plan-with-team
Pass
Audited by Gen Agent Trust Hub on Oct 31, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill implements a multi-agent orchestration workflow where the primary agent spawns specialized sub-agents based on user-provided requirements. This architecture presents an indirect prompt injection surface where instructions from the initial requirements phase or broadcasts between agents could be used to influence sub-agent behavior.\n
- Evidence: Phase 2 in
SKILL.mdspawns agents using prompts that incorporate requirements gathered from the user in Phase 1.\n - Evidence: The collaborative workflow in Phases 3 and 4 involves agents broadcasting proposals to each other, creating a path for instruction leakage across agent contexts.\n
- Mitigation: The skill includes specific behavioral constraints labeled as 'CRITICAL RULES' (e.g., 'No code is written during planning') and requires the use of structured output templates defined in
resources/plan-template.md. The inclusion of 'The Skeptic' role provides a dedicated layer for identifying security and architectural flaws before consolidation.\n- [SAFE]: The skill provides instructions for the user to manually enable experimental platform features by editing.claude/settings.json. This is a standard procedure for configuring the agent environment and does not involve automated or unauthorized system modifications.
Audit Metadata