stable-harness 0.0.88 → 0.0.90

package/node_modules/@stable-harness/adapter-deepagents/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@stable-harness/adapter-deepagents",
-  "version": "0.0.88",
+  "version": "0.0.90",
   "type": "module",
   "files": [
     "dist/**/*.js",
@@ -15,7 +15,7 @@
     "@langchain/node-vfs": "^0.1.4",
     "@langchain/ollama": "^1.2.7",
     "@langchain/openai": "^1.4.5",
-    "@stable-harness/core": "0.0.88",
+    "@stable-harness/core": "0.0.90",
     "deepagents": "^1.10.1",
     "langchain": "^1.4.0"
   },

package/node_modules/@stable-harness/adapter-langgraph/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@stable-harness/adapter-langgraph",
-  "version": "0.0.88",
+  "version": "0.0.90",
   "type": "module",
   "files": [
     "dist/**/*.js",
@@ -11,6 +11,6 @@
   "types": "dist/src/index.d.ts",
   "peerDependencies": {
     "@langchain/langgraph": "^1.3.0",
-    "@stable-harness/core": "0.0.88"
+    "@stable-harness/core": "0.0.90"
   }
 }

package/node_modules/@stable-harness/core/dist/quality/event-evidence.js

@@ -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),r=readToolSource(e);return isToolResultEvent(e)&&r?!isSuccessfulEvidenceEvent(e)||function isPlanningTool(t){return/^(?:write_todos|read_todos)$/u.test(t)}(r)?[]:stringifyEvidence(e.evidenceOutput??e.output).filter(t=>function isUsableEvidenceOutput(t,e){return"task"!==t||!function looksLikeUnexecutedToolIntent(t){const e=t.trim();return!(e.length>4e3)&&[/\b(?:I need to|I will|I'll|I am going to|I'm going to|Let me)\s+(?:call|use|run|invoke|get|fetch|check)\b[\s\S]{0,1200}\b(?:tool|function|system_time|web_search|url_text_fetch|task)\b/iu,/(?:我需要|我要|我会|我将|让我|我来|接下来我(?:会|将)?)\s*(?:先)?(?:调用|使用|运行|执行|检查|获取|查看)\b[\s\S]{0,1200}\b(?:工具|函数|system_time|web_search|url_text_fetch|task)\b/iu,/```(?:json)?\s*(?:system_time|web_search|url_text_fetch|task)\s*```/iu].some(t=>t.test(e))}(e)&&!function looksLikeDelegatedCommentary(t){const e=t.trim();return!(e.length>8e3)&&[/(?:这篇文章|this article)[\s\S]{0,1200}(?:我认为|我认同|打动我|值得深入探讨|特别认同|what I find|I think|I agree)/iu,/(?:你还有什么|还想深入探讨|would you like|do you want).{0,120}[?？]\s*$/iu].some(t=>t.test(e))}(e)}(r,t)).map(t=>({source:r,output:t})):[]})}export function controlBlockers(t){const e=successfulEventIndexesBySource(t);return t.flatMap((t,r)=>{const n=readAdapterEvent(t),o=readControlStatus(n),s=readControlSource(n)??"tool";return o&&isBlockerStatus(o)?isResolvedByLaterCompletion(o)&&completedAfter(e,s,r)?[]:[`${s}:${o}`]:[]})}export function controlGaps(t){const e=new Set(successfulEvidenceItems(t).map(t=>t.source));return t.flatMap(t=>{const r=readAdapterEvent(t),n=readControlStatus(r),o=readControlSource(r)??"tool";return n&&isGapStatus(n)?e.has(o)&&isResolvedByLaterCompletion(n)?[]:[`${o}:${n}`]:[]})}export function omittedControlGaps(t,e){const r=successfulEventIndexesBySource(t),n=successfulEvidenceItems(t),o=function outputHasUnsupportedEvidenceClaims(t,e,r){const n=function evidenceCorpus(t,e){return[...e.map(t=>t.output),...t.flatMap(t=>"runtime.request.started"===t.type?[t.input??""]:[]),...t.flatMap(t=>"runtime.memory.recall.completed"===t.type?[t.context]:[])].join("\n")}(e,r).toLowerCase();return unsupportedTokens(t,n,/\b[A-Z][A-Z0-9]{1,12}-\d+\b/gu).length>0||unsupportedTokens(t,n,/\b[A-Za-z_$][A-Za-z0-9_$]*[A-Z][A-Za-z0-9_$]*\b/gu).filter(isCodeLikeIdentifier).length>=2}(e,t,n);return t.flatMap((t,s)=>{const u=readAdapterEvent(t),i=readControlStatus(u),a=readControlSource(u)??"tool";return i&&isGapStatus(i)?isResolvedByLaterCompletion(i)&&completedAfter(r,a,s)||!o&&function outputUsesPriorEvidence(t,e,r){const n=t.toLowerCase();return r.filter(t=>t.source===e).some(t=>{const e=t.output.trim();return!!e&&(!!n.includes(e.slice(0,500).toLowerCase())||function matchingEvidenceTokens(t,e){const r=new Set(["status","completed","success","recorded"]);return[...new Set(e.toLowerCase().match(/[a-z0-9][a-z0-9_.-]{1,}/gu)??[])].filter(t=>!r.has(t)).filter(e=>t.includes(e)).length}(n,e)>=2)})}(e,a,n)?[]:[`${a}:${i}`]:[]})}function stringifyEvidence(t){return"string"==typeof t?t.trim()?[t]:[]:null==t?[]:[JSON.stringify(t)]}function readAdapterEvent(t){if("runtime.adapter.event"===t.type&&isRecord(t.event))return t.event;const e=t;return isRecord(e)&&isToolResultEvent(e)?e:void 0}function isSuccessfulEvidenceEvent(t){const e=readEventStatus(t);return!e||/^(?:completed|success|ok|recorded)$/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 successfulEventIndexesBySource(t){const e=new Map;return t.forEach((t,r)=>{const n=function successfulEventSource(t){if("runtime.tool.direct.completed"===t.type)return t.toolId;const e=readAdapterEvent(t),r=readToolSource(e);return isToolResultEvent(e)&&r&&isSuccessfulEvidenceEvent(e)?r:void 0}(t);n&&e.set(n,[...e.get(n)??[],r])}),e}function isToolResultEvent(t){return"deepagents.tool_execution.result"===t?.eventType||"agent.tool.result"===t?.phase||"agent.tool.result"===t?.type||"deepagents.tool_execution.result"===t?.type}function completedAfter(t,e,r){return(t.get(e)??[]).some(t=>t>r)}function unsupportedTokens(t,e,r){return[...new Set(t.match(r)??[])].filter(t=>!e.includes(t.toLowerCase()))}function isCodeLikeIdentifier(t){return!(t.length<8)&&(/(?:Service|Controller|Repository|Manager|Client|Resolver|Agent|Tool|Async)$/u.test(t)||/[a-z][A-Z]/u.test(t))}function readOutputStatus(t){if(isRecord(t)&&"string"==typeof t.status)return t.status;if("string"==typeof t){const e=parseJsonRecord(t);return"string"==typeof e?.status?e.status:t.match(/^Status:\s*([A-Za-z0-9_-]+)/imu)?.[1]}}function parseJsonRecord(t){try{const e=JSON.parse(t);return isRecord(e)?e:void 0}catch{return}}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)??readOutputStatus(t?.output)}function readToolSource(t){return readString(t?.toolId)??readString(t?.toolName)??readString(t?.name)}function readControlStatus(t){const e=readEventStatus(t);return e&&isControlStatus(e)?e:findNestedControlStatus(t)}function readControlSource(t){return readToolSource(t)??findNestedToolSource(t)}function isControlStatus(t){return isGapStatus(t)||isBlockerStatus(t)}function findNestedControlStatus(t,e=0){if(e>5||null==t)return;if("string"==typeof t)return function readNestedStringStatus(t,e){const r=parseJsonRecord(t);if(r)return findNestedControlStatus(r,e+1);const n=t.match(/^Status:\s*([A-Za-z0-9_-]+)/imu)?.[1];return n&&isControlStatus(n)?n:function readControlToken(t){const e=t.match(/\b(?:blocked|approval_required|schema_repair_failed|tool_argument_error|invalid_input|dependency_required|plan_required|repeated_tool_call_limit|duplicate_tool_call)\b/iu)?.[0];return e&&isControlStatus(e)?e:void 0}(t)}(t,e);if(Array.isArray(t))return findFirstNested(t,t=>findNestedControlStatus(t,e+1));if(!isRecord(t))return;const r=readString(t.controlStatus)??readString(t.status);if(r&&isControlStatus(r))return r;const n=readOutputStatus(t.output);return n&&isControlStatus(n)?n:findFirstNested(Object.values(t),t=>findNestedControlStatus(t,e+1))}function findNestedToolSource(t,e=0){if(!(e>5||null==t)){if("string"==typeof t)return function readToolSourceFromText(t){const e=t.match(/["'](?:toolId|toolName|name)["']\s*:\s*["']([^"']+)["']/u)?.[1];return readString(e)}(t);if(Array.isArray(t))return findFirstNested(t,t=>findNestedToolSource(t,e+1));if(isRecord(t))return(readString(t.toolId)??readString(t.toolName)??readString(t.name))||findFirstNested(Object.values(t),t=>findNestedToolSource(t,e+1))}}function findFirstNested(t,e){for(const r of t){const t=e(r);if(void 0!==t)return t}}
+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),r=readToolSource(e);return isToolResultEvent(e)&&r?!isSuccessfulEvidenceEvent(e)||function isPlanningTool(t){return/^(?:write_todos|read_todos)$/u.test(t)}(r)?[]: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*$/iu].some(t=>t.test(e))}(e)}(r,t)).map(t=>({source:r,output:t})):[]})}export function controlBlockers(t){const e=successfulEventIndexesBySource(t);return t.flatMap((t,r)=>{const n=readAdapterEvent(t),o=readControlStatus(n),s=readControlSource(n)??"tool";return o&&isBlockerStatus(o)?isResolvedByLaterCompletion(o)&&completedAfter(e,s,r)||isResolvedByLaterAnyCompletion(o)&&completedAfterAny(e,r)?[]:[`${s}:${o}`]:[]})}export function controlGaps(t){const e=new Set(successfulEvidenceItems(t).map(t=>t.source));return t.flatMap(t=>{const r=readAdapterEvent(t),n=readControlStatus(r),o=readControlSource(r)??"tool";return n&&isGapStatus(n)?e.has(o)&&isResolvedByLaterCompletion(n)?[]:[`${o}:${n}`]:[]})}export function omittedControlGaps(t,e){const r=successfulEventIndexesBySource(t),n=successfulEvidenceItems(t),o=function outputHasUnsupportedEvidenceClaims(t,e,r){const n=function evidenceCorpus(t,e){return[...e.map(t=>t.output),...t.flatMap(t=>"runtime.request.started"===t.type?[t.input??""]:[]),...t.flatMap(t=>"runtime.memory.recall.completed"===t.type?[t.context]:[])].join("\n")}(e,r).toLowerCase();return unsupportedTokens(t,n,/\b[A-Z][A-Z0-9]{1,12}-\d+\b/gu).length>0||unsupportedTokens(t,n,/\b[A-Za-z_$][A-Za-z0-9_$]*[A-Z][A-Za-z0-9_$]*\b/gu).filter(isCodeLikeIdentifier).length>=2}(e,t,n);return t.flatMap((t,s)=>{const u=readAdapterEvent(t),i=readControlStatus(u),a=readControlSource(u)??"tool";return i&&function isOmittedControlStatus(t){return isGapStatus(t)||isBlockerStatus(t)}(i)?isResolvedByLaterCompletion(i)&&completedAfter(r,a,s)||isResolvedByLaterAnyCompletion(i)&&completedAfterAny(r,s)||!o&&function outputUsesPriorEvidence(t,e,r){const n=t.toLowerCase();return r.filter(t=>t.source===e).some(t=>{const e=t.output.trim();return!!e&&(!!n.includes(e.slice(0,500).toLowerCase())||function matchingEvidenceTokens(t,e){const r=new Set(["status","completed","success","recorded"]);return[...new Set(e.toLowerCase().match(/[a-z0-9][a-z0-9_.-]{1,}/gu)??[])].filter(t=>!r.has(t)).filter(e=>t.includes(e)).length}(n,e)>=2)})}(e,a,n)?[]:[`${a}:${i}`]:[]})}function stringifyEvidence(t){return"string"==typeof t?t.trim()?[t]:[]:null==t?[]:[JSON.stringify(t)]}function readAdapterEvent(t){if("runtime.adapter.event"===t.type&&isRecord(t.event))return t.event;const e=t;return isRecord(e)&&isToolResultEvent(e)?e:void 0}function isSuccessfulEvidenceEvent(t){const e=readEventStatus(t);return!e||/^(?:completed|success|ok|recorded)$/iu.test(e)}function isBlockerStatus(t){return/^(?:blocked|approval_required|schema_repair_failed|tool_argument_error|invalid_input|task_inventory_blocked)$/iu.test(t)}function isGapStatus(t){return/^(?:dependency_required|plan_required|repeated_tool_call_limit|duplicate_tool_call)$/iu.test(t)}function isResolvedByLaterCompletion(t){return isGapStatus(t)||isBlockerStatus(t)}function isResolvedByLaterAnyCompletion(t){return"task_inventory_blocked"===t}function successfulEventIndexesBySource(t){const e=new Map;return t.forEach((t,r)=>{const n=function successfulEventSource(t){if("runtime.tool.direct.completed"===t.type)return t.toolId;const e=readAdapterEvent(t),r=readToolSource(e);return isToolResultEvent(e)&&r&&isSuccessfulEvidenceEvent(e)?r:void 0}(t);n&&e.set(n,[...e.get(n)??[],r])}),e}function isToolResultEvent(t){return"deepagents.tool_execution.result"===t?.eventType||"agent.tool.result"===t?.phase||"agent.tool.result"===t?.type||"deepagents.tool_execution.result"===t?.type}function completedAfter(t,e,r){return(t.get(e)??[]).some(t=>t>r)}function completedAfterAny(t,e){return[...t.values()].some(t=>t.some(t=>t>e))}function unsupportedTokens(t,e,r){return[...new Set(t.match(r)??[])].filter(t=>!e.includes(t.toLowerCase()))}function isCodeLikeIdentifier(t){return!(t.length<8)&&(/(?:Service|Controller|Repository|Manager|Client|Resolver|Agent|Tool|Async)$/u.test(t)||/[a-z][A-Z]/u.test(t))}function readOutputStatus(t){if(isRecord(t)&&"string"==typeof t.status)return t.status;if("string"==typeof t){const e=parseJsonRecord(t);return"string"==typeof e?.status?e.status:t.match(/^Status:\s*([A-Za-z0-9_-]+)/imu)?.[1]}}function parseJsonRecord(t){try{const e=JSON.parse(t);return isRecord(e)?e:void 0}catch{return}}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)??readOutputStatus(t?.output)}function readToolSource(t){return readString(t?.toolId)??readString(t?.toolName)??readString(t?.name)}function readControlStatus(t){const e=function readInventoryRepairControlStatus(t){if("inventory.repair"===t?.phase&&"blocked"===t.status)return"task"===readInventoryRepairSource(t)?"task_inventory_blocked":"blocked"}(t);if(e)return e;const r=readEventStatus(t);return r&&isControlStatus(r)?r:findNestedControlStatus(t)}function readControlSource(t){return readToolSource(t)??readInventoryRepairSource(t)??findNestedToolSource(t)}function isControlStatus(t){return isGapStatus(t)||isBlockerStatus(t)}function findNestedControlStatus(t,e=0){if(e>5||null==t)return;if("string"==typeof t)return function readNestedStringStatus(t,e){const r=parseJsonRecord(t);if(r)return findNestedControlStatus(r,e+1);const n=t.match(/^Status:\s*([A-Za-z0-9_-]+)/imu)?.[1];return n&&isControlStatus(n)?n:function readControlToken(t){const e=t.match(/\b(?:blocked|approval_required|schema_repair_failed|tool_argument_error|invalid_input|task_inventory_blocked|dependency_required|plan_required|repeated_tool_call_limit|duplicate_tool_call)\b/iu)?.[0];return e&&isControlStatus(e)?e:void 0}(t)}(t,e);if(Array.isArray(t))return findFirstNested(t,t=>findNestedControlStatus(t,e+1));if(!isRecord(t))return;const r=readString(t.controlStatus)??readString(t.status);if(r&&isControlStatus(r))return r;const n=readOutputStatus(t.output);return n&&isControlStatus(n)?n:findFirstNested(Object.values(t),t=>findNestedControlStatus(t,e+1))}function findNestedToolSource(t,e=0){if(!(e>5||null==t)){if("string"==typeof t)return function readToolSourceFromText(t){const e=t.match(/["'](?:toolId|toolName|name)["']\s*:\s*["']([^"']+)["']/u)?.[1];return readString(e)}(t);if(Array.isArray(t))return findFirstNested(t,t=>findNestedToolSource(t,e+1));if(isRecord(t))return(readString(t.toolId)??readString(t.toolName)??readString(t.name))||findFirstNested(Object.values(t),t=>findNestedToolSource(t,e+1))}}function findFirstNested(t,e){for(const r of t){const t=e(r);if(void 0!==t)return t}}function readInventoryRepairSource(t){const e=isRecord(t?.diagnostic)?t.diagnostic:void 0;return readString(e?.layer)}

