swarm
Pass
Audited by Gen Agent Trust Hub on Sep 21, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADSPRIVILEGE_ESCALATION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill possesses a significant attack surface by ingesting and processing external data like user tasks, codebases, and PR content across multiple files. It establishes a multi-stage pipeline where data processed by one agent informs the context of others. While the skill defines 'Output Contracts' as coordination delimiters, it utilizes powerful tools for agent spawning, filesystem modification, and command execution. Security relies on the lead agent's review of teammate outputs before consolidation.
- [COMMAND_EXECUTION]: Implementing roles are authorized to use shell execution tools (Bash) to perform development and validation tasks.
- [DYNAMIC_EXECUTION]: The documentation outlines a dynamic workflow mechanism using JavaScript for runtime orchestration of agent behaviors.
- [EXTERNAL_DOWNLOADS]: The skill includes references to official platform documentation at code.claude.com for technical guidance.
- [PRIVILEGE_ESCALATION]: The skill describes the inheritance of high-privilege permission modes (such as bypassPermissions) when creating sub-agents, enabling automated tool execution without user confirmation.
Audit Metadata