feat: add intent-aware tauri reply policy

Author: Jacobinwwey

docs/diataxis/en/explanation/development-progress-dashboard.md

@@ -72,6 +72,7 @@ The next real improvement beyond that baseline is now also in code:
 - `src/learning/KnowledgeLearningPlatform.ts` no longer treats `assistantBlocks` as a thin transport wrapper around the same old answer string,
 - the scoped conversation reply is now organized into explicit overview / explanation / evidence summary / memory notice / action guidance sections before citations and knowledge-action affordances are appended,
 - those sections now also consume real scoped data instead of only templated filler: explanation is anchored to the strongest scoped point, evidence summary reflects actual citations, and action guidance now carries memory follow-through hints,
+- the reply policy is now also intent-aware, so comparison-style and how-to-style prompts can shape the explanation and next-action sections differently,
 - which means the Tauri agent surface can now look materially different even when the underlying knowledge result set is unchanged.
 
 The next gap is narrower now:

docs/diataxis/zh/explanation/development-progress-dashboard.md

@@ -72,6 +72,7 @@ Tauri-first reply rendering 基线已交付：
 - `src/learning/KnowledgeLearningPlatform.ts` 不再把 `assistantBlocks` 当成“同一段旧 answer 的运输包装层”，
 - scoped conversation reply 现在会先组织成明确的 overview / explanation / evidence summary / memory notice / action guidance，再追加 citations 与 knowledge-action affordance，
 - 这些 section 现在也会吃进真实 scoped 数据，而不再只是模板填充：explanation 会锚定最强 scoped point，evidence summary 反映真实 citation，action guidance 还会带上 memory follow-through hint，
+- reply policy 现在也具备 intent awareness，因此 comparison-style 与 how-to-style prompt 可以把 explanation 和 next-action section 引导成不同语气与用途，
 - 这意味着即便底层知识命中集合没变，Tauri 中的 agent 输出也已经可以在结构上明显不同。
 
 当前剩余缺口已经收窄为：

docs/en/implementation_plan.md

@@ -70,6 +70,7 @@ Align the active implementation plan with current code reality:
   - artifact-style handling for large HTML assistant outputs through sandboxed preview.
   - structured reply composition now splits the assistant output into overview / explanation / evidence summary / memory notice / action guidance blocks instead of emitting only one wrapped markdown answer.
   - those sections are now also semantically richer: the explanation is anchored to the strongest scoped knowledge point, the evidence summary reflects real scoped citations, and next-action guidance incorporates both scoped nodes and memory-action follow-through.
+  - reply composition is now intent-aware as well: comparison-style and how-to-style prompts no longer reuse the exact same explanation/action phrasing as plain explanatory prompts.
 
 #### Next execution order
 

docs/en/task.md

@@ -11,6 +11,7 @@
 - [x] The Tauri-first plan to evolve toward shared Reader-aligned rich reply rendering is now implemented as the current baseline while preserving knowledge-point/capability compatibility.
 - [x] The backend reply composer now uses `assistantBlocks` as a real organization layer in Tauri: overview, explanation, evidence summary, memory notice, and next-action guidance are emitted as distinct blocks instead of only wrapping the old answer text.
 - [x] The new Tauri reply sections are no longer template-only: explanation, evidence, and next-action guidance now derive from actual knowledge points, citations, and memory-action hints.
+- [x] The reply policy is now intent-aware for Tauri agent output: comparison-style and how-to-style prompts can yield different explanation/action phrasing instead of one generic section style.
 - [x] FR-010 is now governed by the current workflow reality instead of the removed transition-era assumption: repo-owned workflows pin `actions/setup-node@v4` to Node 24 and no longer rely on `FORCE_JAVASCRIPT_ACTIONS_TO_NODE24`.
 - [~] Remote CI closure remains a live release bar rather than a static claim: `Fixrisk Operational Readiness` must stay green on `main`, while residual Node 20 deprecation annotations from marketplace actions remain documented as non-blocking external debt.
 

