diverse-plan
Pass
Audited by Gen Agent Trust Hub on Sep 8, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to execute a local discovery script (
discover_runners.py) and several CLI-based runner tools (claude-runner,codex-runner,gemini-runner,grok-runner,pi-runner) to interface with various AI providers. These commands are executed via the shell to coordinate the multi-model planning workflow. - [INDIRECT_PROMPT_INJECTION]: The skill processes external data including project source code,
CONTEXT.mdfiles, and Architectural Decision Records (ADRs). These inputs are interpolated into prompts for multiple external models without explicit sanitization or strict boundary markers, creating a surface where instructions embedded in code comments or documentation could influence the agent's planning logic. - Ingestion points: Reads target code files,
CONTEXT.md, and ADRs as specified in the 'Frame' step. - Boundary markers: Not explicitly defined in the prompting instructions.
- Capability inventory: Shell execution of discovery scripts and model runners.
- Sanitization: No specific sanitization or validation of ingested file content is documented before it is passed to LLM models.
Audit Metadata