registry / micro-models-agent / 0.2.35

micro-models-agent@0.2.35

Micro Models Agent (MMA) — LLM agent harness for small models (Qwen3.5-9B, 32K-64K context)

AI Security Review

scanned 2h ago · by lpm-firewall-ai

A user-invoked coding agent can let model output trigger shell, filesystem, web, and configured MCP actions. No install-time or covert attack chain is established.

Static reason

One or more suspicious static signals were detected.; previous stored version diff introduced dangerous source

Trigger

User runs `mma` with a prompt or enters the interactive REPL.

Impact

Prompt injection or unsafe model output could cause arbitrary local commands under the invoking user's permissions.

Mechanism

Model-generated tool calls execute PowerShell commands and configured MCP commands.

Rationale

This is not concrete malware, but it exposes broad local command execution through an LLM agent with insufficiently enforced approval boundaries. Flag as a warning for dangerous capability rather than publish-blocking malicious behavior.

Evidence

package.jsondist/index.jsdist/agent/loop.jsdist/tools/bash.jsdist/tools/mcp.jsdist/tools/syntax-check.jsdist/updater.js

Network endpoints1

lite.duckduckgo.com/lite/

Decision evidence

public snapshot

AI called this at 91.0% confidence with false-positive risk.

micro-models-agent@0.2.35

AI Security Review

Decision evidence

Source & flagged code

Source & flagged code

Findings

Source & flagged code