A developer's bug report that Postman's desktop app quietly installed plugins into two AI coding agents, Anthropic's Claude Code and OpenAI's Codex, has become one of the most discussed AI threads on Hacker News this week. The thread had more than 1,400 points and nearly 500 comments by 5 November 2026. Postman had not publicly responded in the coverage reviewed.
What the report says
A developer posting as "indicava" filed GitHub issue #14370 on 25 October after Codex began showing authentication errors and repeatedly loading Postman tools without being asked. According to the report, as covered by The New Stack and daily.dev, opening Postman desktop version 12.33.4 on macOS triggered an installation event on 24 October, logged as "auto-existing-upgrade." The log shows plugins installed for both Claude Code and Codex.
The reported changes to Codex's configuration file include:
- an enabled Postman plugin and a new plugin marketplace entry;
- a session-start hook that loads a Postman "skill" every time Codex starts;
- a remote connection to Postman's MCP server, which failed to authenticate and caused the errors.
The self-approval problem
The part that alarmed developers most is that Codex normally asks the user to approve a new hook before it can run, to stop third parties from silently running code inside the agent. According to the issue, Postman's installer code calls a function that approves its own hook's trust record, skipping that manual review. The report also says the standard uninstall does not fully undo the changes to agent settings, saved connections and installed command-line tools.
What developers are asking for
The thread, which had gathered 99 reactions on GitHub, asks Postman to stop installing agent plugins for existing users without an explicit opt-in, to remove the installer-side self-approval, and to limit the startup instruction to cases where a user asks for it. Reaction on Hacker News was mostly sceptical, with some users saying their organisations were reviewing their use of Postman and others pointing to alternatives such as Bruno, Insomnia and Yaak, according to a summary by daily.dev. Postman's tool may also be useful to developers who choose to install the plugin. The complaint is about consent, not about the plugin existing.
How to check your setup
Developers who use Postman alongside coding agents can audit their own machines. Coverage suggests:
- Look in Codex's
~/.codex/config.tomland Claude Code's~/.claude/settings.jsonfor plugins, marketplace entries, startup hooks or MCP connections you did not add. - Remove anything unfamiliar and check for leftover command-line tools.
- Consider requiring explicit approval for new hooks and plugins in your team's agent settings.
Why it matters
As coding agents gain access to files, credentials and commands, other software on the same machine can change how they behave. Silent changes to an agent's instructions are a form of the "supply chain" risk that security teams worry about: a trusted tool becomes a way to steer a powerful assistant. The episode also lands as developers ask for more control over what their agents load.
Anthropic, the maker of Claude Code, is the company behind this assistant. This article reports the GitHub issue and press coverage. These are allegations from a user's logs, and Postman had not issued a public statement in the sources reviewed.
Sources: GitHub issue postmanlabs/postman-app-support #14370; The New Stack; daily.dev; AI Weekly; DEV.ua; Braindetox.
