doubt-driven-development
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill requires the agent to run shell commands to verify the availability of external tools (
which gemini,which codex) and their versions. It also involves executing these tools with user-supplied code artifacts. - [DYNAMIC_EXECUTION]: The skill provides instructions for generating and executing shell commands dynamically, including writing content to temporary files and piping it to external CLI tools to avoid shell injection risks.
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process external code artifacts which may contain malicious instructions intended to influence the reviewer's output. The documentation explicitly acknowledges this risk and provides mitigation strategies.
- Ingestion points: The
ARTIFACTandCONTRACTcontent blocks passed to the adversarial reviewer in Step 3. - Boundary markers: The skill uses clear labels and isolation of reasoning to separate untrusted input from instructions.
- Capability inventory: Includes the ability to execute shell commands and write to temporary files (e.g.,
/tmp/doubt-prompt.md). - Sanitization: The documentation prescribes using stdin/heredocs instead of shell arguments and utilizes read-only sandboxing for CLI tools to prevent the execution of injected commands within the artifacts.
Audit Metadata