stable-harness 0.0.69 → 0.0.71
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
|
@@ -1 +1 @@
|
|
|
1
|
-
export function hasPlanningEvidence(t){return t.some(t=>{const e=readAdapterEvent(t);return!!e&&("plan"===e.traceType||String(e.traceLabel??"").startsWith("plan.")||"write_todos"===e.toolId&&String(e.phase??"").startsWith("agent.tool."))})}export function successfulEvidenceToolIds(t){const e=t.flatMap(t=>{if("runtime.tool.direct.completed"===t.type)return[t.toolId];const e=readAdapterEvent(t);return isToolResultEvent(e)&&"string"==typeof e.toolId&&isSuccessfulEvidenceEvent(e)?[e.toolId]:[]});return[...new Set(e)]}export function successfulEvidenceOutputs(t){return successfulEvidenceItems(t).map(t=>t.output)}export function successfulEvidenceItems(t){return t.flatMap(t=>{if("runtime.tool.direct.completed"===t.type)return stringifyEvidence(t.output).map(e=>({source:t.toolId,output:e}));const e=readAdapterEvent(t);return isToolResultEvent(e)&&
|
|
1
|
+
export function hasPlanningEvidence(t){return t.some(t=>{const e=readAdapterEvent(t);return!!e&&("plan"===e.traceType||String(e.traceLabel??"").startsWith("plan.")||"write_todos"===e.toolId&&String(e.phase??"").startsWith("agent.tool."))})}export function successfulEvidenceToolIds(t){const e=t.flatMap(t=>{if("runtime.tool.direct.completed"===t.type)return[t.toolId];const e=readAdapterEvent(t);return isToolResultEvent(e)&&"string"==typeof e.toolId&&isSuccessfulEvidenceEvent(e)?[e.toolId]:[]});return[...new Set(e)]}export function successfulEvidenceOutputs(t){return successfulEvidenceItems(t).map(t=>t.output)}export function successfulEvidenceItems(t){return t.flatMap(t=>{if("runtime.tool.direct.completed"===t.type)return stringifyEvidence(t.output).map(e=>({source:t.toolId,output:e}));const e=readAdapterEvent(t),n=readToolSource(e);return isToolResultEvent(e)&&n?!isSuccessfulEvidenceEvent(e)||function isPlanningTool(t){return/^(?:write_todos|read_todos)$/u.test(t)}(n)?[]:stringifyEvidence(e.evidenceOutput??e.output).filter(t=>function isUsableEvidenceOutput(t,e){return"task"!==t||!function looksLikeUnexecutedToolIntent(t){const e=t.trim();return!(e.length>4e3)&&[/\b(?:I need to|I will|I'll|I am going to|I'm going to|Let me)\s+(?:call|use|run|invoke|get|fetch|check)\b[\s\S]{0,1200}\b(?:tool|function|system_time|web_search|url_text_fetch|task)\b/iu,/(?:我需要|我要|我会|我将|让我|我来|接下来我(?:会|将)?)\s*(?:先)?(?:调用|使用|运行|执行|检查|获取|查看)\b[\s\S]{0,1200}\b(?:工具|函数|system_time|web_search|url_text_fetch|task)\b/iu,/```(?:json)?\s*(?:system_time|web_search|url_text_fetch|task)\s*```/iu].some(t=>t.test(e))}(e)&&!function looksLikeDelegatedCommentary(t){const e=t.trim();return!(e.length>8e3)&&[/(?:这篇文章|this article)[\s\S]{0,1200}(?:我认为|我认同|打动我|值得深入探讨|特别认同|what I find|I think|I agree)/iu,/(?:你还有什么|还想深入探讨|would you like|do you want).{0,120}[?？]\s*$/iu].some(t=>t.test(e))}(e)}(n,t)).map(t=>({source:n,output:t})):[]})}export function controlBlockers(t){const e=function successfulEventIndexesBySource(t){const e=new Map;return t.forEach((t,n)=>{const r=function successfulEventSource(t){if("runtime.tool.direct.completed"===t.type)return t.toolId;const e=readAdapterEvent(t),n=readToolSource(e);return isToolResultEvent(e)&&n&&isSuccessfulEvidenceEvent(e)?n:void 0}(t);r&&e.set(r,[...e.get(r)??