brainstorming
Pass
Audited by Gen Agent Trust Hub on Oct 12, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONMETADATA_POISONING
Full Analysis
- Indirect Prompt Injection Surface: The skill instructions specify checking the 'current project state in working directory' during the Understanding phase. If the workspace contains untrusted files with adversarial content, those instructions could influence the agent's behavior during the brainstorming process. This is a common surface for development-oriented agents and is mitigated by the model's internal guardrails.
- Instructional Metadata: The metadata description field contains imperative language ('IMMEDIATELY USE THIS SKILL when creating or develop anything'). While intended to define when the skill is applicable, using the description field to mandate specific agent behavior is an instructional pattern that should be noted.
- Modular Skill Integration: The skill references external components located in '.claude/skills/using-git-worktrees' and '.claude/skills/writing-plans'. These represent functional dependencies on other skills within the environment to complete the design-to-implementation workflow.
Audit Metadata