security-threat-model

TL;DR

An application-security threat model from OpenAI's catalog of Codex skills. The agent builds a model of the system from the repo: components, data stores, integrations, how it runs and its entry points, keeping runtime apart from CI, build tooling and tests, and claiming nothing it can't point to. It lists trust boundaries as concrete edges with their protocol, auth, encryption, validation and rate limits, the assets that drive risk, and realistic attacker capabilities, including what an attacker can't do, so severity isn't inflated. Threats are written as abuse paths tied to assets and ranked low to critical by likelihood times impact, adjusted for existing controls. Before the final report it pauses to confirm assumptions with you; whatever stays unanswered is kept as an explicit assumption and makes recommendations conditional. Mitigations separate existing controls, with evidence, from recommended ones tied to specific locations. A quality check confirms every entry point and trust boundary is covered. The report follows a fixed template: executive summary, scope and assumptions, system model, assets, attacker model, entry points, top abuse paths, the threat table, criticality calibration and focus paths. The openai/skills repository is marked deprecated; its README points to openai/plugins for current Codex skills.

Summarizes the skill at 49f948f.