Miggo at BlackHat 2026!
Easily map the attack path and prioritize which CVEs are a threat to your organization
Miggo AIGenerate vendor-ready rules for the observed attack patterns, plus reasoning and safe deployment guidance
| Package Name | Ecosystem | Vulnerable Versions | First Patched Version |
|---|---|---|---|
| aider-chat | pip | <= 0.86.2 |
Miggo AIRoot Cause AnalysisThe vulnerability lies in the trust boundary between the 'architect' and 'editor' models in Aider's architect mode. The output of the architect model is passed to the editor model without proper sanitization. The GitHub issue and the source code of aider/coders/architect_coder.py confirm that the editor_coder.run function is called with preproc=False. This disables the input preprocessing that would normally handle potentially malicious code, such as slash commands. An attacker can exploit this by providing malicious content to the architect model (for example, in a README file), which then generates malicious code that the editor model will execute. The vulnerable function is ArchitectCoder.reply_completed because it contains the insecure call to editor_coder.run.
ArchitectCoder.reply_completedaider/coders/architect_coder.py