diff --git a/apps/server/src/modules/ai/ai.controller.ts b/apps/server/src/modules/ai/ai.controller.ts
index 40a8933..4f43c7a 100644
--- a/apps/server/src/modules/ai/ai.controller.ts
+++ b/apps/server/src/modules/ai/ai.controller.ts
@@ -1,7 +1,13 @@
 import { Body, Controller, Post } from '@nestjs/common';
 import { AiService } from './ai.service';
 import { DecomposeDto } from './dto/decompose.dto';
-import type { AgentDecomposeResponse, AgentDecomposeError } from '@ftb/shared';
+import { RiskInterpretDto } from './dto/risk-interpret.dto';
+import type {
+  AgentDecomposeResponse,
+  AgentDecomposeError,
+  AgentRiskInterpretResponse,
+  AgentRiskInterpretError,
+} from '@ftb/shared';
 
 @Controller('ai')
 export class AiController {
@@ -11,4 +17,9 @@ export class AiController {
   async decompose(@Body() dto: DecomposeDto): Promise<AgentDecomposeResponse | AgentDecomposeError> {
     return this.aiService.decompose(dto);
   }
+
+  @Post('risk-interpret')
+  async interpretRisk(@Body() dto: RiskInterpretDto): Promise<AgentRiskInterpretResponse | AgentRiskInterpretError> {
+    return this.aiService.interpretRisk(dto);
+  }
 }
diff --git a/apps/server/src/modules/ai/ai.service.spec.ts b/apps/server/src/modules/ai/ai.service.spec.ts
index 7e57b79..957ff29 100644
--- a/apps/server/src/modules/ai/ai.service.spec.ts
+++ b/apps/server/src/modules/ai/ai.service.spec.ts
@@ -1,11 +1,230 @@
 import { AiService } from './ai.service';
 import type { AiGatewayService } from './ai-gateway.service';
 import { DECOMPOSE_SYSTEM_PROMPT, DECOMPOSE_TOOL_INPUT_SCHEMA } from './prompts/decompose';
+import { RISK_INTERPRET_TOOL_INPUT_SCHEMA } from './prompts/risk-interpret';
+import { plainToInstance } from 'class-transformer';
+import { validateSync } from 'class-validator';
+import { RiskInterpretDto } from './dto/risk-interpret.dto';
+import type { AgentRiskInterpretRequest } from '@ftb/shared';
 
 const missingToolUseMessage =
   'Anthropic 未通过 tool_use 返回结果；stop_reason=tool_use；content=text("⚠️ 上游模型未返回任何内容。可能原因：触发了安全策略、上游限流、或模型对当前输入直接结束。")';
 
+function createRiskRequest(patch: Partial<AgentRiskInterpretRequest> = {}): AgentRiskInterpretRequest {
+  return {
+    versionId: 'ver-1',
+    versionName: 'V1.0',
+    riskScore: 70,
+    riskLevel: 'at_risk',
+    delayDays: 1,
+    confidence: 80,
+    signals: {
+      unfinishedCount: 4,
+      openBugCount: 3,
+      criticalBugCount: 3,
+      failedTestCount: 1,
+      blockedCount: 0,
+      silentRiskCount: 0,
+      daysToExpectedRelease: 1,
+    },
+    trendSummary: 'risk rising',
+    reasons: [{ key: 'critical_bug', label: 'Critical Bug', severity: 'critical', detail: '3 P1 bugs' }],
+    silentRisks: [],
+    dailyEvidence: {
+      todayDeliveries: [],
+      todayProgress: [],
+      todayRisks: ['tests failed'],
+      progressNotes: [],
+      needsProgressItems: [],
+      recentActivityCount: 1,
+    },
+    ...patch,
+  };
+}
+
 describe('AiService', () => {
+  it('interprets version risk through the active provider', async () => {
+    const callTool = jest.fn().mockResolvedValue({
+      toolName: 'submit_risk_interpretation',
+      toolInput: {
+        summary: '预计延期',
+        why: ['P1 Bug 增加'],
+        forecast: '预计 2026-07-04 可发',
+        recommendedReleaseWindow: '2026-07-04 之后',
+        suggestedActions: ['优先修复 P1 Bug'],
+        ownerHints: ['测试负责人同步失败用例'],
+        generatedAt: '2026-06-29T01:00:00.000Z',
