agent-insight 0.3.0-beta → 0.3.1-beta

package/.next/standalone/.next/server/chunks/_73cb0d03._.js

@@ -108,7 +108,7 @@ ${e.traceSummaryText?.trim()||"（无可用轨迹）"}
 ## Skill 上下文
 ${s}
 
-请你自行逐条检查关键观点覆盖情况，并在不派发任何子代理的前提下完成任务完成度评测和关键观点根因分析。`}function E(e,t="skill-aware"){var a;let r,n="no-skill"===t?{...e,key_point_findings:(0,h.stripSkillAttributionFromKeyPointFindings)(e.key_point_findings)}:e,s=(0,h.deriveTaskCompletionScoreFromFindings)(n.key_point_findings),i=s.score??0,o=i>=.8,l=String(e.reason||"").trim()||"任务完成度评测已完成，但未返回理由。",u=void 0===e.score?void 0:(a=e.score,r="number"==typeof a?a:"string"==typeof a?Number(a):NaN,Number.isFinite(r)?Math.max(0,Math.min(1,r)):0),{result_issues:c,resultIssues:d,...p}=n;return{isCorrect:o,score:i,reason:l,rawAnalysis:{...p,score:i,is_correct:o,llm_reported_score:u,score_computation:{method:"equal_weight_average",item_count:s.itemCount},skillAttributionMode:t,key_point_findings:s.findings}}}async function x(e,t,a,r,n){let s=new d.ChatOpenAI({apiKey:t.apiKey||"no-api-key",model:t.model||"deepseek-chat",configuration:{baseURL:t.baseUrl||"https://api.deepseek.com"},temperature:0,topP:1,timeout:18e4,maxRetries:2,modelKwargs:{seed:42}}),i=S(e,a),o=await s.invoke([new m.SystemMessage(v(e.skillAttributionMode||"skill-aware")),new p.HumanMessage(i)]),l="string"==typeof o.content?o.content:JSON.stringify(o.content),u=(0,w.parseLooseJson)(l);if(!u||!A(u))throw Error(`任务完成度直接 LLM 评测未产出有效 JSON。模型输出前 800 字符：${l.slice(0,800)}`);let y=E(u),h=`${R}-${(0,f.randomUUID)()}`,_=(0,c.findSystemAgentDefinition)("opencode",R);return await (0,g.recordDirectEvaluatorExecution)({taskId:h,agentName:R,user:n,query:e.caseInput,userMessage:i,assistantOutput:l,usage:function(e){let t=e?.usage_metadata;if(t&&("number"==typeof t.input_tokens||"number"==typeof t.output_tokens))return{input:t.input_tokens,output:t.output_tokens,total:t.total_tokens};let a=e?.response_metadata?.tokenUsage;return a&&("number"==typeof a.promptTokens||"number"==typeof a.completionTokens)?{input:a.promptTokens,output:a.completionTokens,total:a.totalTokens}:null}(o),modelID:t.model,skill:_?.traceSkill??null}).catch(e=>{console.warn("[opencode-task-completion] failed to record direct evaluator trace:",e?.message||e)}),{...y,rawAnalysis:{...y.rawAnalysis||{},evaluatorSessionId:h,root_cause_source:r}}}async function b(e,t,a,d){return(0,s.withBackgroundOpencodeSlot)(async()=>{if(!(0,g.shouldForceOpencodeEvalTransport)()){let{rootCauses:a,source:r}=await k(e,t),n=await (0,o.getActiveConfig)(t);if(!n)return{isCorrect:!1,score:0,reason:"请先在模型配置中激活一个评测模型，才能执行结果评测。",rawAnalysis:{root_cause_source:r}};try{return await x(e,n,a,r,t)}catch(e){console.warn("[opencode-task-completion] direct LLM path failed, falling back to opencode transport:",e?.message||e)}}return(0,n.runWithEphemeralOpencodeServer)({user:t||void 0,verbose:!1,isolateHome:!0},async a=>{let{rootCauses:n,source:s}=await k(e,t),d=e.skillAttributionMode||"skill-aware",p=await (0,o.getActiveConfig)(t);if(!p)return{isCorrect:!1,score:0,reason:"请先在模型配置中激活一个评测模型，才能执行结果评测。",rawAnalysis:{root_cause_source:s}};let m=t?await (0,l.loadServerModelForUser)(t):null,f=m?.providerID||(0,l.normalizeProviderID)(p.provider||(0,l.inferProviderFromBaseUrl)(p.baseUrl)),y=m?.modelID||p.model||"deepseek-chat",h={text:S(e,n),agent:"build",model:{providerID:f,modelID:y,apiKey:m?.apiKey||p.apiKey,baseURL:m?.baseURL||p.baseUrl},modelOptions:{temperature:0,top_p:1,seed:42},system:v(d),permission:(0,i.buildEvaluatorPermissions)()},_="",x=null,b="",I=null,C=null;try{C=new r.AgentInsight({baseURL:a,logLevel:"warn"});let n=await C.createSession({title:`${R}-${Date.now()}`,directory:"/tmp"}),i=String(n?.id||n?.ID||"");if(!i)throw Error("Failed to create opencode session for task completion evaluation");b=i;let o=await (0,c.getSystemAgentId)("opencode",R),l=(0,c.findSystemAgentDefinition)("opencode",R);if((0,u.tagOpencodeSession)(i,{agentName:R,agentId:o,skill:l?.traceSkill,displayQuery:e.caseInput,user:t||void 0}),_=(await C.chat(i,h,{onText:e=>{_+=e.delta},onError:e=>{x=e},onSubagent:e=>{I=e.agent||e.sessionID||"unknown-subagent",console.warn(`[opencode-task-completion] unexpected subagent spawned: ${I}`)}},{streamTimeoutMs:6e5,idleTimeoutMs:18e4})).text||_,await (0,g.recordEvaluatorExecution)(C,{taskId:i,agentName:R,user:t,query:e.caseInput}),I)throw Error(`任务完成度评估器不允许派发子代理，但实际派发了：${I}`);let p=(0,w.parseLooseJson)(_);if(p&&A(p)){let e=E(p,d);return{...e,rawAnalysis:{...e.rawAnalysis||{},evaluatorSessionId:b,root_cause_source:s}}}}catch(e){x=e instanceof Error?e:Error(String(e))}if(b&&C)try{await (0,g.recordEvaluatorExecution)(C,{taskId:b,agentName:R,user:t,query:e.caseInput})}catch(e){console.warn("[opencode-task-completion] failed to persist evaluator execution:",e?.message||e)}if(I)throw x||Error(`任务完成度评估器不允许派发子代理，但实际派发了：${I}`);let N=function(e,...t){for(let a of t){let t=(0,w.parseLooseJson)(String(a||""));if(t&&A(t))return E(t,e)}return null}(d,_,x?.message);if(N)return{...N,rawAnalysis:{...N.rawAnalysis||{},evaluatorSessionId:b||void 0,root_cause_source:s}};let $=x?.message||`Agent 输出前 800 字符：${_.slice(0,800)}`;throw Error(`任务完成度评估器未产出有效 JSON。opencode 评测失败：${$}`)})},{taskType:"task-completion-eval",user:t??void 0,skill:a??void 0,skillVersion:d??null,label:`task-completion: ${(e.caseInput||"").slice(0,40)}`,silent:!0})}new Date().toISOString(),v("skill-aware"),v("no-skill"),e.s(["evaluateTaskCompletionViaOpencode",()=>b]),a()}catch(e){a(e)}},!1),508439,e=>e.a(async(t,a)=>{try{e.i(889228);var r=e.i(88045),n=e.i(322632),s=e.i(637351);e.i(384660);var i=e.i(106411),o=t([i]);[i]=o.then?(await o)():o;let f=n.z.object({normalized_input:n.z.string().default(""),confidence:n.z.number().min(0).max(1).default(0),ignored_parts:n.z.array(n.z.string()).default([]),reason:n.z.string().default("")}),g=n.z.object({best_case_id:n.z.string().nullable().default(null),best_confidence:n.z.number().min(0).max(1).default(0),reason:n.z.string().default("")});function l(e){let t=e.trim(),a=t.match(/```(?:json)?