doubt-driven-development
Pass
Audited by Gen Agent Trust Hub on Oct 24, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill provides detailed instructions for safely invoking external CLI tools like 'gemini' and 'codex'. It mandates several security checkpoints: verifying the tool exists via 'which', testing binary versions, confirming exact command syntax with the user, and requiring explicit human authorization before every run.
- [INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted 'artifacts' and 'contracts'. It includes proactive security guidance to mitigate indirect injection risks, specifically recommending the use of read-only sandboxes and passing input via stdin to ensure that any instructions embedded in the analyzed data cannot execute shell commands or modify the workspace.
- [PROMPT_INJECTION]: Uses an adversarial persona to find faults in decisions. While it instructs the agent to override default 'balanced' response shapes, this is restricted to the context of a logical review step and does not attempt to bypass core safety or ethical guardrails.
Audit Metadata