Postman accused of auto-installing AI plugins and approving Codex hooks during desktop app upgrades

Postman accused of auto-installing AI plugins and approving Codex hooks during desktop app upgrades

N
News Editor
2026-11-04 11:06:41
Postman is facing scrutiny after GitHub users reported that its desktop app automatically installed plugins for OpenAI Codex and Anthropic Claude Code during upgrades, while also adding Postman CLI. A GitHub issue filed by user indicava on Oct. 25 said Postman Desktop 12.33.4 on Mac installed the integrations without explicit approval and altered Codex settings by adding a SessionStart hook tied to Postman’s api-engineer skill. Source code cited in the report showed an approvePostmanHook function that writes a hook hash into Codex’s trusted_hash field, effectively handling a review step that Codex would normally present to the user. The issue author said the report did not establish that data had been exfiltrated. Another user, ronthedrummer, also said they had no evidence that data was uploaded. The controversy spread to Hacker News, where a post about the issue reached the top spot on Nov. 3. Postman later responded through a GitHub account named zecoquit, saying the bundled installation was meant to help users who had trouble setting up the tools manually, and that the company would review its hook approval and session-start behavior. As of Nov. 4, Postman’s plugin repository still contained the hook approval logic, and release notes for desktop version 12.35.2 did not mention changes to plugin installation behavior.

Postman has been accused by users of automatically installing AI coding-agent plugins during desktop app upgrades, including integrations for OpenAI Codex and Anthropic Claude Code, along with Postman CLI.

The complaint was laid out in a GitHub issue opened on Oct. 25 by user indicava. In that report, indicava said Postman Desktop 12.33.4 on Mac installed plugins for Codex and Claude Code and added the Postman CLI command-line tool. He described the behavior as a "serious consent and trust boundary issue," arguing that an API tool should not make itself part of another tool’s default workflow simply because a user opened the app.

Logs shared in the issue showed the Oct. 24 installation source as "auto-existing-upgrade," indicating an automatic install during an upgrade for an existing user. The result fields included claude-code:installed and codex:installed. According to indicava, the installation happened after he opened Postman for the first time in a while. Another user, ronthedrummer, reported on Nov. 3 that Postman auto-updated from 12.31.3 to 12.35.1 and installed the CLI and both agent plugins during the same launch.

Codex settings were modified with a SessionStart hook

The issue went beyond plugin installation. Codex’s global configuration was also modified with a hook, a command that runs automatically at specific moments. This hook was attached to the SessionStart event, which fires whenever Codex starts a new conversation, clears a conversation, or compacts one. The command prints a message instructing the agent to load the postman:api-engineer skill before handling "any non-trivial API engineering task."

A review of the public postman-plugin source code showed that the message is wrapped in an "EXTREMELY_IMPORTANT" label and applies across API work from design through deployment. The text also says that instructions in AGENTS.md, CLAUDE.md, and direct user requests take priority. Indicava said that helps explain why Codex repeatedly routed API-related work toward Postman’s skill.

Codex normally places a confirmation step in front of hook activation. Postman’s own installation documentation says that when users install through the Codex app or command line, the hook is untrusted by default. On the next Codex launch, users should see a "Hooks need review" prompt and approve it themselves. By contrast, documentation for Postman’s installer, which detects installed agents on a machine, says it will "approve this hook for you."

In source code from codex-hooks.ts cited by indicava, the approvePostmanHook function first asks Codex for the hook’s hash, then writes that value into the trusted_hash field in the configuration file. A code comment says, "Approves only Postman’s current hook after installation." The feature came from PR #124, merged on Oct. 21. That pull request also stated that "Hooks can run outside the sandbox." Based on Postman’s installation flow, a step that Codex would normally require the user to confirm was handled by the installer on this path.

Indicava also drew a limit around what the evidence showed. He said the source code proves only that the installer "attempts" to approve the hook. The trust record in the configuration file, by itself, does not prove who wrote it. On ronthedrummer’s machine running 12.35.1, that trust record was not present. He later corrected his own comment and said he did not see the Codex hook actually execute.

The plugin also connected Codex to a Postman-hosted MCP server. MCP is a protocol that lets AI agents connect to external services. The configuration pointed to mcp.postman.com. Logs shared by indicava showed the connection failed to start because "Auth required." For Claude Code, the enabled integration was Postman plugin version 2.0.1 from Anthropic’s official plugin marketplace. Another user, ck-tecso, said a Postman plugin was also present in a GitHub Copilot configuration file.

