From 7870827eaa0c2fc6b52dff468c25875485501984 Mon Sep 17 00:00:00 2001
From: Jack Zhuang <277994282+os-zhuang@users.noreply.github.com>
Date: Sun, 21 Jun 2026 14:29:55 +0800
Subject: [PATCH] chore(spec): drop the dead agent `memory.shortTerm` config
 (ADR-0013 D3, cloud#339)
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

`memory.shortTerm` ({maxMessages, maxTokens}) declared a working-memory window
that NOTHING in the runtime ever consumed — a config that lies. Per cloud
ADR-0013 D3, cross-turn grounding is done by tools reading live state and the
context budget is governed by the per-request token guardrail, not by this
field. The only place it was set was data-chat-agent's definition (also dead).

- Remove `shortTerm` from the agent `memory` zod schema (keeping `longTerm` /
  `reflectionInterval` — forward-looking, off-by-default, not misleading).
- Remove the dead `memory.shortTerm` declaration from data-chat-agent.
- Update the memory-config spec test accordingly.

No runtime behavior change (nothing read it). Public API surface unchanged
(the gate tracks export names, not zod field internals); spec agent tests 45/45.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---
 .../service-ai/src/agents/data-chat-agent.ts       |  7 -------
 packages/spec/src/ai/agent.test.ts                 |  5 -----
 packages/spec/src/ai/agent.zod.ts                  | 14 ++++++--------
 3 files changed, 6 insertions(+), 20 deletions(-)

diff --git a/packages/services/service-ai/src/agents/data-chat-agent.ts b/packages/services/service-ai/src/agents/data-chat-agent.ts
index 5cb0bc8b73..37a6f0a0a7 100644
--- a/packages/services/service-ai/src/agents/data-chat-agent.ts
+++ b/packages/services/service-ai/src/agents/data-chat-agent.ts
@@ -96,12 +96,5 @@ Always answer in the same language the user is using. Detailed tool-usage guidan
     maxIterations: 10,
     allowReplan: true,
   },
-
-  memory: {
-    shortTerm: {
-      maxMessages: 20,
-      maxTokens: 4096,
-    },
-  },
 };
 
diff --git a/packages/spec/src/ai/agent.test.ts b/packages/spec/src/ai/agent.test.ts
index 9b3b4235fa..ada2df3e5d 100644
--- a/packages/spec/src/ai/agent.test.ts
+++ b/packages/spec/src/ai/agent.test.ts
@@ -589,10 +589,6 @@ Be precise, data-driven, and clear in your explanations.`,
         role: 'Persistent Assistant',
         instructions: 'Remember across sessions.',
         memory: {
-          shortTerm: {
-            maxMessages: 100,
-            maxTokens: 4096,
-          },
           longTerm: {
             enabled: true,
             store: 'vector',
@@ -602,7 +598,6 @@ Be precise, data-driven, and clear in your explanations.`,
         },
       });
 
-      expect(agent.memory?.shortTerm?.maxMessages).toBe(100);
       expect(agent.memory?.longTerm?.enabled).toBe(true);
       expect(agent.memory?.longTerm?.store).toBe('vector');
       expect(agent.memory?.reflectionInterval).toBe(5);
diff --git a/packages/spec/src/ai/agent.zod.ts b/packages/spec/src/ai/agent.zod.ts
index 868d651033..1fbe0d1058 100644
--- a/packages/spec/src/ai/agent.zod.ts
+++ b/packages/spec/src/ai/agent.zod.ts
@@ -173,14 +173,12 @@ export const AgentSchema = lazySchema(() => z.object({
 
   /** Memory Management */
   memory: z.object({
-    /** Short-term (working) memory configuration */
-    shortTerm: z.object({
-      /** Maximum number of recent messages to retain */
-      maxMessages: z.number().int().min(1).default(50).describe('Max recent messages in working memory'),
-
-      /** Maximum token budget for short-term context */
-      maxTokens: z.number().int().min(100).optional().describe('Max tokens for short-term context window'),
-    }).optional().describe('Short-term / working memory'),
+    // NOTE: `shortTerm` ({maxMessages,maxTokens}) was removed (ADR-0013 D3,
+    // cloud#339). It declared a working-memory window that NOTHING in the
+    // runtime consumed — a config that lies. Cross-turn grounding is done by
+    // tools reading live state, and the context budget is governed elsewhere
+    // (the per-request token guardrail), not by this field. `longTerm` /
+    // `reflectionInterval` are kept as forward-looking, off-by-default config.
 
     /** Long-term (persistent) memory configuration */
     longTerm: z.object({