choosing-swarm-patterns
Pass
Audited by Gen Agent Trust Hub on Aug 15, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill documentation describes multi-agent orchestration patterns that interpolate untrusted external data (e.g., {{task}}, {{request}}, {{question}}) into agent task prompts. This pattern establishes an attack surface for indirect prompt injection where malicious input could influence the behavior of downstream agents or tools.\n
- Ingestion points: Variable interpolation is used in several YAML examples in SKILL.md (e.g., handoff, cascade, and pipeline patterns).\n
- Boundary markers: The provided examples do not demonstrate the use of delimiters or isolation markers to separate untrusted data from instructions.\n
- Capability inventory: The agents described in the patterns have access to shell execution (npm test, git status) and communication tools (send_dm, post_message).\n
- Sanitization: The skill suggests logic-based routing tokens but does not provide examples of input sanitization or validation to prevent prompt injection.\n- [COMMAND_EXECUTION]: The workflow examples include the execution of shell commands, specifically 'git status --porcelain' for preflight checks and 'npm test' for verification steps. While contextually appropriate for a development workflow, these instructions promote the use of capabilities that interact directly with the host system.
Audit Metadata