package/node_modules/@stable-harness/core/dist/quality/execution-review.js

@@ -1 +1 @@
-import{controlBlockers as e,controlGaps as t,successfulEvidenceOutputs as n,successfulEvidenceToolIds as o}from"./event-evidence.js";const r=/(?<![\w.])(?:\d{1,3}(?:,\d{3})+|\d+)(?:\.\d+)?[%kKmMbBtTxX]?(?!\w)/gu;export const defaultExecutionEvaluatorRules=[function blockerIssues(t,n){return n.executionReview.stopOnBlocker?e(t.events).map(e=>({code:"control_blocker",message:`Execution produced a control blocker: ${e}`,recoverable:!1})):[]},function controlGapIssues(e,n){if(!n.executionReview.stopOnBlocker||!e.output?.text.trim())return[];const o=t(e.events).filter(t=>!function mentionsGap(e,t){const[n,o]=t.split(":"),r=e.toLowerCase();return r.includes(t.toLowerCase())||Boolean(n&&o&&r.includes(n.toLowerCase())&&r.includes(o.toLowerCase()))}(e.output?.text??"",t));return 0===o.length?[]:[{code:"unresolved_control_gap",message:`Final answer omitted unresolved runtime evidence gap(s): ${o.slice(0,8).join(", ")}`,recoverable:!0}]},function emptyFinalIssues(e,t){return!t.executionReview.rejectEmptyFinal||e.output?.text.trim()?[]:[{code:"empty_final_answer",message:"The final answer is empty.",recoverable:!0}]},function toolEvidenceIssues(e,t){return!t.executionReview.requireToolEvidence||o(e.events).length>0?[]:[{code:"missing_tool_evidence",message:"No successful tool or delegated-task evidence was observed.",recoverable:!0}]},function ungroundedNumberIssues(e,t){if(!t.executionReview.rejectUngroundedNumbers||!e.output?.text.trim())return[];const o=numberSet(n(e.events).join("\n"));if(0===o.size)return[];const r=[...numberSet(e.output.text)].filter(e=>!function isSupportedNumber(e,t){if(t.has(e))return!0;const n=Number.parseFloat(e);if(!Number.isFinite(n))return!1;for(const e of t){const t=Number.parseFloat(e);if(Number.isFinite(t)&&Math.abs(t-n)<=roundingTolerance(n))return!0}return!1}(e,o));return 0===r.length?[]:[{code:"ungrounded_numeric_claim",message:`Final answer contains numeric claims not found in successful tool evidence: ${r.slice(0,12).join(", ")}`,recoverable:!1}]}];export function reviewExecutionEvidence(e,t,n=defaultExecutionEvaluatorRules){if(!t.enabled||!t.executionReview.enabled)return{verdict:"pass",issues:[]};const o=evaluateExecutionRules(e,t,n);return 0===o.length?{verdict:"pass",issues:[]}:{verdict:o.some(e=>!e.recoverable)?"blocked":"continue_react",issues:o}}export function evaluateExecutionRules(e,t,n=defaultExecutionEvaluatorRules){return n.flatMap(n=>n(e,t))}function numberSet(e){const t=new Set;for(const n of e.matchAll(r)){const e=normalizeNumber(n[0]);e&&t.add(e)}return t}function normalizeNumber(e){const t=e.replace(/,/gu,"").replace(/^\+/u,"").replace(/[%kKmMbBtTxX]$/u,"").trim();if(t){if(/^\d+$/u.test(t)){const e=Number.parseInt(t,10);if(e>=1&&e<=20)return;return String(e)}return/^\d+\.\d+$/u.test(t)?t.replace(/0+$/u,"").replace(/\.$/u,""):void 0}}function roundingTolerance(e){return Math.abs(e)>=1e3?1:Math.abs(e)>=100?.1:Math.abs(e)>=10?.05:.005}
+import{controlBlockers as e,controlGaps as t,successfulEvidenceOutputs as n,successfulEvidenceToolIds as o}from"./event-evidence.js";const r=/(?<![\w.])(?:\d{1,3}(?:,\d{3})+|\d+)(?:\.\d+)?[%kKmMbBtTxX]?(?!\w)/gu;export const defaultExecutionEvaluatorRules=[function blockerIssues(t,n){return n.executionReview.stopOnBlocker?e(t.events).map(e=>({code:"control_blocker",message:`Execution produced a control blocker: ${e}`,recoverable:!1})):[]},function controlGapIssues(e,n){if(!n.executionReview.stopOnBlocker||!e.output?.text.trim())return[];const o=t(e.events).filter(t=>!function mentionsGap(e,t){const[n,o]=t.split(":"),r=e.toLowerCase();return r.includes(t.toLowerCase())||function statusPhrases(e){if(!e)return[];const t=e.toLowerCase().replaceAll("_"," ");return"task_inventory_blocked"===e?[t,"workspace inventory","not in the workspace inventory"]:[t]}(o).some(e=>r.includes(e))||Boolean(n&&o&&r.includes(n.toLowerCase())&&r.includes(o.toLowerCase()))}(e.output?.text??"",t));return 0===o.length?[]:[{code:"unresolved_control_gap",message:`Final answer omitted unresolved runtime evidence gap(s): ${o.slice(0,8).join(", ")}`,recoverable:!0}]},function emptyFinalIssues(e,t){return!t.executionReview.rejectEmptyFinal||e.output?.text.trim()?[]:[{code:"empty_final_answer",message:"The final answer is empty.",recoverable:!0}]},function toolEvidenceIssues(e,t){return!t.executionReview.requireToolEvidence||o(e.events).length>0?[]:[{code:"missing_tool_evidence",message:"No successful tool or delegated-task evidence was observed.",recoverable:!0}]},function ungroundedNumberIssues(e,t){if(!t.executionReview.rejectUngroundedNumbers||!e.output?.text.trim())return[];const o=numberSet(n(e.events).join("\n"));if(0===o.size)return[];const r=[...numberSet(e.output.text)].filter(e=>!function isSupportedNumber(e,t){if(t.has(e))return!0;const n=Number.parseFloat(e);if(!Number.isFinite(n))return!1;for(const e of t){const t=Number.parseFloat(e);if(Number.isFinite(t)&&Math.abs(t-n)<=roundingTolerance(n))return!0}return!1}(e,o));return 0===r.length?[]:[{code:"ungrounded_numeric_claim",message:`Final answer contains numeric claims not found in successful tool evidence: ${r.slice(0,12).join(", ")}`,recoverable:!1}]}];export function reviewExecutionEvidence(e,t,n=defaultExecutionEvaluatorRules){if(!t.enabled||!t.executionReview.enabled)return{verdict:"pass",issues:[]};const o=evaluateExecutionRules(e,t,n);return 0===o.length?{verdict:"pass",issues:[]}:{verdict:o.some(e=>!e.recoverable)?"blocked":"continue_react",issues:o}}export function evaluateExecutionRules(e,t,n=defaultExecutionEvaluatorRules){return n.flatMap(n=>n(e,t))}function numberSet(e){const t=new Set;for(const n of e.matchAll(r)){const e=normalizeNumber(n[0]);e&&t.add(e)}return t}function normalizeNumber(e){const t=e.replace(/,/gu,"").replace(/^\+/u,"").replace(/[%kKmMbBtTxX]$/u,"").trim();if(t){if(/^\d+$/u.test(t)){const e=Number.parseInt(t,10);if(e>=1&&e<=20)return;return String(e)}return/^\d+\.\d+$/u.test(t)?t.replace(/0+$/u,"").replace(/\.$/u,""):void 0}}function roundingTolerance(e){return Math.abs(e)>=1e3?1:Math.abs(e)>=100?.1:Math.abs(e)>=10?.05:.005}