\s*([\s\S]*?)\s*```/i);if(a)t=a[1];else{let e=t.indexOf("{"),a=t.lastIndexOf("}");-1!==e&&-1!==a&&a>=e&&(t=t.substring(e,a+1))}return JSON.parse(t)}async function u(e){var t;let a,n=await (0,i.getActiveConfig)(e);if(!n)return{client:null,model:null};let o=n.apiKey||"no-api-key-required",l=(t=n.baseUrl||"https://api.deepseek.com",String(t||"").replace(/\/chat\/completions\/?$/,"")),{customFetch:u}=(0,s.getProxyConfig)();return{client:new r.OpenAI({apiKey:o,baseURL:l,fetch:u,timeout:(a=Number(process.env.SEMANTIC_MATCH_TIMEOUT_MS||process.env.JUDGMENT_TIMEOUT_MS||3e5),Number.isFinite(a)&&a>0?a:3e5)}),model:n.model||"deepseek-chat"}}async function c(e){let{client:t,model:a}=await u(e);if(!t||!a)return"No active evaluation model configured";try{let e=await t.chat.completions.create({messages:[{role:"user",content:"Hi"}],model:a,max_tokens:5});if(e?.choices)return null;return"No response from model"}catch(a){let e,t;return e=a?.message||"LLM request failed",t=a?.cause?.message?` (${a.cause.message})`:"",`${e}${t}`}}async function d(e,t){let a={normalized_input:String(e||"").trim(),confidence:0,ignored_parts:[],reason:"LLM extraction unavailable, fallback to raw input"};if(!a.normalized_input)return a;let{client:r,model:n}=await u(t);if(!r||!n)return a;try{var s;let e=await r.chat.completions.create({model:n,temperature:0,top_p:1,seed:42,messages:[{role:"user",content:(s=a.normalized_input,`你是一个 agent 评测系统中的“真实用户输入抽取器”。
+请你自行逐条检查关键观点覆盖情况，并在不派发任何子代理的前提下完成任务完成度评测和关键观点根因分析。`}function E(e,t="skill-aware"){var a;let r,n="no-skill"===t?{...e,key_point_findings:(0,h.stripSkillAttributionFromKeyPointFindings)(e.key_point_findings)}:e,s=(0,h.deriveTaskCompletionScoreFromFindings)(n.key_point_findings),i=s.score??0,o=i>=.8,l=String(e.reason||"").trim()||"任务完成度评测已完成，但未返回理由。",u=void 0===e.score?void 0:(a=e.score,r="number"==typeof a?a:"string"==typeof a?Number(a):NaN,Number.isFinite(r)?Math.max(0,Math.min(1,r)):0),{result_issues:c,resultIssues:d,...p}=n;return{isCorrect:o,score:i,reason:l,rawAnalysis:{...p,score:i,is_correct:o,llm_reported_score:u,score_computation:{method:"equal_weight_average",item_count:s.itemCount},skillAttributionMode:t,key_point_findings:s.findings}}}async function x(e,t,a,r,n){let s=new d.ChatOpenAI({apiKey:t.apiKey||"no-api-key",model:t.model||"deepseek-chat",configuration:{baseURL:t.baseUrl||"https://api.deepseek.com"},temperature:0,topP:1,timeout:18e4,maxRetries:2,modelKwargs:{seed:42}}),i=S(e,a),o=v(e.skillAttributionMode||"skill-aware"),l=await s.invoke([new m.SystemMessage(o),new p.HumanMessage(i)]),u="string"==typeof l.content?l.content:JSON.stringify(l.content),y=(0,w.parseLooseJson)(u);if(!y||!A(y))throw Error(`任务完成度直接 LLM 评测未产出有效 JSON。