teams-setup
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted project data (source code, build files, and documentation) during Phase 1 analysis to generate specialized agent instructions, which could be exploited via malicious content in the processed files.\n
- Ingestion points: Phase 1: Project Analysis in SKILL.md reads files from the project root and .claude/ directory.\n
- Boundary markers: Uses structured templates and specific return contracts in references/agent-template.md, but lacks explicit ignore-instruction delimiters for external content.\n
- Capability inventory: Employs Bash, Write, Edit, and Task tools to manage files and spawn subagents.\n
- Sanitization: Provides JSON escaping and text truncation in the trace-ops.sh script.\n- [COMMAND_EXECUTION]: Orchestrates various local shell scripts (detect-mode.sh, verify-team.sh, trace-ops.sh, toggle-team.sh) to handle argument parsing, team verification, and activity logging.\n- [DYNAMIC_EXECUTION]: Dynamically constructs and writes agent markdown files in .claude/agents/ by populating a predefined template with project-specific facts harvested during analysis.
Audit Metadata