ns-agent-adaptation
Pass
Audited by Gen Agent Trust Hub on Sep 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill is strictly limited to architectural guidance and documentation. It explicitly prohibits the generation of implementation code, framework recommendations, or the execution of external commands. The frontmatter restricts the execution scope to the 'ns-harness' dependency, and the core behavior enforces safety-first design principles such as Human-In-The-Loop (HITL) for write actions and validation-before-execution for agents.
- [INDIRECT_PROMPT_INJECTION]: The skill identifies a theoretical attack surface as it ingests user input to generate the 'agent-design.md' specification document. ● Ingestion points: User-provided answers during the multi-step adaptation interview described in SKILL.md. ● Boundary markers: The skill uses a strict turn-based protocol ('one question per turn') and explicit 'lock before next' instructions to isolate user inputs. ● Capability inventory: The primary capability is writing to the 'docs/specs/agent-design.md' file; the skill lacks network exfiltration or shell execution capabilities. ● Sanitization: No specific sanitization or escaping mechanisms are described for the markdown output, though the risk is mitigated by the skill's restriction to conceptual design only.
Audit Metadata