[],n])}),e}(t);return t.flatMap((t,n)=>{const r=readAdapterEvent(t),o=readEventStatus(r),s=readToolSource(r)??"tool";return o&&isBlockerStatus(o)?isResolvedByLaterCompletion(o)&&function completedAfter(t,e,n){return(t.get(e)??[]).some(t=>t>n)}(e,s,n)?[]:[`${s}:${o}`]:[]})}export function controlGaps(t){const e=new Set(successfulEvidenceItems(t).map(t=>t.source));return t.flatMap(t=>{const n=readAdapterEvent(t),r=readEventStatus(n),o=readToolSource(n)??"tool";return r&&isGapStatus(r)?e.has(o)&&isResolvedByLaterCompletion(r)?[]:[`${o}:${r}`]:[]})}function stringifyEvidence(t){return"string"==typeof t?t.trim()?[t]:[]:null==t?[]:[JSON.stringify(t)]}function readAdapterEvent(t){return"runtime.adapter.event"===t.type&&isRecord(t.event)?t.event:void 0}function isSuccessfulEvidenceEvent(t){const e=readEventStatus(t);return!e||/^(?:completed|success|ok|recorded)$/iu.test(e)}function isBlockerStatus(t){return/^(?:blocked|approval_required|schema_repair_failed|tool_argument_error|invalid_input)$/iu.test(t)}function isGapStatus(t){return/^(?:dependency_required|plan_required|repeated_tool_call_limit|duplicate_tool_call)$/iu.test(t)}function isResolvedByLaterCompletion(t){return isGapStatus(t)||isBlockerStatus(t)}function isToolResultEvent(t){return"deepagents.tool_execution.result"===t?.eventType||"agent.tool.result"===t?.phase}function isRecord(t){return"object"==typeof t&&null!==t&&!Array.isArray(t)}function readString(t){return"string"==typeof t&&t.trim()?t:void 0}function readEventStatus(t){return readString(t?.controlStatus)??readString(t?.status)??function readOutputStatus(t){if(isRecord(t)&&"string"==typeof t.status)return t.status;if("string"==typeof t){const e=function parseJsonRecord(t){try{const e=JSON.parse(t);return isRecord(e)?e:void 0}catch{return}}(t);return"string"==typeof e?.status?e.status:t.match(/^Status:\s*([A-Za-z0-9_-]+)/imu)?.[1]}}(t?.output)}function readToolSource(t){return readString(t?.toolId)??readString(t?.toolName)??readString(t?.name)}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
export function completeRun(
|
|
1
|
+
import{controlGaps as t}from"../quality/event-evidence.js";export function completeRun(e){const r=e.store.getRun(e.requestId);if("cancelled"===r?.state)return response(e,"cancelled",r.artifacts);!function assertDeliverableOutput(e,r){if(!e.text.trim())throw new Error("runtime_empty_output: Runtime request produced no user-facing output.");!function assertNoOmittedControlGaps(e,r){if(function resultWasSynthesizedFromEvidence(t){const e=t.metadata?.toolCallRecovery;return"object"==typeof e&&null!==e&&"synthesized"in e&&!0===e.synthesized}(e))return;const s=t(r).filter(t=>!function mentionsGap(t,e){const[r,s]=e.split(":"),n=t.toLowerCase();return n.includes(e.toLowerCase())||Boolean(r&&s&&n.includes(r.toLowerCase())&&n.includes(s.toLowerCase()))}(e.text,t));if(0!==s.length)throw new Error(`runtime_unresolved_control_gap: Runtime request produced unresolved control gap(s): ${s.slice(0,8).join(", ")}`)}(e,r)}(e.result,r?.events??