docs/zh/implementation_plan.md

@@ -71,6 +71,7 @@
   - 面向大型 HTML assistant output 的 artifact-style 隔离路径（sandboxed preview）。
   - assistant 输出现在按 overview / explanation / evidence summary / memory notice / action guidance 分块组织，而不再只是单个 markdown answer 的包装。
   - 这些 section 现在也具备更实在的语义内容：explanation 会锚定最强 scoped knowledge point，evidence summary 会反映真实 scoped citation，next-action guidance 也会吸收 scoped node 与 memory action 的 follow-through 建议。
+  - reply composition 现在也具备 query intent awareness：comparison-style 与 how-to-style prompt 不再复用与普通 explanatory prompt 完全相同的 explanation / action phrasing。
 
 #### 下一步执行顺序
 

docs/zh/task.md

@@ -7,6 +7,7 @@
 - [x] 这条 Tauri-first 路线已作为当前基线落地：在保持现有 knowledge-point / capability 兼容的前提下，引入了共享 Reader-aligned rich reply rendering。
 - [x] 后端回复组织层现在已经把 `assistantBlocks` 用成真正的 Tauri 输出结构：overview、explanation、evidence summary、memory notice、next-action guidance 已分块输出，而不再只是旧 answer 文本的包装。
 - [x] 新的 Tauri reply sections 已不再只是模板文本：explanation、evidence、next-action guidance 现在会由真实 knowledge point、citation 和 memory action hint 驱动。
+- [x] Tauri agent 输出的 reply policy 现在已具备 intent awareness：comparison-style 与 how-to-style prompt 会得到不同的 explanation / action phrasing，而不再只有一套通用 section 风格。
 - [x] FR-010 现已按当前工作流现实治理，而不是继续沿用已删除的过渡期假设：仓库自有工作流固定为 `actions/setup-node@v4` + Node 24，且不再依赖 `FORCE_JAVASCRIPT_ACTIONS_TO_NODE24`。
 - [~] remote CI 是否闭环仍以 `main` 上的实时 `Fixrisk Operational Readiness` 结果为准；其余 marketplace action 带来的 Node 20 弃用注解继续作为非阻塞外部债务记录。
 

src/agent_workspace.frontend.test.ts

@@ -2784,6 +2784,12 @@ describe('agent workspace learning-path integration', () => {
                     );
                 }
                 if (markdown.includes('## Next Actions')) {
+                    if (markdown.includes('side by side')) {
+                        return (
+                            '<h2>Next Actions</h2>'
+                            + '<p>inspect the strongest nodes side by side</p>'
+                        );
+                    }
                     return (
                         '<h2>Next Actions</h2>'
                         + '<p>Persist the latest user focus to scoped conversation memory.</p>'

src/learning/KnowledgeLearningPlatform.test.ts

@@ -1496,6 +1496,81 @@ describe('KnowledgeLearningPlatform', () => {
         expect(persistedMemory.results[0]?.tags).toContain('scope_corpus:optics');
     });
 