package/node_modules/@stable-harness/core/dist/recovery/control-omission.d.ts

@@ -0,0 +1,6 @@
+import type { RuntimeEvent } from "../types.js";
+export declare function omittedControlRecoveryLines(input: {
+    output: string;
+    events: RuntimeEvent[];
+    availableToolIds?: string[];
+}): string[] | undefined;

package/node_modules/@stable-harness/core/dist/recovery/control-omission.js

@@ -0,0 +1 @@
+import{omittedControlGaps as e}from"../quality/event-evidence.js";export function omittedControlRecoveryLines(t){const o=e(t.events,t.output).filter(e=>!function mentionsGap(e,t){const[o,r]=t.split(":"),n=e.toLowerCase(),i=function statusPhrases(e){if(!e)return[];const t=e.toLowerCase().replaceAll("_"," ");return"task_inventory_blocked"===e?[t,"workspace inventory","not in the workspace inventory"]:"repeated_tool_call_limit"===e?[t,"repeat limit"]:[t]}(r);return n.includes(t.toLowerCase())||Boolean(r&&n.includes(r.toLowerCase()))||i.some(e=>n.includes(e))||Boolean(o&&r&&n.includes(o.toLowerCase())&&n.includes(r.toLowerCase()))}(t.output,e));if(0===o.length)return;const r=function recentToolEvidence(e,t){return e.flatMap(e=>{const o="runtime.adapter.event"===e.type&&isRecord(e.event)?e.event:void 0;return function isToolResultEvent(e){return"deepagents.tool_execution.result"===e?.eventType||"agent.tool.result"===e?.phase}(o)&&"string"==typeof o.toolId?function isControlToolOutput(e){if("string"!=typeof e||!e.trim().startsWith("{"))return!1;try{const t=JSON.parse(e),o=isRecord(t)?t.status:void 0;return"duplicate_tool_call"===o||"repeated_tool_call_limit"===o||"tool_argument_error"===o||"task_inventory_blocked"===o}catch{return!1}}(o.output)?[]:[`- ${o.toolId}: ${formatToolEvidence(o,t)}`]:[]}).slice(-5)}(t.events,1e3);return["Stable runtime recovery: your previous final answer omitted unresolved runtime control gap(s) or blocker(s).",`Unresolved control gap(s): ${o.slice(0,8).join(", ")}`,"Continue the same user request from the completed evidence.","Use one remaining declared tool or subagent action only when it is required and available in the runtime inventory.",...t.availableToolIds?.length?[`Available configured tools: ${t.availableToolIds.join(", ")}`,"Do not invent, print, or call tools that are not in this list."]:[],"If no declared action can safely continue, return a grounded blocker that explicitly preserves the unresolved runtime control state.",...r.length>0?["","Recent executed tool evidence:",...r]:[],"","Previous invalid final answer:",t.output]}function formatToolEvidence(e,t=1e3){return"string"==typeof e.output&&e.output.trim()?e.output.slice(0,t):"string"==typeof e.error&&e.error.trim()?`error: ${e.error.slice(0,t)}`:isRecord(e.args)?`completed with args: ${JSON.stringify(e.args).slice(0,t)}`:"completed"}function isRecord(e){return"object"==typeof e&&null!==e&&!Array.isArray(e)}

