Codex Security: README vs what HN readers say
The README for openai/codex-security promises a chain: scan a whole repository, selected paths or Git changes, validate candidate findings, generate patches, and verify existing fixes. It also lists a TypeScript SDK, a GitHub Action, and exports to SARIF, JSON and CSV. Whether that adds up to more than a thin wrapper is the question the Hacker News comments circle.
On Hacker News, @AmazingTurtle, 2026-07-29 called it "Really just a (not so) fancy CLI wrapper about a prompt and a skill". @petesergeant, 2026-07-28 wrote "I don't think there's much to this other than it being a convenient CI wrapper around their existing models", then added in an edit that a linked part of the repo has "a little bit more meat". @varenc, 2026-07-28 raised a related question: "how much of the value here is providing the english Skill definitions that tell the LLM what to do", and said some of those approaches could be useful in other contexts. Read together, these three see a thin wrapper around prompts and OpenAI's existing models. In the excerpts here none of them says it fails to run; the open point is where the value sits.
The README's quick start is concrete: install the package globally, run cs login, then cs scan . with Node.js 22.13.0+ within 22.x, or 24.x or 26.x. @ryanto, 2026-07-29 tried a scan on a small open-source library and pasted the terminal log, which shows the CLI picking up stored Codex credentials and reaching a worker-delegation preflight. The excerpt available here is cut off before the outcome, so this review cannot say how that run ended.
GitHub dates the repository to 2026-07-13, with a latest push on 2026-11-13, and in about four months the project collected 12,659 stars, 989 forks and an Apache-2.0 license. The opinions above are more than three months old: all of them come from 2026-07-28 and 2026-07-29, from a single platform. The X posts collected for this repo are mostly announcements and summaries of the README.
Left open: varenc's question of how much a scan's quality owes to the SDK's prompts rather than the code around them.
Related repositories
Is codex-security worth your time?
ChatGPT, Claude and Perplexity can all read this page. Ask one of them what it makes of codex-security.
