Inspectable by design
Framework, scoring logic, schema, examples, and implementation are available for review.
ARIMAS open project
Open
Evidence
Governance
Make agent capability visible.
Arima AiCMM is an open-source framework for classifying, evaluating, and communicating what AI agents can and cannot do. Evidence becomes a comparable capability profile instead of an opaque claim.
Claims need support
Twelve universal dimensions produce capability fingerprints grounded in observable indicators.
Trust is structural
Safety, explainability, interoperability, cost, and domain fit participate in deployment decisions.
A shared language for responsible agents.
AiCMM separates universal capability from domain-specific requirements, then packages the result in a machine-readable Agent Card.
12 dimensions
Autonomy, reasoning, memory, learning, tools, collaboration, embodiment, explainability, safety, interoperability, cost efficiency, and domain alignment.
Agency classification
A derived qualification layer distinguishes scripted automation, assistants, and genuinely agentic systems.
Seven governance rules
Cross-dimensional rules prevent strong headline scores from hiding deployment-critical weaknesses.
Agent Cards
Portable JSON records communicate scores, evidence, constraints, metadata, and intended use.
Capability fingerprints
Profiles make different agents comparable without flattening capability into one marketing label.
Domain deep-dives
Level 1 profiles add domain-specific requirements while retaining the universal baseline.
One installation agent. Three controlled phases.
Copy the short project-aware starter or the complete installation agent into your preferred AI CLI. It explains AiCMM, loads the guidance, announces each phase, and stops at every approval gate.
01
Screening always comes first.
Phase 1 is read-only. The CLI cannot install or run AiCMM until it presents provenance, dependency, security, runtime, and existing-installation findings for your approval.
Evaluate and screen
Verify source, license, commit, dependencies, permissions, sensitive features, loopback safety, and any existing installation.
Install and verify
After approval, install the screened revision, test it, prove loopback-only binding, and document rollback. Skip only when an existing installation passes every check.
Demonstrate and learn
Start the verified local system and guide the user through dimensions, evidence, governance, Agent Cards, validation, and cleanup.
Install with an agentic CLI
Paste this short starter into Copilot CLI, Claude, Google AGY, or another web-enabled agentic CLI. The linked Markdown installation agent is self-contained.
I want to evaluate and install Arima AiCMM, the open-source Agent Capability Maturity Model for evidence-based agent evaluation, Agent Cards, and governance.
Read and follow the complete installation agent at:
https://arimas.ai/AiCMM/ArimaAiCMMInstallationAgent.md
Explain the project and present its three phases. Do not install or run anything until Phase 1 screening is complete and I explicitly approve proceeding.Install manually
Complete the same screening first, then clone, verify, require explicit loopback binding, start locally, and follow the tutorial.
git clone https://github.com/snchande/Arima-AiCMM.git
cd Arima-AiCMM
git remote -v
git rev-parse HEAD
mvn clean verify
# Review the loopback-binding requirement
# in the full manual guide before starting.Phase 2 installs only when needed.
The same online guide continues after you accept Phase 1. There is no second prompt to copy and no silent transition between phases.
02
Existing installation?
If Phase 1 verifies its official source, screened commit, prerequisites, tests, configuration, and loopback safety, the CLI offers to skip Phase 2 and move directly to the tutorial.
Preserve the screened revision
The CLI stops if source or online instructions changed after Phase 1.
Build, test, and contain
Use normal permissions, run verification, and patch-or-stop when explicit loopback binding is absent.
Report before teaching
Record commands, files, listener addresses, warnings, and rollback, then ask before Phase 3.
Verified project reference
Java 17+ and Maven 3.8+mvn clean verifymvn clean package -DskipTestshttp://localhost:8080 after verified loopback bindingLearn by examining evidence.
These tutorials are designed for a local AiCMM installation. They move from framework concepts to a reviewable Agent Card without sending private source, credentials, or evaluation data to ARIMAS.
Read a capability fingerprint
Walk through the twelve universal dimensions, the derived agency classification, and the difference between capability and deployment readiness.
Inspect an Agent Card
Trace every score to observable evidence, identify unsupported claims, and review the JSON representation against the schema.
Evaluate an agent safely
Ask your CLI to inspect public documentation and source, draft evidence with citations, then validate the card locally before accepting scores.
03
The guide starts this tutorial automatically after approval.
Phase 3 uses a disposable workspace and public or fictional evidence. It does not publish cards, overwrite examples, connect providers, or invoke sensitive runtime features.
Framework orientation and capability fingerprint.
A complete transparent evaluation workflow.
Using cross-dimensional rules to review readiness.