doubt-driven-development
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to utilize external CLI tools (such as
geminiorcodex) for cross-model verification. The workflow includes mandatory safety protocols: verifying tool existence viawhich, performing functionality tests, and obtaining explicit user confirmation for every command execution. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted data ('artifacts') which may contain malicious instructions. The documentation explicitly identifies the risk of prompt injection and recommends using read-only sandboxes to isolate the reviewer sub-agents.
- Ingestion points: External code and contracts are provided to reviewers in
SKILL.md. - Boundary markers: The skill suggests isolating content via temp files and stdin to prevent it from being interpreted by the shell.
- Capability inventory: Uses agent orchestration (network) and shell command execution for external model analysis.
- Sanitization: Mandates the use of stdin/heredocs to avoid argument-based shell injection vulnerabilities.
- [DYNAMIC_EXECUTION]: The process involves dynamically assembling prompts and artifact data to be passed into external processes, though it emphasizes secure methods like stdin over unsafe argument interpolation.
Audit Metadata