Skill Threat Model

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

• Do not run unreviewed third-party scripts to learn what they do.
• Do not access secrets or private data during threat modeling.
• Do not approve broad write/network permissions without a clear workflow need and explicit user approval path.
• Treat skills with executable code as higher risk than instruction-only skills.

Deliverable Shape

For threat models, provide:

• Review scope
• Asset and capability inventory
• Trust boundaries
• Threat paths
• Risk matrix
• Required mitigations
• Approval recommendation

References

• Read references/skill-threat-model-checklist.md when threat modeling a skill, bundle, script, MCP preset, or install path.