agent-swarm-deployer

Pass

Audited by Gen Agent Trust Hub on Sep 15, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and orchestrate parallel processing of large sets of untrusted external data items, such as rows, datasets, and documents, which exposes a surface for indirect prompt injection if input data contains malicious instructions targeting the sub-agents.
  • Ingestion points: Bulk datasets are read and processed during the intake and distribution workflow outlined in SKILL.md and references/agent-brief.md.
  • Boundary markers: Structural delimiters and JSON schemas are specified for agent briefs, but there are no explicit prompt guidelines instructing sub-agents to ignore text-based commands embedded in the processed items.
  • Capability inventory: The execution environment grants access to powerful tools like Bash, Write, and Agent invocation, creating a functional impact surface if an indirect injection successfully controls sub-agent execution.
  • Sanitization: Content sanitization or parsing defenses are not defined prior to data interpolation into the sub-agent briefs.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 15, 2026, 01:01 PM
Security Audit — agent-trust-hub — agent-swarm-deployer