package/node_modules/@stable-harness/core/dist/recovery/tool-call.js

@@ -1 +1 @@
-import{controlGaps as e}from"../quality/event-evidence.js";import{containsProgressOnlyToolIntent as t,progressOnlyToolIntentMessage as o}from"./progress-intent.js";import{containsRawToolResultText as n}from"./tool-result.js";export function toolCallRecoveryEnabled(e){return!0===readToolCallRecovery(e).enabled}export function isRecoverableAdapterError(e,t){const o=readToolCallRecovery(t);if(!0!==o.enabled)return!1;const n=e instanceof Error?e.message:String(e);return readRegexps(o.adapterErrorPatterns,[/XML syntax error|tool.?call.*syntax|malformed.*(?:XML|tool)|Non string tool message content|repeat limit reached for tool/iu]).some(e=>e.test(n))}export function buildAdapterErrorRecoveryPrompt(e,t,o){const n=t instanceof Error?t.message:String(t),r=readToolCallRecovery(o).instruction;return recoverRequest(e,["Stable runtime recovery: the backend failed while parsing a tool call.",`Parser error: ${n}`,"string"==typeof r?r:"Continue the same user request using the backend's normal tool-calling mechanism, then return a final human-readable answer.","Do not print raw tool-call markup, JSON tool-call envelopes, or pseudo tool-call text in the final answer."])}export function buildResultRecoveryRequest(e){const o=readToolCallRecovery(e.policy);if(!0!==o.enabled)return;if(containsRawToolCallText(e.output,o)){const t=recentToolEvidence(e.events,1e3);return recoverRequest(e.request,["Stable runtime recovery: your previous final answer printed raw tool-call markup instead of executing the tool.","Continue the same user request by calling the available upstream tool normally when more evidence is required.","If you call a tool, the next assistant action must be the backend's structured tool call itself, with no prose before it.",...e.availableToolIds?.length?[`Available configured tools: ${e.availableToolIds.join(", ")}`,"Do not invent, print, or call tools that are not in this list."]:[],"If the conversation context already contains enough evidence to answer, synthesize the final answer from that context instead.","Do not print XML, JSON, markdown fences, pseudo tool-call text, plans, or future-intent text such as saying you will call or wait for a tool.",...t.length>0?["","Recent executed tool evidence:",...t]:[],"","Previous invalid final answer:",e.output])}if(t(e.output,e.availableToolIds)){const t=recentToolEvidence(e.events,1e3);return recoverRequest(e.request,["Stable runtime recovery: your previous final answer was progress-only text that declared a future tool call instead of executing it.","Continue the same user request by calling the named available tool through the backend's normal structured mechanism.","If the conversation context already contains enough evidence to answer, synthesize the final answer from that context instead.",...e.availableToolIds?.length?[`Available configured tools: ${e.availableToolIds.join(", ")}`,"Do not invent, print, or call tools that are not in this list."]:[],"Do not narrate intended future tool calls as a final answer.",...t.length>0?["","Recent executed tool evidence:",...t]:[],"","Previous invalid final answer:",e.output])}if(n(e.output,e.events)){const t=recentToolEvidence(e.events,1e3);return recoverRequest(e.request,["Stable runtime recovery: your previous final answer copied an executed tool result JSON as the user-facing answer.","Continue the same user request from the existing tool evidence.","If more evidence is required, use one remaining declared tool or subagent action through the backend's normal structured mechanism.",...e.availableToolIds?.length?[`Available configured tools: ${e.availableToolIds.join(", ")}`,"Do not invent, print, or call tools that are not in this list."]:[],"Otherwise synthesize a human-readable final answer from the executed tool evidence.","Do not return the raw tool result JSON as the final answer.",...t.length>0?["","Recent executed tool evidence:",...t]:[],"","Previous invalid final answer:",e.output])}if(containsToolExecutionErrorText(e.output,o)){const t=recentToolEvidence(e.events,1e3);return recoverRequest(e.request,["Stable runtime recovery: your previous final answer exposed a backend tool execution error instead of handling it.","Continue the same user request using the backend's normal structured tool-calling mechanism.","Do not retry the same invalid tool arguments. If the failed tool is not required to answer the user, synthesize the final answer from the available context instead.",...e.availableToolIds?.length?[`Available configured tools: ${e.availableToolIds.join(", ")}`,"Do not invent, print, or call tools that are not in this list."]:[],"Do not print tool error stacks, schema validation diagnostics, raw tool-call markup, JSON tool-call envelopes, or pseudo tool-call text in the final answer.",...t.length>0?["","Recent executed tool evidence:",...t]:[],"","Previous invalid final answer:",e.output])}const r=function lastConfiguredEventHint(e,t){const o=function readEventRecoveryHints(e){return(Array.isArray(e)?e:[]).flatMap(e=>isRecord(e)&&"string"==typeof e.instruction?[{..."string"==typeof e.toolId?{toolId:e.toolId}:{},..."string"==typeof e.phase?{phase:e.phase}:{},..."string"==typeof e.outputIncludes?{outputIncludes:e.outputIncludes}:{},..."string"==typeof e.outputMatches?{outputMatches:e.outputMatches}:{},instruction:e.instruction}]:[])}(t.eventRecoveryHints);if(0!==o.length)return e.flatMap(e=>function readMatchingHints(e,t){const o="runtime.adapter.event"===e.type&&isRecord(e.event)?e.event:void 0;return o?t.filter(e=>function eventMatchesHint(e,t){return(!t.toolId||e.toolId===t.toolId)&&(!t.phase||e.phase===t.phase)&&(t.outputIncludes?"string"==typeof e.output&&e.output.includes(t.outputIncludes):!t.outputMatches||"string"==typeof e.output&&new RegExp(t.outputMatches,"u").test(e.output))}(o,e)).map(e=>({output:"string"==typeof o.output?o.output:"Adapter event matched configured recovery hint.",instruction:e.instruction})):[]}(e,o)).at(-1)}(e.events,o);return r?recoverRequest(e.request,["Stable runtime recovery: a previous adapter event matched a configured recovery hint.",r.output,r.instruction]):void 0}export function buildExecutionContractRecoveryRequest(e){if(!0!==readToolCallRecovery(e.policy).enabled)return;const t=function lastMissingEvidenceTools(e){for(let t=e.length-1;t>=0;t-=1){const o=e[t];if("runtime.execution.contract.failed"===o?.type)return readStringArray(o.missingEvidenceTools)}return[]}(e.events);return 0!==t.length?recoverRequest(e.request,["Stable runtime recovery: the execution contract was not satisfied.",`Required evidence tool(s) were missing: ${t.join(", ")}`,"Continue the same user request by calling the missing required evidence tool(s) through the backend's normal structured tool-calling mechanism.","Do not produce a final answer until the required evidence tool call has executed and you have synthesized its result.","Do not print XML, JSON, markdown fences, pseudo tool-call text, plans, or future-intent text in the final answer."]):void 0}export function assertNoRawToolCallOutput(e,t){if(containsRawToolCallOutput(e,t))throw new Error(`Adapter returned raw tool-call text as the final answer after recovery. The backend must execute tools instead of printing tool-call markup. Output preview: ${previewOutput(e)}`)}export function containsRawToolCallOutput(e,t){const o=readToolCallRecovery(t);return!0===o.enabled&&containsRawToolCallText(e,o)}export function containsRecoverableResultOutput(e,t){const o=readToolCallRecovery(t);return!0===o.enabled&&containsRecoverableResultFailureText(e,o)}export function assertNoRawToolResultOutput(e,t,o){if(!0===readToolCallRecovery(o).enabled&&n(e,t))throw new Error(`Adapter returned raw tool result JSON as the final answer after recovery. The backend must synthesize a user-facing answer. Output preview: ${previewOutput(e)}`)}export function assertNoToolExecutionErrorOutput(e,t){const o=readToolCallRecovery(t);if(!0===o.enabled&&containsToolExecutionErrorText(e,o))throw new Error(`Adapter returned a tool execution error as the final answer after recovery. Output preview: ${previewOutput(e)}`)}export function assertNoProgressOnlyToolIntentOutput(e,n,r){if(!0===readToolCallRecovery(r).enabled&&t(e,n))throw new Error(o(e))}export function rawToolCallFailureMessage(){return["The model attempted to call a tool but returned the tool call as text instead of executing it.","Please retry the request or use a model/backend configuration with reliable tool calling for this workspace."].join(" ")}export function buildEvidenceSynthesisOutput(t){const o=readToolCallRecovery(t.policy);if(!0!==o.enabled||!1===o.synthesizeFromEvidenceOnFailure||!containsRecoverableResultFailureText(t.output,o))return;const n=function latestDelegatedTaskReport(e){return e.flatMap(e=>{const t="runtime.adapter.event"===e.type&&isRecord(e.event)?e.event:void 0;if(!isToolResultEvent(t)||"task"!==t.toolId)return[];const o="string"==typeof t.output?t.output.trim():"";return function looksLikeFinalReport(e){const t=e.trim();return!(t.length<80)&&(/^#{1,3}\s+\S/mu.test(t)||/\n#{1,3}\s+\S/mu.test(t)||/\n-{3,}\n/u.test(t))}(o)?[o]:[]}).at(-1)||void 0}(t.events);if(n)return n;const r=recentToolEvidence(t.events,6e3);if(0===r.length)return;const s=/\p{Script=Han}/u.test(t.request.input)?"zh":"en",a=e(t.events);return"zh"===s?function buildChineseEvidenceSynthesis(e,t,o){return["上游模型在已有工具证据后仍输出了伪工具调用；runtime 已拒绝该 raw 输出，并直接交付已执行工具返回的证据结果。","","已执行的工具证据：",...t,...o.length>0?["","未解决的证据缺口：",...o.map(e=>`- ${e}`)]:[],"",`被拒绝的最终输出预览：${previewRejectedOutput(e)}`].join("\n")}(t.output,r,a):function buildEnglishEvidenceSynthesis(e,t,o){return["The upstream model still returned pseudo tool-call text after tool evidence was available. The runtime rejected that raw output and is returning the executed tool evidence directly.","","Executed tool evidence:",...t,...o.length>0?["","Unresolved evidence gaps:",...o.map(e=>`- ${e}`)]:[],"",`Rejected final output preview: ${previewRejectedOutput(e)}`].join("\n")}(t.output,r,a)}function previewOutput(e){const t=e.replace(/\s+/gu," ").trim();return t.length>300?`${t.slice(0,297)}...`:t}function previewRejectedOutput(e){return previewOutput(e).replace(/[<>]/gu,"")}export function rawToolCallOutputPreview(e){return previewOutput(e)}function recoverRequest(e,t){return{...e,input:[e.input,"",...t].join("\n"),metadata:{...e.metadata,stableHarnessRecovery:"tool_call"}}}function containsRawToolCallText(e,t){const o=readRegexps(t.rawOutputPatterns,[/\{\s*"name"\s*:\s*"[^"]+"\s*,\s*"arguments"\s*:/iu,/\{\s*"tool_name"\s*:\s*"[^"]+"\s*,\s*"parameters"\s*:/iu,/\{\s*"type"\s*:\s*"[^"]+"\s*,\s*"args"\s*:/iu,/^\s*[A-Za-z_][\w.-]*\s*\([^)]{0,2000}\)\s*$/iu,/^\s*[A-Za-z_][\w.-]*(?:_command|_tool|_analysis|_investigate|task)\s*$/iu,/```(?:json)?[\s\S]{0,4000}"(?:tool_name|tool|name|subagent_type)"\s*:/iu,/```(?:json)?[\s\S]{0,4000}"(?:tool_name|tool|name|subagent_type)"\s*:[\s\S]{0,4000}"(?:arguments|parameters|task)"\s*:/iu,/```(?:json)?[\s\S]{0,2000}"query"\s*:[\s\S]{0,2000}"(?:max_results|count|freshness|market)"\s*:/iu]);return!![/<\s*(?:tool_call|task)\b[^>]*>/iu,/<\s*\/\s*(?:tool_call|task)\s*>/iu,/<\s*\/?\s*tool_code\b[^>]*>/iu,/<\s*[A-Za-z_][\w.-]*\s*\([^>]{0,2000}\)\s*>/iu,/<\s*\/?\s*[A-Za-z_][\w.-]*(?:_command|_tool|_analysis|_investigate|_todos|task)\b[^>]*>/iu].some(t=>t.test(e))||function looksLikeStandaloneRecoveryCandidate(e){const t=e.trim();return t.length<=6e3||/^\s*(?:```|\{|\[|[A-Za-z_][\w.-]*\s*\()/u.test(t)}(e)&&(o.some(t=>t.test(e))||[/^[\s\S]{0,2400}\b(?:I need to|I will|I'll|I am going to|I'm going to)\s+(?:call|use|invoke|delegate)\b[\s\S]{0,1200}\b(?:tool|function|specialist|subagent)\b[\s\S]{0,400}$/iu,/^[\s\S]{0,2400}\b(?:I will|I'll|I am going to|I'm going to)\s+(?:investigate|gather|check)\b[\s\S]{0,1200}\b(?:evidence|cluster|system|results?)\b[\s\S]{0,400}$/iu,/^[\s\S]{0,2400}\b(?:waiting for|wait for)\b[\s\S]{0,1200}\b(?:tool|function|specialist|subagent)\b[\s\S]{0,400}$/iu,/^[\s\S]{0,2400}\b(?:Would you like me to|Do you want me to|Should I|I can help with)\b[\s\S]{0,1200}\?[\s\S]{0,400}$/iu,/^[\s\S]{0,2400}\bCould you please provide\b[\s\S]{0,1200}\b(?:tool|function|specialist|subagent|task)\b[\s\S]{0,400}$/iu,/^[\s\S]{0,2400}\b(?:I don't|I do not) have enough information\b[\s\S]{0,1200}\b(?:tool|function|specialist|subagent|task|context)\b[\s\S]{0,400}$/iu,/^[\s\S]{0,2400}\bLet me\s+(?:call|use|invoke|delegate|check|run|verify|gather|inspect)\b[\s\S]{0,1200}\b(?:tool|function|specialist|subagent|results?|data|evidence|commands?)\b[\s\S]{0,400}$/iu,/^[\s\S]{0,2400}\b(?:Let me|I'll|I will|I am going to|I'm going to)\s+(?:start\s+by\s+)?(?:read(?:ing)?|access(?:ing)?|gather(?:ing)?|collect(?:ing)?|fetch(?:ing)?|check(?:ing)?|inspect(?:ing)?)\b[\s\S]{0,1200}\b(?:context|instructions?|workflow|pull request|PR|issue|data|evidence|details?)\b[\s\S]{0,400}$/iu,/^[\s\S]{0,2400}(?: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|读取|收集|调查|验证|查看)[\s\S]{0,1200}$/iu,/^[\s\S]{0,2400}(?:要不要|是否需要|需要我|你想让我)[\s\S]{0,1200}(?:继续|进一步|帮你|分析|检查)[\s\S]{0,1200}[?？][\s\S]{0,400}$/iu].some(t=>t.test(e)))}function containsRecoverableResultFailureText(e,t){return containsRawToolCallText(e,t)||containsToolExecutionErrorText(e,t)}function containsToolExecutionErrorText(e,t){return readRegexps(t.toolFailureOutputPatterns,[/^Error invoking tool ['"][^'"]+['"] with kwargs /iu,/Received tool input did not match expected schema/iu,/ToolMessage.*status.*error/iu]).some(t=>t.test(e))}function recentToolEvidence(e,t){return e.flatMap(e=>{const o="runtime.adapter.event"===e.type&&isRecord(e.event)?e.event:void 0;return isToolResultEvent(o)&&"string"==typeof o.toolId?function isControlToolOutput(e){if("string"!=typeof e||!e.trim().startsWith("{"))return!1;try{const t=JSON.parse(e),o=isRecord(t)?t.status:void 0;return"duplicate_tool_call"===o||"repeated_tool_call_limit"===o||"tool_argument_error"===o}catch{return!1}}(o.output)?[]:[`- ${o.toolId}: ${formatToolEvidence(o,t)}`]:[]}).slice(-5)}function isToolResultEvent(e){return"deepagents.tool_execution.result"===e?.eventType||"agent.tool.result"===e?.phase}function formatToolEvidence(e,t=1e3){return"string"==typeof e.output&&e.output.trim()?e.output.slice(0,t):"string"==typeof e.error&&e.error.trim()?`error: ${e.error.slice(0,t)}`:isRecord(e.args)?`completed with args: ${previewOutput(JSON.stringify(e.args))}`:"completed"}function readToolCallRecovery(e){if(!isRecord(e))return{};const t=isRecord(e.recovery)?e.recovery:{};return isRecord(t.toolCall)?t.toolCall:{}}function readRegexps(e,t){const o=(Array.isArray(e)?e:[]).filter(e=>"string"==typeof e&&e.length>0).map(e=>new RegExp(e,"iu"));return o.length>0?o:t}function readStringArray(e){return Array.isArray(e)?e.filter(e=>"string"==typeof e&&e.length>0):[]}function isRecord(e){return"object"==typeof e&&null!==e&&!Array.isArray(e)}