模型输出前 800 字符：${u.slice(0,800)}`);let h=E(y),_=`${R}-${(0,f.randomUUID)()}`,k=(0,c.findSystemAgentDefinition)("opencode",R);return await (0,g.recordDirectEvaluatorExecution)({taskId:_,agentName:R,user:n,query:e.caseInput,systemPrompt:o,userMessage:i,assistantOutput:u,usage:function(e){let t=e?.usage_metadata;if(t&&("number"==typeof t.input_tokens||"number"==typeof t.output_tokens))return{input:t.input_tokens,output:t.output_tokens,total:t.total_tokens};let a=e?.response_metadata?.tokenUsage;return a&&("number"==typeof a.promptTokens||"number"==typeof a.completionTokens)?{input:a.promptTokens,output:a.completionTokens,total:a.totalTokens}:null}(l),modelID:t.model,skill:k?.traceSkill??null}).catch(e=>{console.warn("[opencode-task-completion] failed to record direct evaluator trace:",e?.message||e)}),{...h,rawAnalysis:{...h.rawAnalysis||{},evaluatorSessionId:_,root_cause_source:r}}}async function b(e,t,a,d){return(0,s.withBackgroundOpencodeSlot)(async()=>{if(!(0,g.shouldForceOpencodeEvalTransport)()){let{rootCauses:a,source:r}=await k(e,t),n=await (0,o.getActiveConfig)(t);if(!n)return{isCorrect:!1,score:0,reason:"请先在模型配置中激活一个评测模型，才能执行结果评测。",rawAnalysis:{root_cause_source:r}};try{return await x(e,n,a,r,t)}catch(e){console.warn("[opencode-task-completion] direct LLM path failed, falling back to opencode transport:",e?.message||e)}}return(0,n.runWithEphemeralOpencodeServer)({user:t||void 0,verbose:!1,isolateHome:!0},async a=>{let{rootCauses:n,source:s}=await k(e,t),d=e.skillAttributionMode||"skill-aware",p=await (0,o.getActiveConfig)(t);if(!p)return{isCorrect:!1,score:0,reason:"请先在模型配置中激活一个评测模型，才能执行结果评测。",rawAnalysis:{root_cause_source:s}};let m=t?await (0,l.loadServerModelForUser)(t):null,f=m?.providerID||(0,l.normalizeProviderID)(p.provider||(0,l.inferProviderFromBaseUrl)(p.baseUrl)),y=m?.modelID||p.model||"deepseek-chat",h={text:S(e,n),agent:"build",model:{providerID:f,modelID:y,apiKey:m?.apiKey||p.apiKey,baseURL:m?.baseURL||p.baseUrl},modelOptions:{temperature:0,top_p:1,seed:42},system:v(d),permission:(0,i.buildEvaluatorPermissions)()},_="",x=null,b="",I=null,C=null;try{C=new r.AgentInsight({baseURL:a,logLevel:"warn"});let n=await C.createSession({title:`${R}-${Date.now()}`,directory:"/tmp"}),i=String(n?.id||n?.ID||"");if(!i)throw Error("Failed to create opencode session for task completion evaluation");b=i;let o=await (0,c.getSystemAgentId)("opencode",R),l=(0,c.findSystemAgentDefinition)("opencode",R);if((0,u.tagOpencodeSession)(i,{agentName:R,agentId:o,skill:l?.traceSkill,displayQuery:e.caseInput,user:t||void 0}),_=(await C.chat(i,h,{onText:e=>{_+=e.delta},onError:e=>{x=e},onSubagent:e=>{I=e.agent||e.sessionID||"unknown-subagent",console.warn(`[opencode-task-completion] unexpected subagent spawned: ${I}`)}},{streamTimeoutMs:6e5,idleTimeoutMs:18e4})).text||_,await (0,g.recordEvaluatorExecution)(C,{taskId:i,agentName:R,user:t,query:e.caseInput}),I)throw Error(`任务完成度评估器不允许派发子代理，但实际派发了：${I}`);let p=(0,w.parseLooseJson)(_);if(p&&A(p)){let e=E(p,d);return{...e,rawAnalysis:{...e.rawAnalysis||{},evaluatorSessionId:b,root_cause_source:s}}}}catch(e){x=e instanceof