[]);const s=(e.result.artifacts??[]).map(toPublicArtifact);r&&e.store.updateRun(e.requestId,{state:"completed",output:e.result.text,metadata:{...r.metadata,...e.result.metadata},artifacts:mergeArtifacts(r.artifacts,s),completedAt:(new Date).toISOString()});for(const t of e.result.artifacts??[])e.artifacts?.createArtifact({...t,requestId:e.requestId,sessionId:e.sessionId,agentId:e.agent.id});for(const t of s)e.emit({type:"runtime.artifact.created",requestId:e.requestId,sessionId:e.sessionId,agentId:e.agent.id,artifact:t});return e.emit({type:"runtime.request.completed",requestId:e.requestId,sessionId:e.sessionId,agentId:e.agent.id,output:e.result.text}),response(e,"completed",e.store.getRun(e.requestId)?.artifacts)}export function failRun(t){const e=t.error instanceof Error?t.error.message:String(t.error);return t.store.getRun(t.requestId)&&t.store.updateRun(t.requestId,{state:"failed",error:e,completedAt:(new Date).toISOString()}),t.emit({type:"runtime.request.failed",requestId:t.requestId,sessionId:t.sessionId,agentId:t.agent.id,error:e}),{requestId:t.requestId,sessionId:t.sessionId,agentId:t.agent.id,state:"failed",output:e}}function response(t,e,r){return{requestId:t.requestId,sessionId:t.sessionId,agentId:t.agent.id,state:e,output:t.result.text,metadata:t.result.metadata,artifacts:r??t.result.artifacts}}function mergeArtifacts(t,e){const r=new Map;for(const s of[...t??[],...e??[]])r.set(s.id,s);return[...r.values()]}function toPublicArtifact(t){return{id:t.id,kind:t.kind,...t.uri?{uri:t.uri}:{},...t.metadata?{metadata:t.metadata}:{}}}
|
package/package.json
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
export function hasPlanningEvidence(t){return t.some(t=>{const e=readAdapterEvent(t);return!!e&&("plan"===e.traceType||String(e.traceLabel??"").startsWith("plan.")||"write_todos"===e.toolId&&String(e.phase??"").startsWith("agent.tool."))})}export function successfulEvidenceToolIds(t){const e=t.flatMap(t=>{if("runtime.tool.direct.completed"===t.type)return[t.toolId];const e=readAdapterEvent(t);return isToolResultEvent(e)&&"string"==typeof e.toolId&&isSuccessfulEvidenceEvent(e)?[e.toolId]:[]});return[...new Set(e)]}export function successfulEvidenceOutputs(t){return successfulEvidenceItems(t).map(t=>t.output)}export function successfulEvidenceItems(t){return t.flatMap(t=>{if("runtime.tool.direct.completed"===t.type)return stringifyEvidence(t.output).map(e=>({source:t.toolId,output:e}));const e=readAdapterEvent(t);return isToolResultEvent(e)&&
|
|
1
|
+
export function hasPlanningEvidence(t){return t.some(t=>{const e=readAdapterEvent(t);return!!e&&("plan"===e.traceType||String(e.traceLabel??"").startsWith("plan.")||"write_todos"===e.toolId&&String(e.phase??"").startsWith("agent.tool."))})}export function successfulEvidenceToolIds(t){const e=t.flatMap(t=>{if("runtime.tool.direct.completed"===t.type)return[t.toolId];const e=readAdapterEvent(t);return isToolResultEvent(e)&&"string"==typeof e.toolId&&isSuccessfulEvidenceEvent(e)?