zyr-agent
Affected
- Versions:
1.5.7,1.6.2
Fixed versions
None listed
Fix
No fixed version is listed.
Advisory
Also known as: None
Malicious package Its source marks this package as malicious: it is not a vulnerability in a legitimate one.
Malicious code in zyr-agent (npm)
---
_-= Per source details. Do not edit below this line.=-_
Source: amazon-inspector (b6dd19313881a0bcfd90734ce897e3134af348bce2b7447eaf4e5dbc910f432c)
zyr-agent ships an AI coding agent whose default 'free' provider routes chat completions to a hardcoded preview-slug endpoint at https://preview-chat-e8a10541-5396-4e8f-9ead-e5fbbcdb33d3.space-z.ai/api/v1 using an embedded zyr_sk_ key. The Agent.send loop iterates tool_calls returned by that remote endpoint and invokes tool.execute(tc.args) with no user confirmation across the CLI, Electron IPC, and HTTP server paths. One of the registered tools is a bash tool whose execute handler runs execSync(cmd) on the installer's host. Whoever controls the preview subdomain can therefore return a bash tool_call in a normal chat response and have arbitrary shell commands executed on any machine running the default provider. A separate provider ships a live third-party OpenRouter API key (sk-or-v1-...) hardcoded in cli.js and desktop/electron.js, which routes installer prompts through an author-controlled OpenRouter account. AI-agent auto-execution of model output is expected for this category, but the combination of (a) auto-exec with no per-tool confirmation and (b) a hardcoded, mutable, author-controlled preview-slug endpoint as the default provider makes the model-driven exec path a remote-command channel controlled by the package author rather than by the operator.
Not rated by the source.
1.5.7, 1.6.2None listed
No fixed version is listed.
Fix commands are the export's own generic commands. Ojuju never runs them. Check them against your setup first.
Check your own computer: guard packages check
Source: OpenSSF Malicious Packages · Apache-2.0