," which are described as arbitrary shell commands, CLI flows, or API calls designed to verify task completion. This grants the agent a broad capability to execute code in the local environment based on its own planning.
- [PROMPT_INJECTION]: The skill processes task instructions from project Markdown files via
mdtask view. This creates an indirect prompt injection surface where malicious instructions in a task could compromise the agent's workflow. - Ingestion points: Project Markdown files and output from the
mdtaskCLI. - Boundary markers: The skill specifies logical task boundaries and requires a checklist of steps, but it lacks technical delimiters or "ignore embedded instructions" warnings for the ingested task content.
- Capability inventory: The agent can modify files, execute arbitrary shell commands for behavior checks, and commit code to the repository.
- Sanitization: The instructions do not define any sanitization, validation, or escaping of the task data before it is interpreted by the agent.
Audit Metadata