Teaching an AI Assistant What It Is Allowed to Know, Claim, and Do

Description and use

Zorgax gained a structured capability layer bounded by evidence, with CI validation and reproducible checkpoints.

Problem

Defining and proving the limits of an AI assistant’s actions and claims based on evidence, rather than simply expanding its knowledge.

Outcome

Zorgax now operates with explicit, evidence‑bound capabilities, reducing ambiguity in what it can claim or do.

Workflow

• Define capability schema with inputs, allowed actions, prohibited claims, evidence requirements, linked evidence, limitations, and state.
• Create initial conservative states for Contributor Profile Assistant, Research Evidence Navigator, and Interoperability Checkpoint Assistant.
• Implement CI validator to check schema, duplicates, registry consistency, status, and linkage to tested checkpoints.
• Add bounded contributor‑scoped checkpoints for each capability.
• Run reproducible bounded tests to transition capabilities to TESTED state.
• Merge PR #1573 and publish final capability states.
• Publish canonical merge commit and documentation.

Evidence

The new schema is: myzubster.zorgax-capability.v1

Only one of them starts as TESTED.

The validator checks things such as: schema validity; duplicate capability IDs; registry ↔ file consistency; status consistency; whether a TESTED capability actually links to bounded TESTED checkpoint evidence.