+      },
+      rawModel: 'test-model',
+      inputTokens: 10,
+      outputTokens: 20,
+    });
+    const gateway = {
+      getActiveProvider: jest.fn().mockResolvedValue({ callTool }),
+      getActiveModel: jest.fn().mockResolvedValue('test-model'),
+    } as unknown as AiGatewayService;
+    const service = new AiService(gateway);
+
+    const result = await service.interpretRisk({
+      versionId: 'ver-1',
+      versionName: 'V1.0',
+      riskScore: 70,
+      riskLevel: 'at_risk',
+      delayDays: 1,
+      confidence: 80,
+      signals: {
+        unfinishedCount: 4,
+        openBugCount: 3,
+        criticalBugCount: 3,
+        failedTestCount: 1,
+        blockedCount: 0,
+        silentRiskCount: 0,
+        daysToExpectedRelease: 1,
+      },
+      trendSummary: '风险上升',
+      reasons: [{ key: 'critical_bug', label: '关键 Bug', severity: 'critical', detail: '3 个 P1 Bug' }],
+      silentRisks: [],
+      dailyEvidence: {
+        todayDeliveries: [],
+        todayProgress: [],
+        todayRisks: ['测试不通过'],
+        progressNotes: [],
+        needsProgressItems: [],
+        recentActivityCount: 1,
+      },
+    });
+
+    expect(result.ok).toBe(true);
+    if (result.ok) expect(result.result.summary).toBe('预计延期');
+    expect(callTool).toHaveBeenCalledWith(expect.objectContaining({
+      forceTool: true,
+      maxTokens: 3000,
+      tool: expect.objectContaining({ name: 'submit_risk_interpretation' }),
+    }));
+    expect(callTool.mock.calls[0][0].userPrompt).toContain('风险上升');
+  });
+
+  it('returns PARSE_ERROR when risk provider returns malformed payload', async () => {
+    const gateway = {
+      getActiveProvider: jest.fn().mockResolvedValue({
+        callTool: jest.fn().mockResolvedValue({
+          toolName: 'submit_risk_interpretation',
+          toolInput: null,
+          rawModel: 'test-model',
+          inputTokens: 10,
+          outputTokens: 20,
+        }),
+      }),
+      getActiveModel: jest.fn().mockResolvedValue('test-model'),
+    } as unknown as AiGatewayService;
+    const service = new AiService(gateway);
+
+    const result = await service.interpretRisk(createRiskRequest());
+
+    expect(result).toMatchObject({ ok: false, code: 'PARSE_ERROR' });
+  });
+
+  it('returns PARSE_ERROR when risk provider returns the wrong tool name', async () => {
+    const gateway = {
+      getActiveProvider: jest.fn().mockResolvedValue({
+        callTool: jest.fn().mockResolvedValue({
+          toolName: 'submit_decompose',
+          toolInput: {
+            summary: 'wrong tool',
+            why: ['bad'],
+            forecast: 'bad',
+            suggestedActions: ['bad'],
+            ownerHints: ['bad'],
+            generatedAt: '2026-06-29T01:00:00.000Z',
+          },
+          rawModel: 'test-model',
+          inputTokens: 10,
+          outputTokens: 20,
+        }),
+      }),
+      getActiveModel: jest.fn().mockResolvedValue('test-model'),
+    } as unknown as AiGatewayService;
+    const service = new AiService(gateway);
+
+    const result = await service.interpretRisk(createRiskRequest());
+
+    expect(result).toMatchObject({ ok: false, code: 'PARSE_ERROR' });
+  });
+
+  it('returns PARSE_ERROR when risk insight arrays contain non-string values', async () => {
+    const gateway = {
+      getActiveProvider: jest.fn().mockResolvedValue({
+        callTool: jest.fn().mockResolvedValue({
