flex-auth/docs/markitect-action-vocabulary.md

Markitect Action Vocabulary

This document defines the action vocabulary for Markitect as a flex-auth protected system. Actions are normalized before policy evaluation so Markitect local behavior maps cleanly to CARING capabilities and exposure modes.

Action	Markitect policy-gateway meaning	CARING capabilities	CARING planes	Exposure modes	Decision effects
read	Render or fetch one document/resource.	View	Data	Metadata, Masked, Plaintext	allow, deny, redact
query	Answer over a bounded resource set.	ViewCollection, Observe	Data	Metadata, Aggregated, Masked	allow, deny, redact
search	Search index or metadata across resources.	ViewCollection, Observe	Data	Metadata, Aggregated, Masked	allow, deny, redact
package	Build a context package from selected resources.	Create, Bind, ViewCollection	Intent, Data	Metadata, Masked	allow, deny, audit_only
activate_context	Activate a prepared context package for model/tool use.	Use, Execute	Intent, Policy	Metadata, Masked	allow, deny, audit_only
export	Materialize or transfer content outside Markitect.	Export	Data, Audit	Exportable, Plaintext	allow, deny, audit_only
workflow_run	Execute a workflow using Markitect resources.	Execute, Operate	Execution, Data, Audit	Metadata, Masked, Plaintext	allow, deny, audit_only
admin	Configure Markitect policy, identity, or resource controls.	Configure, Grant, Revoke, Audit	Configuration, Identity, Policy, Audit	Metadata, Plaintext	allow, deny, audit_only

read, query, and search never imply Export. Export is separate because it changes the exposure mode to Exportable and usually requires explicit conditions such as MFA and logging.

The code-level source of truth is internal/markitect/actions.go. The pinned manifest example in examples/markitect/protected_system_manifest.yaml mirrors that vocabulary as protected-system action definitions.