Error?e:Error(String(e))}if(b&&C)try{await (0,g.recordEvaluatorExecution)(C,{taskId:b,agentName:R,user:t,query:e.caseInput})}catch(e){console.warn("[opencode-task-completion] failed to persist evaluator execution:",e?.message||e)}if(I)throw x||Error(`任务完成度评估器不允许派发子代理，但实际派发了：${I}`);let N=function(e,...t){for(let a of t){let t=(0,w.parseLooseJson)(String(a||""));if(t&&A(t))return E(t,e)}return null}(d,_,x?.message);if(N)return{...N,rawAnalysis:{...N.rawAnalysis||{},evaluatorSessionId:b||void 0,root_cause_source:s}};let $=x?.message||`Agent 输出前 800 字符：${_.slice(0,800)}`;throw Error(`任务完成度评估器未产出有效 JSON。opencode 评测失败：${$}`)})},{taskType:"task-completion-eval",user:t??void 0,skill:a??void 0,skillVersion:d??null,label:`task-completion: ${(e.caseInput||"").slice(0,40)}`,silent:!0})}new Date().toISOString(),v("skill-aware"),v("no-skill"),e.s(["evaluateTaskCompletionViaOpencode",()=>b]),a()}catch(e){a(e)}},!1),508439,e=>e.a(async(t,a)=>{try{e.i(889228);var r=e.i(88045),n=e.i(322632),s=e.i(637351);e.i(384660);var i=e.i(106411),o=t([i]);[i]=o.then?(await o)():o;let f=n.z.object({normalized_input:n.z.string().default(""),confidence:n.z.number().min(0).max(1).default(0),ignored_parts:n.z.array(n.z.string()).default([]),reason:n.z.string().default("")}),g=n.z.object({best_case_id:n.z.string().nullable().default(null),best_confidence:n.z.number().min(0).max(1).default(0),reason:n.z.string().default("")});function l(e){let t=e.trim(),a=t.match(/```(?:json)?\s*([\s\S]*?)\s*```/i);if(a)t=a[1];else{let e=t.indexOf("{"),a=t.lastIndexOf("}");-1!==e&&-1!==a&&a>=e&&(t=t.substring(e,a+1))}return JSON.parse(t)}async function u(e){var t;let a,n=await (0,i.getActiveConfig)(e);if(!n)return{client:null,model:null};let o=n.apiKey||"no-api-key-required",l=(t=n.baseUrl||"https://api.deepseek.com",String(t||"").replace(/\/chat\/completions\/?$/,"")),{customFetch:u}=(0,s.getProxyConfig)();return{client:new r.OpenAI({apiKey:o,baseURL:l,fetch:u,timeout:(a=Number(process.env.SEMANTIC_MATCH_TIMEOUT_MS||process.env.JUDGMENT_TIMEOUT_MS||3e5),Number.isFinite(a)&&a>0?a:3e5)}),model:n.model||"deepseek-chat"}}async function c(e){let{client:t,model:a}=await u(e);if(!t||!a)return"No active evaluation model configured";try{let e=await t.chat.completions.create({messages:[{role:"user",content:"Hi"}],model:a,max_tokens:5});if(e?.choices)return null;return"No response from model"}catch(a){let e,t;return e=a?.message||"LLM request failed",t=a?.cause?.message?` (${a.cause.message})`:"",`${e}${t}`}}async function d(e,t){let a={normalized_input:String(e||"").trim(),confidence:0,ignored_parts:[],reason:"LLM extraction unavailable, fallback to raw input"};if(!a.normalized_input)return a;let{client:r,model:n}=await u(t);if(!r||!n)return a;try{var s;let e=await r.chat.completions.create({model:n,temperature:0,top_p:1,seed:42,messages:[{role:"user",content:(s=a.normalized_input,`你是一个 agent 评测系统中的“真实用户输入抽取器”。
 
 你的任务：
 从 raw_input 中抽取真正应该用于评测数据集匹配的用户输入。