+          toolName: 'submit_risk_interpretation',
+          toolInput: {
+            summary: 'bad array',
+            why: [123],
+            forecast: 'bad',
+            suggestedActions: ['bad'],
+            ownerHints: ['bad'],
+            generatedAt: '2026-06-29T01:00:00.000Z',
+          },
+          rawModel: 'test-model',
+          inputTokens: 10,
+          outputTokens: 20,
+        }),
+      }),
+      getActiveModel: jest.fn().mockResolvedValue('test-model'),
+    } as unknown as AiGatewayService;
+    const service = new AiService(gateway);
+
+    const result = await service.interpretRisk(createRiskRequest());
+
+    expect(result).toMatchObject({ ok: false, code: 'PARSE_ERROR' });
+  });
+
+  it('validates risk daily evidence array items as strings', () => {
+    const dto = plainToInstance(RiskInterpretDto, {
+      ...createRiskRequest(),
+      dailyEvidence: {
+        todayDeliveries: [],
+        todayProgress: [],
+        todayRisks: [123],
+        progressNotes: [],
+        needsProgressItems: [],
+        recentActivityCount: 1,
+      },
+    });
+
+    const errors = validateSync(dto);
+
+    expect(JSON.stringify(errors)).toContain('todayRisks');
+  });
+
+  it('requires risk signals and daily evidence objects in DTO validation', () => {
+    const { signals: _signals, dailyEvidence: _dailyEvidence, ...payload } = createRiskRequest();
+    const dto = plainToInstance(RiskInterpretDto, payload);
+
+    const errors = validateSync(dto);
+    const serialized = JSON.stringify(errors);
+
+    expect(serialized).toContain('signals');
+    expect(serialized).toContain('dailyEvidence');
+  });
+
+  it('forbids extra fields in risk interpretation tool schema', () => {
+    expect((RISK_INTERPRET_TOOL_INPUT_SCHEMA as any).additionalProperties).toBe(false);
+  });
+
+  it('returns PARSE_ERROR when provider reports invalid tool arguments', async () => {
+    const gateway = {
+      getActiveProvider: jest.fn().mockResolvedValue({
+        callTool: jest.fn().mockRejectedValue(new Error('OpenAI tool_call arguments is not valid JSON')),
+      }),
+      getActiveModel: jest.fn().mockResolvedValue('test-model'),
+    } as unknown as AiGatewayService;
+    const service = new AiService(gateway);
+
+    const result = await service.interpretRisk(createRiskRequest());
+
+    expect(result).toMatchObject({ ok: false, code: 'PARSE_ERROR' });
+  });
+
   it('retries once with shorter prototype context when Anthropic returns no tool_use', async () => {
     const callTool = jest
       .fn()
diff --git a/apps/server/src/modules/ai/ai.service.ts b/apps/server/src/modules/ai/ai.service.ts
index 655c8f4..c3684ba 100644
--- a/apps/server/src/modules/ai/ai.service.ts
+++ b/apps/server/src/modules/ai/ai.service.ts
@@ -6,6 +6,12 @@ import {
   DECOMPOSE_TOOL_DESCRIPTION,
   DECOMPOSE_TOOL_INPUT_SCHEMA,
 } from './prompts/decompose';
+import {
+  RISK_INTERPRET_SYSTEM_PROMPT,
+  RISK_INTERPRET_TOOL_NAME,
+  RISK_INTERPRET_TOOL_DESCRIPTION,
+  RISK_INTERPRET_TOOL_INPUT_SCHEMA,
+} from './prompts/risk-interpret';
 import { buildPrototypeContext } from './prototype-context';
 import type {
   AgentDecomposeRequest,
@@ -13,6 +19,10 @@ import type {
   AgentDecomposeError,
   AgentDecomposeResult,
   AgentDecomposeTarget,
+  AgentRiskInsight,
+  AgentRiskInterpretError,
+  AgentRiskInterpretRequest,
+  AgentRiskInterpretResponse,
 } from '@ftb/shared';
 
