You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
- Product-specific consent UX, support routing, escalation targets, or transcript-sharing policy.
45
47
46
48
Products can require Agents API because they build on the substrate. Agents API must not depend on any product plugin, import product classes, mirror a product source tree, or encode product vocabulary as generic runtime API.
47
49
@@ -107,6 +109,10 @@ wp_register_agent(
107
109
-`AgentsAPI\AI\IterationBudget`
108
110
-`AgentsAPI\AI\AgentConversationResult`
109
111
-`AgentsAPI\AI\AgentConversationLoop`
112
+
-`WP_Agent_Consent_Policy_Interface`
113
+
-`WP_Agent_Default_Consent_Policy`
114
+
-`AgentsAPI\AI\Consent\AgentConsentOperation`
115
+
-`AgentsAPI\AI\Consent\AgentConsentDecision`
110
116
-`AgentsAPI\AI\Tools\RuntimeToolDeclaration`
111
117
-`AgentsAPI\AI\Tools\ToolCall`
112
118
-`AgentsAPI\AI\Tools\ToolSourceRegistry`
@@ -142,6 +148,42 @@ add_filter(
142
148
);
143
149
```
144
150
151
+
## Consent Policy Boundary
152
+
153
+
Agents API owns a generic consent contract for runtime operations that carry different user expectations:
154
+
155
+
-`store_memory` — store consolidated agent memory.
156
+
-`use_memory` — use existing agent memory during a run.
157
+
-`store_transcript` — store a raw conversation transcript.
158
+
-`share_transcript` — share a raw transcript outside its owning context.
159
+
-`escalate_to_human` — escalate a run or transcript to a human/support adapter.
160
+
161
+
Memory consent and transcript consent are intentionally separate. Allowing an agent to store or use consolidated memory does not imply consent to persist or share raw transcripts. Escalation is also separate so support-mode adapters can ask their own product questions without adding support-product logic to Agents API.
162
+
163
+
Policies implement `WP_Agent_Consent_Policy_Interface` and return `AgentConsentDecision` values with `allowed`, `operation`, `reason`, and `audit_metadata` fields. Consumers should store those decision arrays alongside any memory write, transcript persistence/share event, or escalation event they apply.
`WP_Agent_Default_Consent_Policy` is conservative: non-interactive modes are denied by default, and interactive modes still require explicit per-operation consent. Products can supply adapter-specific policies for their own UX, authorization ceilings, support routing, retention rules, and audit stores.
186
+
145
187
## Conversation Compaction
146
188
147
189
Agents can declare support for runtime conversation compaction without tying Agents API to a provider or model executor:
0 commit comments