The issue did not claim proven data exfiltration

The GitHub issue sat without comments for eight days, then began drawing responses on Nov. 2 before being posted to Hacker News the next day. The Hacker News submission said Postman had "silently auto-installed AI plugins and attempted to share all API data." It reached the No. 1 spot on the site’s front page on Nov. 3. As of 3:30 p.m. Taiwan time on Nov. 4, the post had 1,197 points and 392 comments. The second-ranked post that day had 957 points. The GitHub issue itself had 209 reactions and 38 comments.

Still, the data-sharing claim in the Hacker News headline was not established in the issue text. Indicava wrote at the end of the report that it documented "unwanted integration and instruction changes" and did not claim that data exfiltration had been proven. Ronthedrummer also said he had no evidence that data was uploaded. As of publication, the Hacker News submitter, sidgarimella, had not posted in the discussion thread to explain the basis for that headline.

Postman’s plugin README describes a different data path. After a folder is connected to a Postman workspace through postman init, the execution results of CLI commands are "published" to that workspace by default. Those published results include collection execution results, lint results, and API interaction logs generated by an agent using postman request. The behavior can be disabled on a per-run basis with --report-events=false.

The README also says tool calls made through the MCP server are sent to Postman "by design." At the same time, the plugin’s api-engineer skill instructs the agent to obtain user consent before pushing workspace content.

Postman said it will review hook approval and session-start behavior

In the early hours of Nov. 4 Taiwan time, a GitHub account named zecoquit replied in the thread using "we" and speaking on behalf of Postman. According to the report, that account was not publicly listed as a member of the postmanlabs organization and had no commit history in the postman-plugin repository. The comment was the account’s only public GitHub activity.

Zecoquit wrote: "We bundled Postman CLI and agent plugins with the desktop app because many users had trouble installing and connecting them on their own." He said the download page and the app update flow mention this behavior, and that users can uninstall any component from within the app. He also said the company would "make the opt-in path clearer, especially for existing users," and that the team was reviewing hook approval and session-start behavior, with updates to be posted in the thread after changes go live.

As of publication, that reply had received 25 downvotes and 2 upvotes. A commenter named ceberttylertech pushed back, saying opt-in should apply to everyone and that this behavior should not appear even on a fresh machine. In the original issue, indicava had asked Postman to stop auto-installing for existing users and to remove the installer’s ability to approve hooks on its own. Zecoquit’s response, at this stage, promised only a clearer opt-in path.

A comparison with snapshots of Postman’s download page preserved by the Wayback Machine showed that the "The Postman app" section on Oct. 24 and Oct. 26 did not mention plugins. The Oct. 26 snapshot had a separate "Postman plugins" section that told users to add plugins to their own AI agents, but it did not say the desktop app would install them automatically.

The current page now includes the sentence, "The Postman app includes Postman CLI & Postman plugins." Official release notes mentioned only that CLI would be installed with the app in version 12.31.7, released on Oct. 9. Later versions did not mention plugins. The report said the app’s update prompt could not be reviewed.

Zecoquit said the plugins can be uninstalled from within the app. User ck-tecso reported that after clicking "Help → Postman CLI and Plugins → Uninstall Postman Coding Agent Plugin," the Postman plugin still remained in Copilot’s configuration. That user did not have Claude Code, Codex, or Cursor installed, so they could not confirm whether the plugin could be fully removed from those agents.

As of the afternoon of Nov. 4 Taiwan time, the main branch of postman-plugin still contained approvePostmanHook, and Postman’s installation documentation still said it would approve the hook for users. Postman released desktop version 12.35.2 on Nov. 4, but the release notes said only "critical bug fixes and enhancements" and did not mention plugin installation behavior.

What the report established

  • According to the GitHub issue, Postman Desktop 12.33.4 automatically installed Postman plugins for Codex and Claude Code, along with Postman CLI, during upgrades for existing users. Another user reported the same behavior when updating from 12.31.3 to 12.35.1.
  • The issue author explicitly said the report did not prove data exfiltration. As of publication, no one in the 38-comment thread had produced evidence showing that data had been uploaded. At the same time, the plugin README says CLI execution results are published to a Postman workspace by default once a folder is connected, and that behavior can be disabled per run.
This article was originally published by Bit.Fan. For more cryptocurrency news and market insights, visit www.bit.fan.
6000

Disclaimer:

The market information, project data, and third-party content displayed on this platform are for industry information sharing only and do not constitute any form of investment advice or return commitment.

Cryptocurrency trading carries high risks. Users should fully assess their risk tolerance and make independent decisions. All profits, losses, and legal responsibilities are borne by the users themselves.

Postman accused of auto-installing AI plugins and approving Codex hooks during desktop app upgrades | Bit.Fan