 const DECOMPOSE_CONTEXT_CHAR_STEPS = [1800, 1200, 800] as const;
@@ -129,6 +139,119 @@ export class AiService {
     };
   }
 
+  async interpretRisk(req: AgentRiskInterpretRequest): Promise<AgentRiskInterpretResponse | AgentRiskInterpretError> {
+    const startedAt = Date.now();
+    let provider;
+    let model: string;
+    try {
+      provider = await this.gateway.getActiveProvider();
+      model = await this.gateway.getActiveModel();
+    } catch (e: any) {
+      return {
+        ok: false,
+        error: e?.message || '未配置 AI 提供商',
+        code: 'NO_PROVIDER',
+      };
+    }
+
+    try {
+      const toolResp = await provider.callTool({
+        systemPrompt: RISK_INTERPRET_SYSTEM_PROMPT,
+        userPrompt: `请基于以下小宝预警规则结果生成解释，禁止编造新事实：\n${JSON.stringify(req, null, 2)}`,
+        tool: {
+          name: RISK_INTERPRET_TOOL_NAME,
+          description: RISK_INTERPRET_TOOL_DESCRIPTION,
+          inputSchema: RISK_INTERPRET_TOOL_INPUT_SCHEMA,
+        },
+        forceTool: true,
+        maxTokens: 3000,
+        model,
+      });
+      const result = this.parseRiskInsightToolResponse(toolResp);
+      if (!result) {
+        return {
+          ok: false,
+          error: 'AI 返回结构不符合预期',
+          code: 'PARSE_ERROR',
+        };
+      }
+      return {
+        ok: true,
+        result,
+        meta: {
+          model: toolResp.rawModel || model,
+          inputTokens: toolResp.inputTokens,
+          outputTokens: toolResp.outputTokens,
+          durationMs: Date.now() - startedAt,
+        },
+      };
+    } catch (e: any) {
+      if (this.isProviderParseError(e)) {
+        return {
+          ok: false,
+          error: e?.message || 'AI 返回结构不符合预期',
+          code: 'PARSE_ERROR',
+        };
+      }
+      this.logger.error(`AI 风险解读调用失败: ${e.message}`);
+      return {
+        ok: false,
+        error: `AI 服务调用失败：${e.message}`,
+        code: 'API_ERROR',
+      };
+    }
+  }
+
+  private parseRiskInsightToolResponse(toolResp: any): AgentRiskInsight | null {
+    if (!toolResp || toolResp.toolName !== RISK_INTERPRET_TOOL_NAME || !this.isRecord(toolResp.toolInput)) {
+      return null;
+    }
+
+    const input = toolResp.toolInput;
+    const summary = this.readString(input.summary);
+    const forecast = this.readString(input.forecast);
+    const generatedAt = this.readString(input.generatedAt);
+    const why = this.readStringArray(input.why);
+    const suggestedActions = this.readStringArray(input.suggestedActions);
+    const ownerHints = this.readStringArray(input.ownerHints);
+    if (!summary || !forecast || !generatedAt || !why || !suggestedActions || !ownerHints) {
+      return null;
+    }
+
+    const insight: AgentRiskInsight = {
+      summary,
+      why,
+      forecast,
+      suggestedActions,
+      ownerHints,
+      generatedAt,
+    };
+    if (input.recommendedReleaseWindow !== undefined) {
+      const recommendedReleaseWindow = this.readString(input.recommendedReleaseWindow);
+      if (!recommendedReleaseWindow) return null;
+      insight.recommendedReleaseWindow = recommendedReleaseWindow;
+    }
+    return insight;
+  }
+
+  private readString(value: unknown): string | null {
+    return typeof value === 'string' ? value : null;
+  }
+
+  private readStringArray(value: unknown): string[] | null {
+    if (!Array.isArray(value) || value.some((item) => typeof item !== 'string')) return null;
+    return value;
+  }
+
