Maintain an agent definition

Core meaning

Capability purpose: Make each persistent agent's intended role and operating boundaries inspectable, portable, versioned, and owner-controlled.
Meaningful outcome: A current agent definition exists with stable identity, purpose, responsibilities, skills, limits, permissions, memory scope, source access, operating instructions, model preferences, lifecycle state, and version history.
Boundaries — includes: definition creation, revision, owner approval, identity reference, purpose, responsibilities, skills, limits, permissions, memory and source scopes, operating instructions, model preferences, status, versioning, rollback, and deployment-target references.
Boundaries — excludes: running the agent, synchronizing runtime copies, assigning a temporary runtime role, storing owner memory, executing work, and choosing outward identity for a specific action.
Terms and concepts: An agent definition describes a software agent as a governed system actor. Where useful, it may map to schema.org/SoftwareApplication; runtime roles may map separately to schema.org/Role.

Actor: The owner or an explicitly delegated agent-governance authority.
Command / intent: Create, revise, activate, suspend, restore, or retire a persistent agent definition.
Current state: The current definition, version history, affected scopes, runtime targets, and change authority can be determined.
Policies / invariants: Canonical definition and runtime copies remain distinguishable; permissions and memory scope are explicit; material changes are versioned and attributable; runtime-local prompts or service memory do not silently become canonical; rollback remains possible.
Transition: Validate the requested changes, create a new definition version or lifecycle state, preserve prior versions and rationale, and record affected runtime targets.
Result: A current, suspended, retired, restored, or rejected versioned agent definition.
Events / effects: May make runtime synchronization, capability discovery, permission review, or deployment actions eligible without performing them.
Unknowns: Universal agent-definition schema, signing, compatibility, and delegation policy are not established.

Agent identity, definition, runtime, memory, activity, and outward representation must not be collapsed into one record.
A definition change must not be represented as active in a runtime until synchronization is verified.

Keep purpose, responsibilities, skills, limits, permissions, memory scope, source access, operating instructions, current state, and evolution history owner-readable.

Portable compatibility rules for agent definitions across runtimes and providers require a later profile.

Statement	Semantic class	Evidence status	Source / note
Persistent agents require canonical, human-visible and agent-readable definitions distinct from runtime, memory, and activity.	rule/invariant	sourced	J01–J03.