MBSE Agents helps an AI assistant investigate systems, understand relationships, explain engineering decisions, develop connected artifacts, and challenge claims against their evidence. It supports work from early concepts through operations, architecture, assurance and change review.
The agent combines a systems-engineering reasoning core with task and domain skills. Use it to think through an unfamiliar problem, examine an existing project, or produce the particular artifacts a decision requires. The skill catalog describes the available entry points.
Launch Codex in this checkout, or install selected skills into an existing
project. The local installer needs Python 3.10+ and no third-party packages or
credentials. Use python3 on Linux/macOS or py on Windows.
git clone https://github.com/ajhcs/mbse-agents.git
cd mbse-agents
python3 scripts/install.py --list
python3 scripts/install.py --project /path/to/project --skill mbse-systems-engineering --skill mbse-requirements --skill mbse-architecture --dry-run
python3 scripts/install.py --project /path/to/project --skill mbse-systems-engineering --skill mbse-requirements --skill mbse-architectureUse $mbse-systems-engineering to investigate this system and the decision we need to make. Work from the supplied material. Explain the boundary, relevant stakeholders, dependencies and competing interpretations. Develop the necessary engineering artifacts, retain source IDs and existing decisions, and identify the evidence that would change your recommendation.
Choose a focused task skill when the work is already clear. Domain skills cover aerospace, defense, automotive, medical devices and electronic systems. The workflows guide explains how to combine them, and the first-use tutorial supplies a brief and a complete path to reviewing the result.
The gallery connects domain baselines, executable teaching models and historical project studies. The historical cases emphasize projects from before 2015 and connect original source material to retrospective questions about system boundaries, interfaces, decisions, operations, organizations and evidence.
The cold-storage alarm walkthrough connects stakeholder needs, requirements, architecture, procedures and saved reference-model observations. Domain simulation evidence identifies the exact models and assumptions behind generated analyses. Each case distinguishes original project evidence, new analysis, executed model results and work that still needs physical or program evidence.
The benchmark and quiz system assesses problem framing, systems thinking, causal reasoning, evidence interpretation, technical judgment and communication. Artifact tasks assess existing-project review and change handling as well as drafting. Automatic quiz results and independently reviewed reasoning are reported separately.
Use the evaluation protocol to compare a model alone with the same model using the agent skills. Instruction changes should be supported by observed performance; a passing package check is not a model-quality result.
Codex, Claude Code,
Databricks, ChatGPT,
and portable hosts
have distinct discovery and installation mechanisms. Guides identify documented
adaptations and the checks actually performed. Plugin distribution
describes the generated portable package; the source checkout keeps its canonical
skills under .agents/skills.
The installer preserves existing project instructions, host settings and local skill edits. CSE provides optional executable engineering analysis and model methods; Ingenium supports editable presentation. Available connectors determine which live tools the host can use. See model and tool crosswalks.
python3 scripts/catalog.py --write
python3 scripts/validate.py
python3 -m unittest discover -s testsModel guidance, source checks, and contribution instructions describe the maintained contracts. The source is MIT-licensed. Referenced standards and historical materials remain under their publishers' terms.