brainstorming
Pass
Audited by Gen Agent Trust Hub on Mar 7, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: In
SKILL.md, the skill ingests user-provided topic strings and local codebase content which are then interpolated into sub-agent prompts. This creates a surface for indirect prompt injection as external content is used to influence agent behavior without sanitization. \n - Ingestion points: User arguments (
$ARGUMENTS) and project files (e.g.,README.md) read viaGrepandGlobduring Phase 0 and Phase 1 discovery. \n - Boundary markers: Not used in sub-agent prompt definitions. \n
- Capability inventory: File reading, task tracking, and sub-agent orchestration. \n
- Sanitization: No filtering or escaping is applied to the ingested data. \n- [COMMAND_EXECUTION]: Documentation templates in
scripts/create-design-doc.mdandscripts/design-doc-template.mdutilize hardcoded shell commands likegit configanddateto auto-populate metadata. These executions are limited to standard system utilities for documentation purposes.
Audit Metadata