A developer reported that Postman added plugins to Claude Code and Codex without permission. This guide explains how to audit your agents and regain control.
A developer reported that after opening the Postman desktop application, they found plugins had been installed in two of their coding agents, Claude Code and OpenAI Codex, without their authorization. The New Stack covered the case, and the affected person reported it on October 25. We won't comment on anyone's intentions; this guide is about how to review what's installed on your agents and how to minimize surprises.
What you need
To follow this guide, you only need access to your development machine, a terminal, and the agents you use (in the original case, Claude Code and Codex). It's helpful to have Postman's documentation on detected agents handy, which explains how the tool recognizes agents present on the system.
It's also useful to know where each agent stores its configuration. The exact paths depend on the agent, operating system, and version, so we won't provide any universal ones here: consult the official documentation for each tool before deleting or modifying anything.
- Back up your agents' configuration folders before making any changes.
- List the plugins, extensions, or hooks each agent has installed and compare them with what you remember adding yourself.
- Identify any you don't recognize and find out which application or provider registered them, by reading their documentation or code if it's public.
- Decide on a case-by-case basis: keep, disable, or remove, and note the reason.
- Reopen the suspicious application and check if it recreates the changes; if it does, look for a disable option or consult the provider.
How to inspect what's installed
The specific mechanism varies by agent. In Codex's case, the Postman installer's behavior can be read in its source code, allowing you to check firsthand what it touches and where, instead of relying on assumptions.
# Backup first (adjust paths to your agent)
cp -r ~/AGENT_CONFIG_PATH ~/AGENT_CONFIG_PATH.bak
# Compare after opening a suspicious application
diff -r ~/AGENT_CONFIG_PATH.bak ~/AGENT_CONFIG_PATHRunning a diff before and after opening an application tells you exactly which files changed. Replace AGENT_CONFIG_PATH with the actual folder documented by your agent; the names in the example are placeholders, not verified paths.
Reducing the attack surface from now on
Review the preferences of any desktop application that integrates with your programming tools and disable automatic agent detection if that option exists. And repeat the audit every time you update a major application: integrations often come with new versions.
If you're interested in delving into the defensive side, roadmap.sh's cybersecurity and DevSecOps learning paths provide a good map of concepts. And to see how other teams are adjusting their relationship with these tools, there's this analysis on developers intentionally reducing their AI's autonomy.
Related articles from LaRebelión
Guide: Prepare Your Android UI for AI Agents with Jetpack Compose
Guide: Evaluate Copilot Agent Harness in Foundry Before Adoption
How to Evaluate Entire, the New GitHub Alternative Focused on Pull RequestsOriginal source: thenewstack.io
Article generated with AI.larebelion