Satna-Project/universal-reverse-engineering-engine

An evidence-driven, dual-agent (Gemini + Claude) reverse-engineering framework for reconstructing software architectures, APIs, protocols, and AI workflows

Summary

UREE is a prompt- and template-driven framework for reverse-engineering software systems, APIs, web applications, protocols, data pipelines, and AI workflows. It assigns Gemini the orchestration and specification-custodian role and Claude an independent analytical and adversarial role, with a gated 10-phase process for producing evidence-backed reconstructions.

What it is useful for

Engineers can use UREE when an existing API, web application, protocol, or legacy system needs to be understood before modernization or interoperability work. It is also relevant to clean-room reconstruction and authorized security or system audits, where conclusions must distinguish direct observations from assumptions.

The process makes uncertainty explicit. Every architectural or behavioral claim must reference an immutable Evidence ID such as EV-001 and a confidence rating: C1 for directly confirmed behavior, C2 or C3 for progressively weaker inference, and C0 when evidence is insufficient. The workflow also models black-box inputs, outputs, and state transitions before proposing internal technical details.

For example, an engineer investigating an undocumented API could define the target, inventory available artifacts, reconstruct its observable behavior, and then produce a technical model and specification. Implementation proceeds in a separate clean-room phase. Completion requires static, unit, integration, end-to-end, regression, and differential validation, plus a requirement traceability matrix. A failed completion gate sends the work back to behavioral reconstruction rather than accepting unresolved gaps.

UREE is explicitly limited to authorized, clean-room analysis. It is a framework for engineering reconstruction work, not a shortcut for analyzing software without evidence or permission.

How engineers can use it

Clone the repository and create an isolated target workspace:

git clone https://github.com/Satna-Project/universal-reverse-engineering-engine.git
cd universal-reverse-engineering-engine
chmod +x scripts/init_workspace.sh
./scripts/init_workspace.sh my-target-system

The script populates the workspace with evidence logs, handoff payloads, and specification templates. Next, load prompts/gemini_orchestrator.md as the system prompt in the Gemini workspace and prompts/claude_analyst.md in the Claude workspace.

To begin Phase 0, fill in workspaces/my-target-system/00_target_definition.md with the artifacts available for the target. Pass that definition to Gemini to generate the initial evidence register and Claude handoff payload. From there, follow the phase gates, using the templates under templates/ and the inspection matrices in profiles/domain_routing_profiles.md.

The README does not document a separate dependency-install step or a single automated command that runs both agents; engineers must provide the artifacts and operate the two workspaces themselves.