+    test('agent conversation explanation and next actions adapt to comparison-style queries', async () => {
+        await platform.ingestKnowledge({
+            incremental: true,
+            documents: [
+                {
+                    documentId: 'doc_compare_scope',
+                    sourcePath: 'Knowledge_Base/optics/reflection.md',
+                    language: 'en',
+                    content: '# Reflection\nReflection and absorption differ in how optical energy is redirected versus dissipated.',
+                },
+                {
+                    documentId: 'doc_compare_scope_2',
+                    sourcePath: 'Knowledge_Base/optics/transmission.md',
+                    language: 'en',
+                    content: '# Transmission\nTransmission complements reflection when comparing optical interface behavior.',
+                },
+            ],
+        });
+
+        const response = await platform.agentConversation({
+            userId: 'agent_compare_user',
+            sessionId: 'session_compare_scope',
+            message: 'compare reflection vs absorption',
+            scope: {
+                corpusId: 'optics',
+                languages: ['en'],
+            },
+            persistMemory: true,
+        });
+
+        const markdownBlocks = (response.assistantBlocks || []).filter((block) => block.type === 'main_markdown');
+        expect(
+            markdownBlocks.some((block) => String((block as { markdown?: string }).markdown || '').includes('comparison baseline'))
+        ).toBe(true);
+        expect(
+            markdownBlocks.some((block) => String((block as { markdown?: string }).markdown || '').includes('Supporting comparison nodes'))
+        ).toBe(true);
+        expect(
+            markdownBlocks.some((block) => String((block as { markdown?: string }).markdown || '').includes('inspect the strongest nodes side by side'))
+        ).toBe(true);
+    });
+
+    test('agent conversation explanation and next actions adapt to how-to queries', async () => {
+        await platform.ingestKnowledge({
+            incremental: true,
+            documents: [
+                {
+                    documentId: 'doc_howto_scope',
+                    sourcePath: 'Knowledge_Base/optics/calibration.md',
+                    language: 'en',
+                    content: '# Calibration\nCalibration sequence requires establishing a baseline measurement, then validating response drift.',
+                },
+            ],
+        });
+
+        const response = await platform.agentConversation({
+            userId: 'agent_howto_user',
+            sessionId: 'session_howto_scope',
+            message: 'how to calibrate optical response',
+            scope: {
+                corpusId: 'optics',
+                languages: ['en'],
+            },
+            persistMemory: true,
+        });
+
+        const markdownBlocks = (response.assistantBlocks || []).filter((block) => block.type === 'main_markdown');
+        expect(
+            markdownBlocks.some((block) => String((block as { markdown?: string }).markdown || '').includes('starting anchor for the next concrete steps'))
+        ).toBe(true);
+        expect(
+            markdownBlocks.some((block) => String((block as { markdown?: string }).markdown || '').includes('move from explanation into concrete guided-learning or focus-mode steps'))
+        ).toBe(true);
+    });
+
     test('agent conversation exposes readiness and miss diagnostics when scoped retrieval is empty', async () => {
         const response = await platform.agentConversation({
             userId: 'agent_miss_user',

src/learning/KnowledgeLearningPlatform.ts

@@ -7624,6 +7624,45 @@ export class KnowledgeLearningPlatform implements KnowledgeLearningPlatformAPI {
         return `${memoryPrefix}The strongest scoped match is ${leadingPoint.title}. I found ${params.knowledgePoints.length} relevant knowledge point(s) and ${params.citations.length} citation(s).\n\nKey evidence:\n${evidenceLines.join('\n')}`;
     }
 
+    private classifyScopedConversationIntent(message: string): 'explain' | 'compare' | 'how_to' | 'generic' {
+        const normalized = String(message || '').trim().toLowerCase();
+        if (!normalized) {
+            return 'generic';
+        }
+        if (
+            normalized.includes('compare')
+            || normalized.includes('difference')
+            || normalized.includes('vs')
+            || normalized.includes('区别')
+            || normalized.includes('对比')
+        ) {
+            return 'compare';
+        }
+        if (
+            normalized.includes('how to')
+            || normalized.includes('how do')
+            || normalized.includes('steps')
+            || normalized.includes('plan')
+            || normalized.includes('如何')
+            || normalized.includes('怎么')
+            || normalized.includes('步骤')
+            || normalized.includes('方案')
+        ) {
+            return 'how_to';
+        }
+        if (
+            normalized.includes('what is')
+            || normalized.includes('why')
+            || normalized.includes('explain')
+            || normalized.includes('解释')
+            || normalized.includes('什么是')
+            || normalized.includes('为什么')
+        ) {
+            return 'explain';
+        }
+        return 'generic';
+    }
+
     private buildScopedConversationOverviewMarkdown(params: {
         knowledgePoints: AgentConversationKnowledgePoint[];
         citations: KnowledgeCitation[];
@@ -7648,19 +7687,29 @@ export class KnowledgeLearningPlatform implements KnowledgeLearningPlatformAPI {
     }
 