+import{controlGaps as e}from"../quality/event-evidence.js";import{omittedControlRecoveryLines as t}from"./control-omission.js";import{containsProgressOnlyToolIntent as o,progressOnlyToolIntentMessage as n}from"./progress-intent.js";import{containsRawToolResultText as r}from"./tool-result.js";export function toolCallRecoveryEnabled(e){return!0===readToolCallRecovery(e).enabled}export function isRecoverableAdapterError(e,t){const o=readToolCallRecovery(t);if(!0!==o.enabled)return!1;const n=e instanceof Error?e.message:String(e);return readRegexps(o.adapterErrorPatterns,[/XML syntax error|tool.?call.*syntax|malformed.*(?:XML|tool)|Non string tool message content|repeat limit reached for tool/iu]).some(e=>e.test(n))}export function buildAdapterErrorRecoveryPrompt(e,t,o){const n=t instanceof Error?t.message:String(t),r=readToolCallRecovery(o).instruction;return recoverRequest(e,["Stable runtime recovery: the backend failed while parsing a tool call.",`Parser error: ${n}`,"string"==typeof r?r:"Continue the same user request using the backend's normal tool-calling mechanism, then return a final human-readable answer.","Do not print raw tool-call markup, JSON tool-call envelopes, or pseudo tool-call text in the final answer."])}export function buildResultRecoveryRequest(e){const n=readToolCallRecovery(e.policy);if(!0!==n.enabled)return;if(containsRawToolCallText(e.output,n)){const t=recentToolEvidence(e.events,1e3);return recoverRequest(e.request,["Stable runtime recovery: your previous final answer printed raw tool-call markup instead of executing the tool.","Continue the same user request by calling the available upstream tool normally when more evidence is required.","If you call a tool, the next assistant action must be the backend's structured tool call itself, with no prose before it.",...e.availableToolIds?.length?[`Available configured tools: ${e.availableToolIds.join(", ")}`,"Do not invent, print, or call tools that are not in this list."]:[],"If the conversation context already contains enough evidence to answer, synthesize the final answer from that context instead.","Do not print XML, JSON, markdown fences, pseudo tool-call text, plans, or future-intent text such as saying you will call or wait for a tool.",...t.length>0?["","Recent executed tool evidence:",...t]:[],"","Previous invalid final answer:",e.output])}if(o(e.output,e.availableToolIds)){const t=recentToolEvidence(e.events,1e3);return recoverRequest(e.request,["Stable runtime recovery: your previous final answer was progress-only text that declared a future tool call instead of executing it.","Continue the same user request by calling the named available tool through the backend's normal structured mechanism.","If the conversation context already contains enough evidence to answer, synthesize the final answer from that context instead.",...e.availableToolIds?.length?[`Available configured tools: ${e.availableToolIds.join(", ")}`,"Do not invent, print, or call tools that are not in this list."]:[],"Do not narrate intended future tool calls as a final answer.",...t.length>0?["","Recent executed tool evidence:",...t]:[],"","Previous invalid final answer:",e.output])}if(r(e.output,e.events)){const t=recentToolEvidence(e.events,1e3);return recoverRequest(e.request,["Stable runtime recovery: your previous final answer copied an executed tool result JSON as the user-facing answer.","Continue the same user request from the existing tool evidence.","If more evidence is required, use one remaining declared tool or subagent action through the backend's normal structured mechanism.",...e.availableToolIds?.length?[`Available configured tools: ${e.availableToolIds.join(", ")}`,"Do not invent, print, or call tools that are not in this list."]:[],"Otherwise synthesize a human-readable final answer from the executed tool evidence.","Do not return the raw tool result JSON as the final answer.",...t.length>0?["","Recent executed tool evidence:",...t]:[],"","Previous invalid final answer:",e.output])}if(containsToolExecutionErrorText(e.output,n)){const t=recentToolEvidence(e.events,1e3);return recoverRequest(e.request,["Stable runtime recovery: your previous final answer exposed a backend tool execution error instead of handling it.","Continue the same user request using the backend's normal structured tool-calling mechanism.","Do not retry the same invalid tool arguments. If the failed tool is not required to answer the user, synthesize the final answer from the available context instead.",...e.availableToolIds?.length?[`Available configured tools: ${e.availableToolIds.join(", ")}`,"Do not invent, print, or call tools that are not in this list."]:[],"Do not print tool error stacks, schema validation diagnostics, raw tool-call markup, JSON tool-call envelopes, or pseudo tool-call text in the final answer.",...t.length>0?["","Recent executed tool evidence:",...t]:[],"","Previous invalid final answer:",e.output])}const s=t(e);if(s)return recoverRequest(e.request,s);const a=function lastConfiguredEventHint(e,t){const o=function readEventRecoveryHints(e){return(Array.isArray(e)?e:[]).flatMap(e=>isRecord(e)&&"string"==typeof e.instruction?[{..."string"==typeof e.toolId?{toolId:e.toolId}:{},..."string"==typeof e.phase?{phase:e.phase}:{},..."string"==typeof e.outputIncludes?{outputIncludes:e.outputIncludes}:{},..."string"==typeof e.outputMatches?{outputMatches:e.outputMatches}:{},instruction:e.instruction}]:[])}(t.eventRecoveryHints);if(0!==o.length)return e.flatMap(e=>function readMatchingHints(e,t){const o="runtime.adapter.event"===e.type&&isRecord(e.event)?e.event:void 0;return o?t.filter(e=>function eventMatchesHint(e,t){return(!t.toolId||e.toolId===t.toolId)&&(!t.phase||e.phase===t.phase)&&(t.outputIncludes?"string"==typeof e.output&&e.output.includes(t.outputIncludes):!t.outputMatches||"string"==typeof e.output&&new RegExp(t.outputMatches,"u").test(e.output))}(o,e)).map(e=>({output:"string"==typeof o.output?o.output:"Adapter event matched configured recovery hint.",instruction:e.instruction})):[]}(e,o)).at(-1)}(e.events,n);return a?recoverRequest(e.request,["Stable runtime recovery: a previous adapter event matched a configured recovery hint.",a.output,a.instruction]):void 0}export function buildExecutionContractRecoveryRequest(e){if(!0!==readToolCallRecovery(e.policy).enabled)return;const t=function lastMissingEvidenceTools(e){for(let t=e.length-1;t>=0;t-=1){const o=e[t];if("runtime.execution.contract.failed"===o?.type)return readStringArray(o.missingEvidenceTools)}return[]}(e.events);return 0!==t.length?recoverRequest(e.request,["Stable runtime recovery: the execution contract was not satisfied.",`Required evidence tool(s) were missing: ${t.join(", ")}`,"Continue the same user request by calling the missing required evidence tool(s) through the backend's normal structured tool-calling mechanism.","Do not produce a final answer until the required evidence tool call has executed and you have synthesized its result.","Do not print XML, JSON, markdown fences, pseudo tool-call text, plans, or future-intent text in the final answer."]):void 0}export function assertNoRawToolCallOutput(e,t){if(containsRawToolCallOutput(e,t))throw new Error(`Adapter returned raw tool-call text as the final answer after recovery. The backend must execute tools instead of printing tool-call markup. Output preview: ${previewOutput(e)}`)}export function containsRawToolCallOutput(e,t){const o=readToolCallRecovery(t);return!0===o.enabled&&containsRawToolCallText(e,o)}export function containsRecoverableResultOutput(e,t){const o=readToolCallRecovery(t);return!0===o.enabled&&containsRecoverableResultFailureText(e,o)}export function assertNoRawToolResultOutput(e,t,o){if(!0===readToolCallRecovery(o).enabled&&r(e,t))throw new Error(`Adapter returned raw tool result JSON as the final answer after recovery. The backend must synthesize a user-facing answer. Output preview: ${previewOutput(e)}`)}export function assertNoToolExecutionErrorOutput(e,t){const o=readToolCallRecovery(t);if(!0===o.enabled&&containsToolExecutionErrorText(e,o))throw new Error(`Adapter returned a tool execution error as the final answer after recovery. Output preview: ${previewOutput(e)}`)}export function assertNoProgressOnlyToolIntentOutput(e,t,r){if(!0===readToolCallRecovery(r).enabled&&o(e,t))throw new Error(n(e))}export function rawToolCallFailureMessage(){return["The model attempted to call a tool but returned the tool call as text instead of executing it.","Please retry the request or use a model/backend configuration with reliable tool calling for this workspace."].join(" ")}export function buildEvidenceSynthesisOutput(t){const o=readToolCallRecovery(t.policy);if(!0!==o.enabled||!1===o.synthesizeFromEvidenceOnFailure||!containsRecoverableResultFailureText(t.output,o))return;const n=function latestDelegatedTaskReport(e){return e.flatMap(e=>{const t="runtime.adapter.event"===e.type&&isRecord(e.event)?e.event:void 0;if(!isToolResultEvent(t)||"task"!==t.toolId)return[];const o="string"==typeof t.output?t.output.trim():"";return function looksLikeFinalReport(e){const t=e.trim();return!(t.length<80)&&(/^#{1,3}\s+\S/mu.test(t)||/\n#{1,3}\s+\S/mu.test(t)||/\n-{3,}\n/u.test(t))}(o)?[o]:[]}).at(-1)||void 0}(t.events);if(n)return n;const r=recentToolEvidence(t.events,6e3);if(0===r.length)return;const s=/\p{Script=Han}/u.test(t.request.input)?"zh":"en",a=e(t.events);return"zh"===s?function buildChineseEvidenceSynthesis(e,t,o){return["上游模型在已有工具证据后仍输出了伪工具调用；runtime 已拒绝该 raw 输出，并直接交付已执行工具返回的证据结果。","","已执行的工具证据：",...t,...o.length>0?["","未解决的证据缺口：",...o.map(e=>`- ${e}`)]:[],"",`被拒绝的最终输出预览：${previewRejectedOutput(e)}`].join("\n")}(t.output,r,a):function buildEnglishEvidenceSynthesis(e,t,o){return["The upstream model still returned pseudo tool-call text after tool evidence was available. The runtime rejected that raw output and is returning the executed tool evidence directly.","","Executed tool evidence:",...t,...o.length>0?["","Unresolved evidence gaps:",...o.map(e=>`- ${e}`)]:[],"",`Rejected final output preview: ${previewRejectedOutput(e)}`].join("\n")}(t.output,r,a)}function previewOutput(e){const t=e.replace(/\s+/gu," ").trim();return t.length>300?`${t.slice(0,297)}...`:t}function previewRejectedOutput(e){return previewOutput(e).replace(/[<>]/gu,"")}export function rawToolCallOutputPreview(e){return previewOutput(e)}function recoverRequest(e,t){return{...e,input:[e.input,"",...t].join("\n"),metadata:{...e.metadata,stableHarnessRecovery:"tool_call"}}}function containsRawToolCallText(e,t){const o=readRegexps(t.rawOutputPatterns,[/\{\s*"name"\s*:\s*"[^"]+"\s*,\s*"arguments"\s*:/iu,/\{\s*"tool_name"\s*:\s*"[^"]+"\s*,\s*"parameters"\s*:/iu,/\{\s*"type"\s*:\s*"[^"]+"\s*,\s*"args"\s*:/iu,/^\s*[A-Za-z_][\w.-]*\s*\([^)]{0,2000}\)\s*$/iu,/^\s*[A-Za-z_][\w.-]*(?:_command|_tool|_analysis|_investigate|task)\s*$/iu,/```(?:json)?[\s\S]{0,4000}"(?:tool_name|tool|name|subagent_type)"\s*:/iu,/```(?:json)?[\s\S]{0,4000}"(?:tool_name|tool|name|subagent_type)"\s*:[\s\S]{0,4000}"(?:arguments|parameters|task)"\s*:/iu,/```(?:json)?[\s\S]{0,2000}"query"\s*:[\s\S]{0,2000}"(?:max_results|count|freshness|market)"\s*:/iu]);return!![/<\s*(?:tool_call|task)\b[^>]*>/iu,/<\s*\/\s*(?:tool_call|task)\s*>/iu,/<\s*\/?\s*tool_code\b[^>]*>/iu,/<\s*[A-Za-z_][\w.-]*\s*\([^>]{0,2000}\)\s*>/iu,/<\s*\/?\s*[A-Za-z_][\w.-]*(?:_command|_tool|_analysis|_investigate|_todos|task)\b[^>]*>/iu].some(t=>t.test(e))||function looksLikeStandaloneRecoveryCandidate(e){const t=e.trim();return t.length<=6e3||/^\s*(?:```|\{|\[|[A-Za-z_][\w.-]*\s*\()/u.test(t)}(e)&&(o.some(t=>t.test(e))||[/^[\s\S]{0,2400}\b(?:I need to|I will|I'll|I am going to|I'm going to)\s+(?:call|use|invoke|delegate)\b[\s\S]{0,1200}\b(?:tool|function|specialist|subagent)\b[\s\S]{0,400}$/iu,/^[\s\S]{0,2400}\b(?:I will|I'll|I am going to|I'm going to)\s+(?:investigate|gather|check)\b[\s\S]{0,1200}\b(?:evidence|cluster|system|results?)\b[\s\S]{0,400}$/iu,/^[\s\S]{0,2400}\b(?:waiting for|wait for)\b[\s\S]{0,1200}\b(?:tool|function|specialist|subagent)\b[\s\S]{0,400}$/iu,/^[\s\S]{0,2400}\b(?:Would you like me to|Do you want me to|Should I|I can help with)\b[\s\S]{0,1200}\?[\s\S]{0,400}$/iu,/^[\s\S]{0,2400}\bCould you please provide\b[\s\S]{0,1200}\b(?:tool|function|specialist|subagent|task)\b[\s\S]{0,400}$/iu,/^[\s\S]{0,2400}\b(?:I don't|I do not) have enough information\b[\s\S]{0,1200}\b(?:tool|function|specialist|subagent|task|context)\b[\s\S]{0,400}$/iu,/^[\s\S]{0,2400}\bLet me\s+(?:call|use|invoke|delegate|check|run|verify|gather|inspect)\b[\s\S]{0,1200}\b(?:tool|function|specialist|subagent|results?|data|evidence|commands?)\b[\s\S]{0,400}$/iu,/^[\s\S]{0,2400}\b(?:Let me|I'll|I will|I am going to|I'm going to)\s+(?:start\s+by\s+)?(?:read(?:ing)?|access(?:ing)?|gather(?:ing)?|collect(?:ing)?|fetch(?:ing)?|check(?:ing)?|inspect(?:ing)?)\b[\s\S]{0,1200}\b(?:context|instructions?|workflow|pull request|PR|issue|data|evidence|details?)\b[\s\S]{0,400}$/iu,/^[\s\S]{0,2400}(?: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|读取|收集|调查|验证|查看)[\s\S]{0,1200}$/iu,/^[\s\S]{0,2400}(?:要不要|是否需要|需要我|你想让我)[\s\S]{0,1200}(?:继续|进一步|帮你|分析|检查)[\s\S]{0,1200}[?？][\s\S]{0,400}$/iu].some(t=>t.test(e)))}function containsRecoverableResultFailureText(e,t){return containsRawToolCallText(e,t)||containsToolExecutionErrorText(e,t)}function containsToolExecutionErrorText(e,t){return readRegexps(t.toolFailureOutputPatterns,[/^Error invoking tool ['"][^'"]+['"] with kwargs /iu,/Received tool input did not match expected schema/iu,/ToolMessage.*status.*error/iu]).some(t=>t.test(e))}function recentToolEvidence(e,t){return e.flatMap(e=>{const o="runtime.adapter.event"===e.type&&isRecord(e.event)?e.event:void 0;return isToolResultEvent(o)&&"string"==typeof o.toolId?function isControlToolOutput(e){if("string"!=typeof e||!e.trim().startsWith("{"))return!1;try{const t=JSON.parse(e),o=isRecord(t)?t.status:void 0;return"duplicate_tool_call"===o||"repeated_tool_call_limit"===o||"tool_argument_error"===o}catch{return!1}}(o.output)?[]:[`- ${o.toolId}: ${formatToolEvidence(o,t)}`]:[]}).slice(-5)}function isToolResultEvent(e){return"deepagents.tool_execution.result"===e?.eventType||"agent.tool.result"===e?.phase}function formatToolEvidence(e,t=1e3){return"string"==typeof e.output&&e.output.trim()?e.output.slice(0,t):"string"==typeof e.error&&e.error.trim()?`error: ${e.error.slice(0,t)}`:isRecord(e.args)?`completed with args: ${previewOutput(JSON.stringify(e.args))}`:"completed"}function readToolCallRecovery(e){if(!isRecord(e))return{};const t=isRecord(e.recovery)?e.recovery:{};return isRecord(t.toolCall)?t.toolCall:{}}function readRegexps(e,t){const o=(Array.isArray(e)?e:[]).filter(e=>"string"==typeof e&&e.length>0).map(e=>new RegExp(e,"iu"));return o.length>0?o:t}function readStringArray(e){return Array.isArray(e)?e.filter(e=>"string"==typeof e&&e.length>0):[]}function isRecord(e){return"object"==typeof e&&null!==e&&!Array.isArray(e)}

