doubt-driven-development

Pass

Audited by Gen Agent Trust Hub on Sep 8, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The skill utilizes external CLI tools (e.g., Gemini or Codex) for cross-model reviews. It provides clear safety templates that mitigate injection risks by instructing the agent to use temporary files and stdin pipes rather than direct shell interpolation of untrusted strings.
  • [INDIRECT_PROMPT_INJECTION]: The skill is specifically designed to analyze untrusted artifacts (code or architectural decisions) which may contain malicious instructions. It enforces several layers of defense against this threat.
  • Ingestion points: The Step 3 (DOUBT) process in SKILL.md ingests an ARTIFACT which can be user-provided code or documentation.
  • Boundary markers: The skill uses explicit ARTIFACT: and CONTRACT: labels in generated prompts to provide context to sub-agents.
  • Capability inventory: The skill has the ability to run shell commands and spawn sub-agent personas.
  • Sanitization: Instructions mandate the use of read-only sandboxes and stdin piping to ensure that the analysis of the artifact cannot lead to unauthorized filesystem writes or command execution.
  • [EXTERNAL_DOWNLOADS]: Reference materials (references/accessibility-checklist.md, references/performance-checklist.md) recommend industry-standard development and auditing tools (e.g., axe-core, lighthouse, pa11y) to be run via npx. These tools are used for their intended purpose in an engineering workflow.
  • [SAFE]: The skill adheres to endorsed orchestration patterns by keeping the user in the loop for all external tool invocations and avoiding persona-to-persona invocation anti-patterns. It effectively guides the AI to maintain a safe, adversarial posture during reviews.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 8, 2026, 06:55 AM
Security Audit — agent-trust-hub — doubt-driven-development