vibe-brainstorm
Pass
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes underspecified creative requests and outputs from delegated model units, establishing a surface where untrusted data could influence agent behavior.
- Ingestion points: User-provided creative requests and specialist context delegations defined in SKILL.md.
- Boundary markers: Implements an 'Effect And Write Boundaries' policy and requires explicit user confirmation for any changes to requested behavior or scope.
- Capability inventory: The skill is capable of writing documentation (ADRs and findings reports) and executing shell commands (e.g., git, sed, rm) when in 'state-changing' mode, as detailed in references/durable-records.md.
- Sanitization: Relies on human-in-the-loop confirmation and the selection of high-capability model tiers for synthesis and review tasks to manage potential injection risks.
- [SAFE]: The skill explicitly defines and enforces operation boundaries (read-only, artifact-only, and state-changing), ensuring the agent does not perform irreversible or outward-facing actions without specific consent.
Audit Metadata