package/node_modules/@stable-harness/core/dist/runtime/completion.js

@@ -1 +1 @@
-import{omittedControlGaps 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){const s=t(r,e.text).filter(t=>!function mentionsGap(t,e){const[r,s]=e.split(":"),n=t.toLowerCase(),o=function statusPhrases(t){if(!t)return[];const e=t.toLowerCase().replaceAll("_"," ");return"repeated_tool_call_limit"===t?[e,"repeat limit"]:[e]}(s);return n.includes(e.toLowerCase())||Boolean(s&&n.includes(s.toLowerCase()))||o.some(t=>n.includes(t))||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}:{}}}
+import{omittedControlGaps 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){const s=t(r,e.text).filter(t=>!function mentionsGap(t,e){const[r,s]=e.split(":"),n=t.toLowerCase(),o=function statusPhrases(t){if(!t)return[];const e=t.toLowerCase().replaceAll("_"," ");return"task_inventory_blocked"===t?[e,"workspace inventory","not in the workspace inventory"]:"repeated_tool_call_limit"===t?[e,"repeat limit"]:[e]}(s);return n.includes(e.toLowerCase())||Boolean(s&&n.includes(s.toLowerCase()))||o.some(t=>n.includes(t))||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/node_modules/@stable-harness/core/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@stable-harness/core",
-  "version": "0.0.88",
+  "version": "0.0.90",
   "type": "module",
   "files": [
     "dist/**/*.js",
@@ -11,7 +11,7 @@
     ".": "./dist/index.js"
   },
   "peerDependencies": {
-    "@stable-harness/governance": "0.0.88",
-    "@stable-harness/memory": "0.0.88"
+    "@stable-harness/governance": "0.0.90",
+    "@stable-harness/memory": "0.0.90"
   }
 }

package/node_modules/@stable-harness/governance/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@stable-harness/governance",
-  "version": "0.0.88",
+  "version": "0.0.90",
   "type": "module",
   "files": [
     "dist/**/*.js",

package/node_modules/@stable-harness/memory/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@stable-harness/memory",
-  "version": "0.0.88",
+  "version": "0.0.90",
   "type": "module",
   "files": [
     "dist/**/*.js",

package/node_modules/@stable-harness/protocols/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@stable-harness/protocols",
-  "version": "0.0.88",
+  "version": "0.0.90",
   "type": "module",
   "files": [
     "dist/**/*.js",
@@ -10,6 +10,6 @@
   "main": "dist/src/index.js",
   "types": "dist/src/index.d.ts",
   "peerDependencies": {
-    "@stable-harness/core": "0.0.88"
+    "@stable-harness/core": "0.0.90"
   }
 }

package/node_modules/@stable-harness/tool-gateway/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@stable-harness/tool-gateway",
-  "version": "0.0.88",
+  "version": "0.0.90",
   "type": "module",
   "files": [
     "dist/**/*.js",

package/node_modules/@stable-harness/workspace-yaml/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@stable-harness/workspace-yaml",
-  "version": "0.0.88",
+  "version": "0.0.90",
   "type": "module",
   "files": [
     "dist/**/*.js",
@@ -11,6 +11,6 @@
     ".": "./dist/index.js"
   },
   "peerDependencies": {
-    "@stable-harness/core": "0.0.88"
+    "@stable-harness/core": "0.0.90"
   }
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "stable-harness",
-  "version": "0.0.88",
+  "version": "0.0.90",
   "type": "module",
   "description": "Stable application runtime and operator control plane for agent workspaces.",
   "license": "Apache-2.0",
@@ -82,14 +82,14 @@
     "@langchain/node-vfs": "^0.1.4",
     "@langchain/ollama": "^1.2.7",
     "@langchain/openai": "^1.4.5",
-    "@stable-harness/adapter-deepagents": "0.0.88",
-    "@stable-harness/adapter-langgraph": "0.0.88",
-    "@stable-harness/core": "0.0.88",
-    "@stable-harness/governance": "0.0.88",
-    "@stable-harness/memory": "0.0.88",
-    "@stable-harness/protocols": "0.0.88",
-    "@stable-harness/tool-gateway": "0.0.88",
-    "@stable-harness/workspace-yaml": "0.0.88",
+    "@stable-harness/adapter-deepagents": "0.0.90",
+    "@stable-harness/adapter-langgraph": "0.0.90",
+    "@stable-harness/core": "0.0.90",
+    "@stable-harness/governance": "0.0.90",
+    "@stable-harness/memory": "0.0.90",
+    "@stable-harness/protocols": "0.0.90",
+    "@stable-harness/tool-gateway": "0.0.90",
+    "@stable-harness/workspace-yaml": "0.0.90",
     "deepagents": "^1.10.1",
     "langchain": "^1.4.0",
     "yaml": "^2.8.2",

package/packages/adapter-deepagents/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@stable-harness/adapter-deepagents",
-  "version": "0.0.88",
+  "version": "0.0.90",
   "type": "module",
   "files": [
     "dist/**/*.js",
@@ -15,7 +15,7 @@
     "@langchain/node-vfs": "^0.1.4",
     "@langchain/ollama": "^1.2.7",
     "@langchain/openai": "^1.4.5",
-    "@stable-harness/core": "0.0.88",
+    "@stable-harness/core": "0.0.90",
     "deepagents": "^1.10.1",
     "langchain": "^1.4.0"
   },

package/packages/adapter-langgraph/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@stable-harness/adapter-langgraph",
-  "version": "0.0.88",
+  "version": "0.0.90",
   "type": "module",
   "files": [
     "dist/**/*.js",
@@ -11,6 +11,6 @@
   "types": "dist/src/index.d.ts",
   "peerDependencies": {
     "@langchain/langgraph": "^1.3.0",
-    "@stable-harness/core": "0.0.88"
+    "@stable-harness/core": "0.0.90"
   }
 }

package/packages/cli/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@stable-harness/cli",
-  "version": "0.0.88",
+  "version": "0.0.90",
   "type": "module",
   "files": [
     "dist/**/*.js",
@@ -14,12 +14,12 @@
   "types": "dist/src/index.d.ts",
   "peerDependencies": {
     "@langchain/langgraph-api": "^1.2.1",
-    "@stable-harness/adapter-deepagents": "0.0.88",
-    "@stable-harness/adapter-langgraph": "0.0.88",
-    "@stable-harness/core": "0.0.88",
-    "@stable-harness/memory": "0.0.88",
-    "@stable-harness/protocols": "0.0.88",
-    "@stable-harness/tool-gateway": "0.0.88",
-    "@stable-harness/workspace-yaml": "0.0.88"
+    "@stable-harness/adapter-deepagents": "0.0.90",
+    "@stable-harness/adapter-langgraph": "0.0.90",
+    "@stable-harness/core": "0.0.90",
+    "@stable-harness/memory": "0.0.90",
+    "@stable-harness/protocols": "0.0.90",
+    "@stable-harness/tool-gateway": "0.0.90",
+    "@stable-harness/workspace-yaml": "0.0.90"
   }
 }

