Postman silently installed itself into Codex and Claude Code
An update quietly added startup hooks to two coding agents and approved its own trust record, no consent screen involved. Postman says it is reverting the change.
Opening Postman desktop 12.33.4 on 24 October installed a plugin into both OpenAI’s Codex and Anthropic’s Claude Code without asking. A developer filed the report as issue 14370 on 25 October, and it has since picked up over 400 reactions.
The installer added a startup hook telling each agent to load a postman:api-engineer skill before handling any API-related task, dropped a Postman CLI into ~/.local/bin, and in Codex’s case wrote the hook’s own trust hash into the config, the step that is supposed to require a user’s manual approval before an agent trusts a new hook. None of it showed a prompt first.
Why it matters: if you run Postman alongside Codex or Claude Code, check your agent configs for plugins or hooks you did not install yourself. A desktop app approving its own hook trust record is exactly the kind of silent privilege escalation that agent sandboxes are meant to catch.
The caveat: a Postman team member has said the change is being reverted, but there is no stated rollback for configs it already modified.