Act on this, mostly signal

Safe REA Skill runs reverse engineering agents in two restricted containers

What to do

Use the Safe REA Skill if you reverse engineer with agents and need the agent sandboxed from its receiver.

Why

Open-source setup with frozen, hash-approved file copies and a hardening update adding 3,702 boundary cases.

What the research found

Open-source REA Safe Agent skill sets up REA and its receiving agent in separate restricted Docker containers, sharing only frozen hash-approved file copies, with a security hardening update adding 3,702 boundary cases.

Claims and how well the evidence supports them

Run REA and the receiving agent in separate restricted Docker containers with network=none, read-only roots and zero host mounts for the agent 99%
Only frozen, hash-approved copies of explicitly selected files are shared with REA 98%
The latest hardening update reports 3,702 reproducible boundary cases 98%
A 27-tool MCP allowlist is preserved in setup 98%

Evidence

github.com https://github.com/dagoSte/rea-safe-agents

Actionability rated medium.