package/packages/core/dist/quality/event-evidence.js

@@ -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),r=readToolSource(e);return isToolResultEvent(e)&&r?!isSuccessfulEvidenceEvent(e)||function isPlanningTool(t){return/^(?:write_todos|read_todos)$/u.test(t)}(r)?[]:stringifyEvidence(e.evidenceOutput??e.output).filter(t=>function isUsableEvidenceOutput(t,e){return"task"!==t||!function looksLikeUnexecutedToolIntent(t){const e=t.trim();return!(e.length>4e3)&&[/\b(?:I need to|I will|I'll|I am going to|I'm going to|Let me)\s+(?:call|use|run|invoke|get|fetch|check)\b[\s\S]{0,1200}\b(?:tool|function|system_time|web_search|url_text_fetch|task)\b/iu,/(?:我需要|我要|我会|我将|让我|我来|接下来我(?:会|将)?)\s*(?:先)?(?:调用|使用|运行|执行|检查|获取|查看)\b[\s\S]{0,1200}\b(?:工具|函数|system_time|web_search|url_text_fetch|task)\b/iu,/```(?:json)?\s*(?:system_time|web_search|url_text_fetch|task)\s*```/iu].some(t=>t.test(e))}(e)&&!function looksLikeDelegatedCommentary(t){const e=t.trim();return!(e.length>8e3)&&[/(?:这篇文章|this article)[\s\S]{0,1200}(?:我认为|我认同|打动我|值得深入探讨|特别认同|what I find|I think|I agree)/iu,/(?:你还有什么|还想深入探讨|would you like|do you want).{0,120}[?？]\s*$/iu].some(t=>t.test(e))}(e)}(r,t)).map(t=>({source:r,output:t})):[]})}export function controlBlockers(t){const e=successfulEventIndexesBySource(t);return t.flatMap((t,r)=>{const n=readAdapterEvent(t),o=readControlStatus(n),s=readControlSource(n)??"tool";return o&&isBlockerStatus(o)?isResolvedByLaterCompletion(o)&&completedAfter(e,s,r)?[]:[`${s}:${o}`]:[]})}export function controlGaps(t){const e=new Set(successfulEvidenceItems(t).map(t=>t.source));return t.flatMap(t=>{const r=readAdapterEvent(t),n=readControlStatus(r),o=readControlSource(r)??"tool";return n&&isGapStatus(n)?e.has(o)&&isResolvedByLaterCompletion(n)?[]:[`${o}:${n}`]:[]})}export function omittedControlGaps(t,e){const r=successfulEventIndexesBySource(t),n=successfulEvidenceItems(t),o=function outputHasUnsupportedEvidenceClaims(t,e,r){const n=function evidenceCorpus(t,e){return[...e.map(t=>t.output),...t.flatMap(t=>"runtime.request.started"===t.type?[t.input??""]:[]),...t.flatMap(t=>"runtime.memory.recall.completed"===t.type?[t.context]:[])].join("\n")}(e,r).toLowerCase();return unsupportedTokens(t,n,/\b[A-Z][A-Z0-9]{1,12}-\d+\b/gu).length>0||unsupportedTokens(t,n,/\b[A-Za-z_$][A-Za-z0-9_$]*[A-Z][A-Za-z0-9_$]*\b/gu).filter(isCodeLikeIdentifier).length>=2}(e,t,n);return t.flatMap((t,s)=>{const u=readAdapterEvent(t),i=readControlStatus(u),a=readControlSource(u)??"tool";return i&&isGapStatus(i)?isResolvedByLaterCompletion(i)&&completedAfter(r,a,s)||!o&&function outputUsesPriorEvidence(t,e,r){const n=t.toLowerCase();return r.filter(t=>t.source===e).some(t=>{const e=t.output.trim();return!!e&&(!!n.includes(e.slice(0,500).toLowerCase())||function matchingEvidenceTokens(t,e){const r=new Set(["status","completed","success","recorded"]);return[...new Set(e.toLowerCase().match(/[a-z0-9][a-z0-9_.-]{1,}/gu)??[])].filter(t=>!r.has(t)).filter(e=>t.includes(e)).length}(n,e)>=2)})}(e,a,n)?[]:[`${a}:${i}`]:[]})}function stringifyEvidence(t){return"string"==typeof t?t.trim()?[t]:[]:null==t?[]:[JSON.stringify(t)]}function readAdapterEvent(t){if("runtime.adapter.event"===t.type&&isRecord(t.event))return t.event;const e=t;return isRecord(e)&&isToolResultEvent(e)?e:void 0}function isSuccessfulEvidenceEvent(t){const e=readEventStatus(t);return!e||/^(?:completed|success|ok|recorded)$/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 successfulEventIndexesBySource(t){const e=new Map;return t.forEach((t,r)=>{const n=function successfulEventSource(t){if("runtime.tool.direct.completed"===t.type)return t.toolId;const e=readAdapterEvent(t),r=readToolSource(e);return isToolResultEvent(e)&&r&&isSuccessfulEvidenceEvent(e)?r:void 0}(t);n&&e.set(n,[...e.get(n)??[],r])}),e}function isToolResultEvent(t){return"deepagents.tool_execution.result"===t?.eventType||"agent.tool.result"===t?.phase||"agent.tool.result"===t?.type||"deepagents.tool_execution.result"===t?.type}function completedAfter(t,e,r){return(t.get(e)??[]).some(t=>t>r)}function unsupportedTokens(t,e,r){return[...new Set(t.match(r)??[])].filter(t=>!e.includes(t.toLowerCase()))}function isCodeLikeIdentifier(t){return!(t.length<8)&&(/(?:Service|Controller|Repository|Manager|Client|Resolver|Agent|Tool|Async)$/u.test(t)||/[a-z][A-Z]/u.test(t))}function readOutputStatus(t){if(isRecord(t)&&"string"==typeof t.status)return t.status;if("string"==typeof t){const e=parseJsonRecord(t);return"string"==typeof e?.status?e.status:t.match(/^Status:\s*([A-Za-z0-9_-]+)/imu)?.[1]}}function parseJsonRecord(t){try{const e=JSON.parse(t);return isRecord(e)?e:void 0}catch{return}}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)??readOutputStatus(t?.output)}function readToolSource(t){return readString(t?.toolId)??readString(t?.toolName)??readString(t?.name)}function readControlStatus(t){const e=readEventStatus(t);return e&&isControlStatus(e)?e:findNestedControlStatus(t)}function readControlSource(t){return readToolSource(t)??findNestedToolSource(t)}function isControlStatus(t){return isGapStatus(t)||isBlockerStatus(t)}function findNestedControlStatus(t,e=0){if(e>5||null==t)return;if("string"==typeof t)return function readNestedStringStatus(t,e){const r=parseJsonRecord(t);if(r)return findNestedControlStatus(r,e+1);const n=t.match(/^Status:\s*([A-Za-z0-9_-]+)/imu)?.[1];return n&&isControlStatus(n)?n:function readControlToken(t){const e=t.match(/\b(?:blocked|approval_required|schema_repair_failed|tool_argument_error|invalid_input|dependency_required|plan_required|repeated_tool_call_limit|duplicate_tool_call)\b/iu)?.[0];return e&&isControlStatus(e)?e:void 0}(t)}(t,e);if(Array.isArray(t))return findFirstNested(t,t=>findNestedControlStatus(t,e+1));if(!isRecord(t))return;const r=readString(t.controlStatus)??readString(t.status);if(r&&isControlStatus(r))return r;const n=readOutputStatus(t.output);return n&&isControlStatus(n)?n:findFirstNested(Object.values(t),t=>findNestedControlStatus(t,e+1))}function findNestedToolSource(t,e=0){if(!(e>5||null==t)){if("string"==typeof t)return function readToolSourceFromText(t){const e=t.match(/["'](?:toolId|toolName|name)["']\s*:\s*["']([^"']+)["']/u)?.[1];return readString(e)}(t);if(Array.isArray(t))return findFirstNested(t,t=>findNestedToolSource(t,e+1));if(isRecord(t))return(readString(t.toolId)??readString(t.toolName)??readString(t.name))||findFirstNested(Object.values(t),t=>findNestedToolSource(t,e+1))}}function findFirstNested(t,e){for(const r of t){const t=e(r);if(void 0!==t)return t}}
+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),r=readToolSource(e);return isToolResultEvent(e)&&r?!isSuccessfulEvidenceEvent(e)||function isPlanningTool(t){return/^(?:write_todos|read_todos)$/u.test(t)}(r)?[]: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*$/iu].some(t=>t.test(e))}(e)}(r,t)).map(t=>({source:r,output:t})):[]})}export function controlBlockers(t){const e=successfulEventIndexesBySource(t);return t.flatMap((t,r)=>{const n=readAdapterEvent(t),o=readControlStatus(n),s=readControlSource(n)??"tool";return o&&isBlockerStatus(o)?isResolvedByLaterCompletion(o)&&completedAfter(e,s,r)||isResolvedByLaterAnyCompletion(o)&&completedAfterAny(e,r)?[]:[`${s}:${o}`]:[]})}export function controlGaps(t){const e=new Set(successfulEvidenceItems(t).map(t=>t.source));return t.flatMap(t=>{const r=readAdapterEvent(t),n=readControlStatus(r),o=readControlSource(r)??"tool";return n&&isGapStatus(n)?e.has(o)&&isResolvedByLaterCompletion(n)?[]:[`${o}:${n}`]:[]})}export function omittedControlGaps(t,e){const r=successfulEventIndexesBySource(t),n=successfulEvidenceItems(t),o=function outputHasUnsupportedEvidenceClaims(t,e,r){const n=function evidenceCorpus(t,e){return[...e.map(t=>t.output),...t.flatMap(t=>"runtime.request.started"===t.type?[t.input??""]:[]),...t.flatMap(t=>"runtime.memory.recall.completed"===t.type?[t.context]:[])].join("\n")}(e,r).toLowerCase();return unsupportedTokens(t,n,/\b[A-Z][A-Z0-9]{1,12}-\d+\b/gu).length>0||unsupportedTokens(t,n,/\b[A-Za-z_$][A-Za-z0-9_$]*[A-Z][A-Za-z0-9_$]*\b/gu).filter(isCodeLikeIdentifier).length>=2}(e,t,n);return t.flatMap((t,s)=>{const u=readAdapterEvent(t),i=readControlStatus(u),a=readControlSource(u)??"tool";return i&&function isOmittedControlStatus(t){return isGapStatus(t)||isBlockerStatus(t)}(i)?isResolvedByLaterCompletion(i)&&completedAfter(r,a,s)||isResolvedByLaterAnyCompletion(i)&&completedAfterAny(r,s)||!o&&function outputUsesPriorEvidence(t,e,r){const n=t.toLowerCase();return r.filter(t=>t.source===e).some(t=>{const e=t.output.trim();return!!e&&(!!n.includes(e.slice(0,500).toLowerCase())||function matchingEvidenceTokens(t,e){const r=new Set(["status","completed","success","recorded"]);return[...new Set(e.toLowerCase().match(/[a-z0-9][a-z0-9_.-]{1,}/gu)??[])].filter(t=>!r.has(t)).filter(e=>t.includes(e)).length}(n,e)>=2)})}(e,a,n)?[]:[`${a}:${i}`]:[]})}function stringifyEvidence(t){return"string"==typeof t?t.trim()?[t]:[]:null==t?[]:[JSON.stringify(t)]}function readAdapterEvent(t){if("runtime.adapter.event"===t.type&&isRecord(t.event))return t.event;const e=t;return isRecord(e)&&isToolResultEvent(e)?e:void 0}function isSuccessfulEvidenceEvent(t){const e=readEventStatus(t);return!e||/^(?:completed|success|ok|recorded)$/iu.test(e)}function isBlockerStatus(t){return/^(?:blocked|approval_required|schema_repair_failed|tool_argument_error|invalid_input|task_inventory_blocked)$/iu.test(t)}function isGapStatus(t){return/^(?:dependency_required|plan_required|repeated_tool_call_limit|duplicate_tool_call)$/iu.test(t)}function isResolvedByLaterCompletion(t){return isGapStatus(t)||isBlockerStatus(t)}function isResolvedByLaterAnyCompletion(t){return"task_inventory_blocked"===t}function successfulEventIndexesBySource(t){const e=new Map;return t.forEach((t,r)=>{const n=function successfulEventSource(t){if("runtime.tool.direct.completed"===t.type)return t.toolId;const e=readAdapterEvent(t),r=readToolSource(e);return isToolResultEvent(e)&&r&&isSuccessfulEvidenceEvent(e)?r:void 0}(t);n&&e.set(n,[...e.get(n)??[],r])}),e}function isToolResultEvent(t){return"deepagents.tool_execution.result"===t?.eventType||"agent.tool.result"===t?.phase||"agent.tool.result"===t?.type||"deepagents.tool_execution.result"===t?.type}function completedAfter(t,e,r){return(t.get(e)??[]).some(t=>t>r)}function completedAfterAny(t,e){return[...t.values()].some(t=>t.some(t=>t>e))}function unsupportedTokens(t,e,r){return[...new Set(t.match(r)??[])].filter(t=>!e.includes(t.toLowerCase()))}function isCodeLikeIdentifier(t){return!(t.length<8)&&(/(?:Service|Controller|Repository|Manager|Client|Resolver|Agent|Tool|Async)$/u.test(t)||/[a-z][A-Z]/u.test(t))}function readOutputStatus(t){if(isRecord(t)&&"string"==typeof t.status)return t.status;if("string"==typeof t){const e=parseJsonRecord(t);return"string"==typeof e?.status?e.status:t.match(/^Status:\s*([A-Za-z0-9_-]+)/imu)?.[1]}}function parseJsonRecord(t){try{const e=JSON.parse(t);return isRecord(e)?e:void 0}catch{return}}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)??readOutputStatus(t?.output)}function readToolSource(t){return readString(t?.toolId)??readString(t?.toolName)??readString(t?.name)}function readControlStatus(t){const e=function readInventoryRepairControlStatus(t){if("inventory.repair"===t?.phase&&"blocked"===t.status)return"task"===readInventoryRepairSource(t)?"task_inventory_blocked":"blocked"}(t);if(e)return e;const r=readEventStatus(t);return r&&isControlStatus(r)?r:findNestedControlStatus(t)}function readControlSource(t){return readToolSource(t)??readInventoryRepairSource(t)??findNestedToolSource(t)}function isControlStatus(t){return isGapStatus(t)||isBlockerStatus(t)}function findNestedControlStatus(t,e=0){if(e>5||null==t)return;if("string"==typeof t)return function readNestedStringStatus(t,e){const r=parseJsonRecord(t);if(r)return findNestedControlStatus(r,e+1);const n=t.match(/^Status:\s*([A-Za-z0-9_-]+)/imu)?.[1];return n&&isControlStatus(n)?n:function readControlToken(t){const e=t.match(/\b(?:blocked|approval_required|schema_repair_failed|tool_argument_error|invalid_input|task_inventory_blocked|dependency_required|plan_required|repeated_tool_call_limit|duplicate_tool_call)\b/iu)?.[0];return e&&isControlStatus(e)?e:void 0}(t)}(t,e);if(Array.isArray(t))return findFirstNested(t,t=>findNestedControlStatus(t,e+1));if(!isRecord(t))return;const r=readString(t.controlStatus)??readString(t.status);if(r&&isControlStatus(r))return r;const n=readOutputStatus(t.output);return n&&isControlStatus(n)?n:findFirstNested(Object.values(t),t=>findNestedControlStatus(t,e+1))}function findNestedToolSource(t,e=0){if(!(e>5||null==t)){if("string"==typeof t)return function readToolSourceFromText(t){const e=t.match(/["'](?:toolId|toolName|name)["']\s*:\s*["']([^"']+)["']/u)?.[1];return readString(e)}(t);if(Array.isArray(t))return findFirstNested(t,t=>findNestedToolSource(t,e+1));if(isRecord(t))return(readString(t.toolId)??readString(t.toolName)??readString(t.name))||findFirstNested(Object.values(t),t=>findNestedToolSource(t,e+1))}}function findFirstNested(t,e){for(const r of t){const t=e(r);if(void 0!==t)return t}}function readInventoryRepairSource(t){const e=isRecord(t?.diagnostic)?t.diagnostic:void 0;return readString(e?.layer)}

