hyperplan
Pass
Audited by Gen Agent Trust Hub on Sep 25, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill implements a multi-stage workflow that ingests untrusted user requests and interpolates them into system prompts for five different adversarial sub-agents (skeptic, validator, researcher, architect, creative).
- Ingestion points: The user's planning request is captured during Phase 0 and Phase 2 and propagated through the Round 1, Round 2, and Round 3 debate cycles.
- Boundary markers: The skill employs XML-like tags such as and to encapsulate the untrusted input, providing minimal defensive separation.
- Capability inventory: The orchestrator agent utilizes team management tools (team_create, team_send_message) and sub-agent dispatching via the task tool; the researcher and architect sub-agents are explicitly instructed to analyze and cite codebase files and line numbers, implying read access to the local filesystem.
- Sanitization: There is no evidence of input validation, sanitization, or escaping of the user's planning request before it is used to influence the behavior of the adversarial team members.
Audit Metadata