Advisories
For your digest

claude-agent-sdk: Claude Agent SDK: Argument Injection via resume Option Allows Arbitrary Command Execution

It touches something Phylax follows. No evidence of use yet.

Does it affect what I run?

claude-agent-sdk (PyPI)

Checked in your browser. The version you type is not sent anywhere.

Who is affected

  • claude-agent-sdk (PyPI): every version, before 0.2.121

The fix

  • claude-agent-sdk (PyPI) fixed in 0.2.121

    pip
    python -m pip install claude-agent-sdk==0.2.121
    uv
    uv add claude-agent-sdk==0.2.121
    poetry
    poetry add claude-agent-sdk@0.2.121
    Check what is installed
    python -m pip show claude-agent-sdk
    Paste into Cursor, Lovable or Claude Code
    Upgrade the PyPI dependency claude-agent-sdk to 0.2.121 or the nearest later release, to fix CVE-2026-96620. Change only what the upgrade needs: the manifest, the lockfile, and any code that breaks because of it. Do not change anything else and do not run commands unrelated to this upgrade. Then run the tests, tell me what changed, and show me the diff before committing.

Fix commands generated from the advisory's versions. No model wrote them.

In plain words

Brief pending: here are the facts and the fix. The explanation in plain words follows when it has been written and checked.

Evidence

No evidence that it is being used, or can be. It is listed because of what it affects.

Timeline

  1. First published here, as "For your digest".