Capability Lifecycle

COP treats capability development as a lifecycle. ARP/JARVIS support this lifecycle with durable artifacts and stable interfaces (for example publishing a capability as a versioned NodeType).

Lifecycle (high level)

1. Define: describe a capability (what it does, schemas, side effects).
2. Implement: write the atomic/composite node implementation.
3. Evaluate: run deterministic tests and/or rubric-based evaluation; collect artifacts.
4. Publish: register the NodeType in Node Registry.
5. Promote: gate “stable” vs “experimental” via scorecards and policy.
6. Reuse: selection + planners prefer promoted capabilities; build larger workflows from smaller proven ones.
7. Operate: monitor regressions, rollback, and re-evaluate continuously.

What the stack provides today (v0.x)

JARVIS v0.3.x focuses on the execution fabric and core artifacts:

• NodeTypes, NodeRuns, Runs
• candidate sets and binding metadata
• durable event streams and artifact references
• policy checkpoints for allow/deny enforcement

Evaluation “promotion gates” are tracked as design intent and will expand as the core stabilizes.

Where to go next

• ARP objects and artifacts: ARP Concepts
• JARVIS components: JARVIS