     private buildScopedConversationExplanationMarkdown(params: {
+        message: string;
         knowledgePoints: AgentConversationKnowledgePoint[];
         citations: KnowledgeCitation[];
         recalledMemories: AgentConversationMemoryRecord[];
     }): string {
         if (params.knowledgePoints.length <= 0) {
             return '## Explanation\n\nThe current scope did not return a strong enough knowledge point to explain the request directly.';
         }
+        const intent = this.classifyScopedConversationIntent(params.message);
         const strongestPoint = params.knowledgePoints[0];
         const explanationLines = [
             '## Explanation',
             '',
-            `**${strongestPoint.title}** is the current best scoped anchor.`,
         ];
+        if (intent === 'compare') {
+            explanationLines.push(`Use **${strongestPoint.title}** as the comparison baseline inside the current scope.`);
+        } else if (intent === 'how_to') {
+            explanationLines.push(`Use **${strongestPoint.title}** as the starting anchor for the next concrete steps.`);
+        } else if (intent === 'explain') {
+            explanationLines.push(`**${strongestPoint.title}** is the current best scoped anchor for the explanation.`);
+        } else {
+            explanationLines.push(`**${strongestPoint.title}** is the current best scoped anchor.`);
+        }
         const summary = normalizeWhitespace(String(strongestPoint.summary || strongestPoint.evidenceSnippet || '').trim());
         if (summary) {
             explanationLines.push('', summary);
@@ -7672,7 +7721,9 @@ export class KnowledgeLearningPlatform implements KnowledgeLearningPlatformAPI {
         if (supportingTitles.length > 0) {
             explanationLines.push(
                 '',
-                `Supporting scoped nodes: ${supportingTitles.join(', ')}.`
+                intent === 'compare'
+                    ? `Supporting comparison nodes: ${supportingTitles.join(', ')}.`
+                    : `Supporting scoped nodes: ${supportingTitles.join(', ')}.`
             );
         }
         if (params.recalledMemories.length > 0) {
@@ -7719,12 +7770,14 @@ export class KnowledgeLearningPlatform implements KnowledgeLearningPlatformAPI {
     }
 
     private buildScopedConversationActionGuideMarkdown(params: {
+        message: string;
         knowledgePoints: AgentConversationKnowledgePoint[];
         memoryActions: AgentConversationMemoryAction[];
     }): string {
         if (params.knowledgePoints.length <= 0) {
             return '## Next Actions\n\nNo actionable scoped knowledge card is available for this turn.';
         }
+        const intent = this.classifyScopedConversationIntent(params.message);
         const topTitles = params.knowledgePoints
             .slice(0, 3)
             .map((point) => `- ${point.title}`);
@@ -7736,7 +7789,11 @@ export class KnowledgeLearningPlatform implements KnowledgeLearningPlatformAPI {
         return [
             '## Next Actions',
             '',
-            'Use the scoped knowledge cards below to continue with focus mode or guided learning for the highest-signal nodes:',
+            intent === 'compare'
+                ? 'Use the scoped knowledge cards below to inspect the strongest nodes side by side before deciding which distinctions matter most:'
+                : intent === 'how_to'
+                    ? 'Use the scoped knowledge cards below to move from explanation into concrete guided-learning or focus-mode steps:'
+                    : 'Use the scoped knowledge cards below to continue with focus mode or guided learning for the highest-signal nodes:',
             ...topTitles,
             ...(actionHints.length > 0
                 ? ['', 'Suggested follow-through from the current turn:', ...actionHints]
@@ -7745,6 +7802,7 @@ export class KnowledgeLearningPlatform implements KnowledgeLearningPlatformAPI {
     }
 
     private buildScopedConversationAssistantBlocks(params: {
+        message: string;
         citations: KnowledgeCitation[];
         knowledgePoints: AgentConversationKnowledgePoint[];
         recalledMemories: AgentConversationMemoryRecord[];
@@ -8212,6 +8270,7 @@ export class KnowledgeLearningPlatform implements KnowledgeLearningPlatformAPI {
             usedScope: traceScope,
         });
         const assistantBlocks = this.buildScopedConversationAssistantBlocks({
+            message,
             citations,
             knowledgePoints,
             recalledMemories,