[e.toolId]:[]});return[...new Set(e)]}export function successfulEvidenceOutputs(t){return successfulEvidenceItems(t).map(t=>t.output)}export function successfulEvidenceItems(t){return t.flatMap(t=>{if("runtime.tool.direct.completed"===t.type)return stringifyEvidence(t.output).map(e=>({source:t.toolId,output:e}));const e=readAdapterEvent(t),n=readToolSource(e);return isToolResultEvent(e)&&n?!isSuccessfulEvidenceEvent(e)||function isPlanningTool(t){return/^(?:write_todos|read_todos)$/u.test(t)}(n)?[]:stringifyEvidence(e.evidenceOutput??e.output).filter(t=>function isUsableEvidenceOutput(t,e){return"task"!==t||!function looksLikeUnexecutedToolIntent(t){const e=t.trim();return!(e.length>4e3)&&[/\b(?:I need to|I will|I'll|I am going to|I'm going to|Let me)\s+(?:call|use|run|invoke|get|fetch|check)\b[\s\S]{0,1200}\b(?:tool|function|system_time|web_search|url_text_fetch|task)\b/iu,/(?:我需要|我要|我会|我将|让我|我来|接下来我(?:会|将)?)\s*(?:先)?(?:调用|使用|运行|执行|检查|获取|查看)\b[\s\S]{0,1200}\b(?:工具|函数|system_time|web_search|url_text_fetch|task)\b/iu,/```(?:json)?\s*(?:system_time|web_search|url_text_fetch|task)\s*```/iu].some(t=>t.test(e))}(e)&&!function looksLikeDelegatedCommentary(t){const e=t.trim();return!(e.length>8e3)&&[/(?:这篇文章|this article)[\s\S]{0,1200}(?:我认为|我认同|打动我|值得深入探讨|特别认同|what I find|I think|I agree)/iu,/(?:你还有什么|还想深入探讨|would you like|do you want).{0,120}[?？]\s*$/iu].some(t=>t.test(e))}(e)}(n,t)).map(t=>({source:n,output:t})):[]})}export function controlBlockers(t){const e=function successfulEventIndexesBySource(t){const e=new Map;return t.forEach((t,n)=>{const r=function successfulEventSource(t){if("runtime.tool.direct.completed"===t.type)return t.toolId;const e=readAdapterEvent(t),n=readToolSource(e);return isToolResultEvent(e)&&n&&isSuccessfulEvidenceEvent(e)?n:void 0}(t);r&&e.set(r,[...e.get(r)??[],n])}),e}(t);return t.flatMap((t,n)=>{const r=readAdapterEvent(t),o=readEventStatus(r),s=readToolSource(r)??"tool";return o&&isBlockerStatus(o)?isResolvedByLaterCompletion(o)&&function completedAfter(t,e,n){return(t.get(e)??[]).some(t=>t>n)}(e,s,n)?[]:[`${s}:${o}`]:[]})}export function controlGaps(t){const e=new Set(successfulEvidenceItems(t).map(t=>t.source));return t.flatMap(t=>{const n=readAdapterEvent(t),r=readEventStatus(n),o=readToolSource(n)??"tool";return r&&isGapStatus(r)?e.has(o)&&isResolvedByLaterCompletion(r)?[]:[`${o}:${r}`]:[]})}function stringifyEvidence(t){return"string"==typeof t?t.trim()?[t]:[]:null==t?[]:[JSON.stringify(t)]}function readAdapterEvent(t){return"runtime.adapter.event"===t.type&&isRecord(t.event)?t.event:void 0}function isSuccessfulEvidenceEvent(t){const e=readEventStatus(t);return!e||/^(?:completed|success|ok|recorded)$/iu.test(e)}function isBlockerStatus(t){return/^(?:blocked|approval_required|schema_repair_failed|tool_argument_error|invalid_input)$/iu.test(t)}function isGapStatus(t){return/^(?:dependency_required|plan_required|repeated_tool_call_limit|duplicate_tool_call)$/iu.test(t)}function isResolvedByLaterCompletion(t){return isGapStatus(t)||isBlockerStatus(t)}function isToolResultEvent(t){return"deepagents.tool_execution.result"===t?.eventType||"agent.tool.result"===t?.phase}function isRecord(t){return"object"==typeof t&&null!