Skill Threat Model

Core Workflow

1. Identify what is being reviewed: skill, bundle, script, MCP preset, install path, or third-party source.
2. Inventory assets, data types, permissions, tools, scripts, dependencies, and host-specific install steps.
3. Map trust boundaries between user instructions, skill instructions, references, scripts, tools, network calls, and external content.
4. Identify threat paths: prompt injection, data exfiltration, privilege escalation, unsafe script execution, secrets exposure, and supply-chain risk.
5. Rate risks by likelihood, impact, evidence, and required mitigation.
6. Recommend approve, approve with changes, block, or defer.

Safety Rules