project-discipline-guides

Pass

Audited by Gen Agent Trust Hub on Nov 10, 2026

Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The script scripts/generate_mirrors.ts executes external CLI tools (defaulting to agy, claude, and codex) using spawnSync. This is the primary mechanism for generating the "knowledge mirrors" used in the refactoring workflow. The calls to the gemini and claude CLIs include the --dangerously-skip-permissions flag, which is a tool-specific parameter used to bypass interactive prompts in automated scripts.
  • [DYNAMIC_EXECUTION]: The skill uses local markdown templates (mirror_prompt.md and intersection_prompt.md) to dynamically construct prompts for the CLI tools. These templates are populated with user-supplied discipline names and content generated by the AI models at runtime.
  • [INDIRECT_PROMPT_INJECTION]: The process for creating the "Unified Redundancy Mirror" involves interpolating output from three different AI models into a single prompt for an intersection task performed by Gemini.
  • Ingestion points: Model outputs are fetched via CLI tools and stored in markdown files within the mirrors/ directory before being read back into the intersection script.
  • Boundary markers: The intersection_prompt.md template utilizes "Source Mirror" headers and markdown horizontal rules to separate content from different model sources.
  • Capability inventory: The script performs local command execution via spawnSync and writes files to the project's local mirrors directory.
  • Sanitization: The script performs simple string replacement to inject model output into the intersection template; it does not explicitly sanitize or escape the content from the mirrors before interpolation.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 10, 2026, 09:41 AM