+  private isRecord(value: unknown): value is Record<string, unknown> {
+    return typeof value === 'object' && value !== null && !Array.isArray(value);
+  }
+
+  private isProviderParseError(error: any): boolean {
+    const message = String(error?.message || '');
+    return message.includes('tool_call arguments') && message.includes('JSON');
+  }
+
   private async fetchPrototype(url: string): Promise<string> {
     if (!url || !url.startsWith('http')) {
       throw new Error('原型链接无效');
diff --git a/apps/server/src/modules/ai/dto/risk-interpret.dto.ts b/apps/server/src/modules/ai/dto/risk-interpret.dto.ts
new file mode 100644
index 0000000..093eeb0
--- /dev/null
+++ b/apps/server/src/modules/ai/dto/risk-interpret.dto.ts
@@ -0,0 +1,142 @@
+import { Type } from 'class-transformer';
+import { IsArray, IsDefined, IsIn, IsNumber, IsObject, IsOptional, IsString, ValidateNested } from 'class-validator';
+
+class RiskReasonDto {
+  @IsString()
+  key!: string;
+
+  @IsString()
+  label!: string;
+
+  @IsIn(['low', 'medium', 'high', 'critical'])
+  severity!: 'low' | 'medium' | 'high' | 'critical';
+
+  @IsString()
+  detail!: string;
+}
+
+class SilentRiskDto {
+  @IsString()
+  key!: string;
+
+  @IsOptional()
+  @IsNumber()
+  days?: number;
+
+  @IsString()
+  detail!: string;
+}
+
+class RiskDailyEvidenceDto {
+  @IsArray()
+  @IsString({ each: true })
+  todayDeliveries!: string[];
+
+  @IsArray()
+  @IsString({ each: true })
+  todayProgress!: string[];
+
+  @IsArray()
+  @IsString({ each: true })
+  todayRisks!: string[];
+
+  @IsArray()
+  @IsString({ each: true })
+  progressNotes!: string[];
+
+  @IsArray()
+  @IsString({ each: true })
+  needsProgressItems!: string[];
+
+  @IsNumber()
+  recentActivityCount!: number;
+
+  @IsOptional()
+  @IsString()
+  lastActivityAt?: string;
+}
+
+class RiskSignalsDto {
+  @IsNumber()
+  unfinishedCount!: number;
+
+  @IsNumber()
+  openBugCount!: number;
+
+  @IsNumber()
+  criticalBugCount!: number;
+
+  @IsNumber()
+  failedTestCount!: number;
+
+  @IsNumber()
+  blockedCount!: number;
+
+  @IsNumber()
+  silentRiskCount!: number;
+
+  @IsOptional()
+  @IsNumber()
+  daysToExpectedRelease?: number;
+}
+
+export class RiskInterpretDto {
+  @IsString()
+  versionId!: string;
+
+  @IsString()
+  versionName!: string;
+
+  @IsOptional()
+  @IsString()
+  productName?: string;
+
+  @IsOptional()
+  @IsString()
+  projectName?: string;
+
+  @IsNumber()
+  riskScore!: number;
+
+  @IsIn(['on_track', 'attention', 'at_risk', 'likely_delayed', 'blocked'])
+  riskLevel!: 'on_track' | 'attention' | 'at_risk' | 'likely_delayed' | 'blocked';
+
+  @IsOptional()
+  @IsString()
+  expectedReleaseDate?: string;
+
+  @IsOptional()
+  @IsString()
+  forecastReleaseDate?: string;
+
+  @IsNumber()
+  delayDays!: number;
+
+  @IsNumber()
+  confidence!: number;
+
+  @IsDefined()
+  @IsObject()
+  @ValidateNested()
+  @Type(() => RiskSignalsDto)
+  signals!: RiskSignalsDto;
+
+  @IsString()
+  trendSummary!: string;
+
+  @IsArray()
+  @ValidateNested({ each: true })
+  @Type(() => RiskReasonDto)
+  reasons!: RiskReasonDto[];
+
+  @IsArray()
+  @ValidateNested({ each: true })
+  @Type(() => SilentRiskDto)
+  silentRisks!: SilentRiskDto[];
+
+  @IsDefined()
+  @IsObject()
+  @ValidateNested()
+  @Type(() => RiskDailyEvidenceDto)
+  dailyEvidence!: RiskDailyEvidenceDto;
+}
diff --git a/apps/server/src/modules/ai/prompts/risk-interpret.ts b/apps/server/src/modules/ai/prompts/risk-interpret.ts
