spec
Pass
Audited by Gen Agent Trust Hub on Oct 29, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill establishes a multi-turn workflow (Spec, Plan, Tickets) designed to separate system intent from implementation details. This structure reduces the risk of accidental complexity and promotes clear boundaries between high-level requirements and executable code.
- [SAFE]: The 'Goldfish' review swarm employs specialized subagent personas (e.g., Design Critic, Zoom Out, Implementation Readiness) to stress-test plans. These subagents are explicitly constrained by 'Core Mandates' that enforce a strict read-only protocol and stateless execution, preventing them from modifying the workspace or carrying unauthorized state between tasks.
- [SAFE]: External dependencies are restricted to vendored documentation and persona instructions derived from the Google Elephant-Goldfish Model (EGM), a well-known and trusted open-source project. Attribution and licensing are correctly documented.
- [SAFE]: The skill uses local CLI tools and internal agent delegation for orchestration. No remote code execution patterns, unauthorized network requests, or hardcoded credentials were found.
- [INDIRECT_PROMPT_INJECTION]: The skill inherently processes untrusted data (user requirements and codebase context) to generate specifications. However, this is the primary purpose of the skill, and it includes robust mitigations such as mandatory templates, 'Constitution Gates' for architectural validation, and a multi-agent review swarm that acts as a sanitization layer for generated plans.
Audit Metadata