agent-squad
Pass
Audited by Gen Agent Trust Hub on Oct 24, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The orchestrator's design facilitates the flow of data across multiple agents, creating a vulnerability surface where one agent's output can influence another's behavior.
- Ingestion points: The Main Agent (SKILL.md) ingest outputs (Reports) from specialized agents such as Rex (Analyst) and Alex (Strategist) into its own context.
- Boundary markers: The "Briefing Packet" format used to pass context to downstream agents (e.g., Mason or Dep) uses simple text headers but lacks robust delimiters or "ignore embedded instructions" warnings to prevent the agent from obeying instructions hidden within the report data.
- Capability inventory: The squad includes agents with powerful capabilities, including code implementation (Mason), system architecture design (Aria), and CI/CD pipeline generation (Dep).
- Sanitization: The process relies on "compressed summaries" created by the Orchestrator, but there is no explicit sanitization logic to filter out potentially malicious instructions provided by a compromised or adversarial input at the requirements stage.
Audit Metadata