new file mode 100644
index 0000000..7f6216e
--- /dev/null
+++ b/apps/server/src/modules/ai/prompts/risk-interpret.ts
@@ -0,0 +1,32 @@
+export const RISK_INTERPRET_TOOL_NAME = 'submit_risk_interpretation';
+export const RISK_INTERPRET_TOOL_DESCRIPTION = '返回小宝预警的结构化解释';
+
+export const RISK_INTERPRET_SYSTEM_PROMPT = `你是 FTB 项目管理系统中的“小宝预警”解释助手。
+
+你只能解释系统给出的规则预测结果和证据，不能编造不存在的数据。
+你需要用项目经理能理解的中文说明：
+1. 当前能不能按期发版；
+2. 为什么有风险；
+3. 预计延期或建议发版窗口；
+4. 需要优先处理的动作；
+5. 哪些负责人或角色需要关注。
+
+输入中的 signals 是规则引擎压缩后的结构化风险信号，优先用它判断 Bug、测试失败、阻塞、静默风险和临近发版的变化。
+如果 confidence 低，必须提醒“预测可信度较低，需补充数据”。
+
+输出必须通过 submit_risk_interpretation 工具返回，不要输出自然语言正文或 Markdown。`;
+
+export const RISK_INTERPRET_TOOL_INPUT_SCHEMA = {
+  type: 'object',
+  properties: {
+    summary: { type: 'string' },
+    why: { type: 'array', items: { type: 'string' } },
+    forecast: { type: 'string' },
+    recommendedReleaseWindow: { type: 'string' },
+    suggestedActions: { type: 'array', items: { type: 'string' } },
+    ownerHints: { type: 'array', items: { type: 'string' } },
+    generatedAt: { type: 'string' },
+  },
+  required: ['summary', 'why', 'forecast', 'suggestedActions', 'ownerHints', 'generatedAt'],
+  additionalProperties: false,
+};
diff --git a/packages/shared/src/agent.ts b/packages/shared/src/agent.ts
index 8dcaa5c..4676b8f 100644
--- a/packages/shared/src/agent.ts
+++ b/packages/shared/src/agent.ts
@@ -114,6 +114,74 @@ export interface AgentDecomposeError {
   code: 'PROTOTYPE_FETCH_FAILED' | 'EMPTY_REQUIREMENTS' | 'API_ERROR' | 'PARSE_ERROR' | 'NO_PROVIDER' | 'UNKNOWN';
 }
 
+export interface AgentRiskReason {
+  key: string;
+  label: string;
+  severity: 'low' | 'medium' | 'high' | 'critical';
+  detail: string;
+}
+
+export interface AgentRiskInterpretRequest {
+  versionId: string;
+  versionName: string;
+  productName?: string;
+  projectName?: string;
+  riskScore: number;
+  riskLevel: 'on_track' | 'attention' | 'at_risk' | 'likely_delayed' | 'blocked';
+  expectedReleaseDate?: string | null;
+  forecastReleaseDate?: string;
+  delayDays: number;
+  confidence: number;
+  signals: {
+    unfinishedCount: number;
+    openBugCount: number;
+    criticalBugCount: number;
+    failedTestCount: number;
+    blockedCount: number;
+    silentRiskCount: number;
+    daysToExpectedRelease?: number;
+  };
+  trendSummary: string;
+  reasons: AgentRiskReason[];
+  silentRisks: Array<{ key: string; days?: number; detail: string }>;
+  dailyEvidence: {
+    todayDeliveries: string[];
+    todayProgress: string[];
+    todayRisks: string[];
+    progressNotes: string[];
+    needsProgressItems: string[];
+    recentActivityCount: number;
+    lastActivityAt?: string;
+  };
+}
+
+export interface AgentRiskInsight {
+  summary: string;
+  why: string[];
+  forecast: string;
+  recommendedReleaseWindow?: string;
+  suggestedActions: string[];
+  ownerHints: string[];
+  generatedAt: string;
+}
+
+export interface AgentRiskInterpretResponse {
+  ok: true;
+  result: AgentRiskInsight;
+  meta: {
+    model: string;
+    inputTokens: number;
+    outputTokens: number;
+    durationMs: number;
+  };
+}
+
+export interface AgentRiskInterpretError {
+  ok: false;
+  error: string;
+  code: 'API_ERROR' | 'PARSE_ERROR' | 'NO_PROVIDER' | 'UNKNOWN';
+}
+
 /* ─────────────────────────  AI Provider 配置  ─────────────────────────── */
 
 export type AiProviderFormat = 'anthropic' | 'openai';