==t&&!Array.isArray(t)}function readString(t){return"string"==typeof t&&t.trim()?t:void 0}function readEventStatus(t){return readString(t?.controlStatus)??readString(t?.status)??function readOutputStatus(t){if(isRecord(t)&&"string"==typeof t.status)return t.status;if("string"==typeof t){const e=function parseJsonRecord(t){try{const e=JSON.parse(t);return isRecord(e)?e:void 0}catch{return}}(t);return"string"==typeof e?.status?e.status:t.match(/^Status:\s*([A-Za-z0-9_-]+)/imu)?.[1]}}(t?.output)}function readToolSource(t){return readString(t?.toolId)??readString(t?.toolName)??readString(t?.name)}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
export function completeRun(
|
|
1
|
+
import{controlGaps as t}from"../quality/event-evidence.js";export function completeRun(e){const r=e.store.getRun(e.requestId);if("cancelled"===r?.state)return response(e,"cancelled",r.artifacts);!function assertDeliverableOutput(e,r){if(!e.text.trim())throw new Error("runtime_empty_output: Runtime request produced no user-facing output.");!function assertNoOmittedControlGaps(e,r){if(function resultWasSynthesizedFromEvidence(t){const e=t.metadata?.toolCallRecovery;return"object"==typeof e&&null!==e&&"synthesized"in e&&!0===e.synthesized}(e))return;const s=t(r).filter(t=>!function mentionsGap(t,e){const[r,s]=e.split(":"),n=t.toLowerCase();return n.includes(e.toLowerCase())||Boolean(r&&s&&n.includes(r.toLowerCase())&&n.includes(s.toLowerCase()))}(e.text,t));if(0!==s.length)throw new Error(`runtime_unresolved_control_gap: Runtime request produced unresolved control gap(s): ${s.slice(0,8).join(", ")}`)}(e,r)}(e.result,r?.events??[]);const s=(e.result.artifacts??[]).map(toPublicArtifact);r&&e.store.updateRun(e.requestId,{state:"completed",output:e.result.text,metadata:{...r.metadata,...e.result.metadata},artifacts:mergeArtifacts(r.artifacts,s),completedAt:(new Date).toISOString()});for(const t of e.result.artifacts??[])e.artifacts?.createArtifact({...t,requestId:e.requestId,sessionId:e.sessionId,agentId:e.agent.id});for(const t of s)e.emit({type:"runtime.artifact.created",requestId:e.requestId,sessionId:e.sessionId,agentId:e.agent.id,artifact:t});return e.emit({type:"runtime.request.completed",requestId:e.requestId,sessionId:e.sessionId,agentId:e.agent.id,output:e.result.text}),response(e,"completed",e.store.getRun(e.requestId)?.artifacts)}export function failRun(t){const e=t.error instanceof Error?t.error.message:String(t.error);return t.store.getRun(t.requestId)&&t.store.updateRun(t.requestId,{state:"failed",error:e,completedAt:(new Date).toISOString()}),t.emit({type:"runtime.request.failed",requestId:t.requestId,sessionId:t.sessionId,agentId:t.agent.id,error:e}),{requestId:t.requestId,sessionId:t.sessionId,agentId:t.agent.id,state:"failed",output:e}}function response(t,e,r){return{requestId:t.requestId,sessionId:t.sessionId,agentId:t.agent.id,state:e,output:t.result.text,metadata:t.result.metadata,artifacts:r??t.result.artifacts}}function mergeArtifacts(t,e){const r=new Map;for(const s of[...t??[],...e??[]])r.set(s.id,s);return[...r.values()]}function toPublicArtifact(t){return{id:t.id,kind:t.kind,...t.uri?{uri:t.uri}:{},...t.metadata?{metadata:t.metadata}:{}}}
|