package/packages/core/dist/quality/execution-review.js

@@ -1 +1 @@
-import{controlBlockers as e,controlGaps as t,successfulEvidenceOutputs as n,successfulEvidenceToolIds as o}from"./event-evidence.js";const r=/(?<![\w.])(?:\d{1,3}(?:,\d{3})+|\d+)(?:\.\d+)?[%kKmMbBtTxX]?(?!\w)/gu;export const defaultExecutionEvaluatorRules=[function blockerIssues(t,n){return n.executionReview.stopOnBlocker?e(t.events).map(e=>({code:"control_blocker",message:`Execution produced a control blocker: ${e}`,recoverable:!1})):[]},function controlGapIssues(e,n){if(!n.executionReview.stopOnBlocker||!e.output?.text.trim())return[];const o=t(e.events).filter(t=>!function mentionsGap(e,t){const[n,o]=t.split(":"),r=e.toLowerCase();return r.includes(t.toLowerCase())||Boolean(n&&o&&r.includes(n.toLowerCase())&&r.includes(o.toLowerCase()))}(e.output?.text??"",t));return 0===o.length?[]:[{code:"unresolved_control_gap",message:`Final answer omitted unresolved runtime evidence gap(s): ${o.slice(0,8).join(", ")}`,recoverable:!0}]},function emptyFinalIssues(e,t){return!t.executionReview.rejectEmptyFinal||e.output?.text.trim()?[]:[{code:"empty_final_answer",message:"The final answer is empty.",recoverable:!0}]},function toolEvidenceIssues(e,t){return!t.executionReview.requireToolEvidence||o(e.events).length>0?[]:[{code:"missing_tool_evidence",message:"No successful tool or delegated-task evidence was observed.",recoverable:!0}]},function ungroundedNumberIssues(e,t){if(!t.executionReview.rejectUngroundedNumbers||!e.output?.text.trim())return[];const o=numberSet(n(e.events).join("\n"));if(0===o.size)return[];const r=[...numberSet(e.output.text)].filter(e=>!function isSupportedNumber(e,t){if(t.has(e))return!0;const n=Number.parseFloat(e);if(!Number.isFinite(n))return!1;for(const e of t){const t=Number.parseFloat(e);if(Number.isFinite(t)&&Math.abs(t-n)<=roundingTolerance(n))return!0}return!1}(e,o));return 0===r.length?[]:[{code:"ungrounded_numeric_claim",message:`Final answer contains numeric claims not found in successful tool evidence: ${r.slice(0,12).join(", ")}`,recoverable:!1}]}];export function reviewExecutionEvidence(e,t,n=defaultExecutionEvaluatorRules){if(!t.enabled||!t.executionReview.enabled)return{verdict:"pass",issues:[]};const o=evaluateExecutionRules(e,t,n);return 0===o.length?{verdict:"pass",issues:[]}:{verdict:o.some(e=>!e.recoverable)?"blocked":"continue_react",issues:o}}export function evaluateExecutionRules(e,t,n=defaultExecutionEvaluatorRules){return n.flatMap(n=>n(e,t))}function numberSet(e){const t=new Set;for(const n of e.matchAll(r)){const e=normalizeNumber(n[0]);e&&t.add(e)}return t}function normalizeNumber(e){const t=e.replace(/,/gu,"").replace(/^\+/u,"").replace(/[%kKmMbBtTxX]$/u,"").trim();if(t){if(/^\d+$/u.test(t)){const e=Number.parseInt(t,10);if(e>=1&&e<=20)return;return String(e)}return/^\d+\.\d+$/u.test(t)?t.replace(/0+$/u,"").replace(/\.$/u,""):void 0}}function roundingTolerance(e){return Math.abs(e)>=1e3?1:Math.abs(e)>=100?.1:Math.abs(e)>=10?.05:.005}
+import{controlBlockers as e,controlGaps as t,successfulEvidenceOutputs as n,successfulEvidenceToolIds as o}from"./event-evidence.js";const r=/(?<![\w.])(?:\d{1,3}(?:,\d{3})+|\d+)(?:\.\d+)?[%kKmMbBtTxX]?(?!\w)/gu;export const defaultExecutionEvaluatorRules=[function blockerIssues(t,n){return n.executionReview.stopOnBlocker?e(t.events).map(e=>({code:"control_blocker",message:`Execution produced a control blocker: ${e}`,recoverable:!1})):[]},function controlGapIssues(e,n){if(!n.executionReview.stopOnBlocker||!e.output?.text.trim())return[];const o=t(e.events).filter(t=>!function mentionsGap(e,t){const[n,o]=t.split(":"),r=e.toLowerCase();return r.includes(t.toLowerCase())||function statusPhrases(e){if(!e)return[];const t=e.toLowerCase().replaceAll("_"," ");return"task_inventory_blocked"===e?[t,"workspace inventory","not in the workspace inventory"]:[t]}(o).some(e=>r.includes(e))||Boolean(n&&o&&r.includes(n.toLowerCase())&&r.includes(o.toLowerCase()))}(e.output?.text??"",t));return 0===o.length?[]:[{code:"unresolved_control_gap",message:`Final answer omitted unresolved runtime evidence gap(s): ${o.slice(0,8).join(", ")}`,recoverable:!0}]},function emptyFinalIssues(e,t){return!t.executionReview.rejectEmptyFinal||e.output?.text.trim()?[]:[{code:"empty_final_answer",message:"The final answer is empty.",recoverable:!0}]},function toolEvidenceIssues(e,t){return!t.executionReview.requireToolEvidence||o(e.events).length>0?[]:[{code:"missing_tool_evidence",message:"No successful tool or delegated-task evidence was observed.",recoverable:!0}]},function ungroundedNumberIssues(e,t){if(!t.executionReview.rejectUngroundedNumbers||!e.output?.text.trim())return[];const o=numberSet(n(e.events).join("\n"));if(0===o.size)return[];const r=[...numberSet(e.output.text)].filter(e=>!function isSupportedNumber(e,t){if(t.has(e))return!0;const n=Number.parseFloat(e);if(!Number.isFinite(n))return!1;for(const e of t){const t=Number.parseFloat(e);if(Number.isFinite(t)&&Math.abs(t-n)<=roundingTolerance(n))return!0}return!1}(e,o));return 0===r.length?[]:[{code:"ungrounded_numeric_claim",message:`Final answer contains numeric claims not found in successful tool evidence: ${r.slice(0,12).join(", ")}`,recoverable:!1}]}];export function reviewExecutionEvidence(e,t,n=defaultExecutionEvaluatorRules){if(!t.enabled||!t.executionReview.enabled)return{verdict:"pass",issues:[]};const o=evaluateExecutionRules(e,t,n);return 0===o.length?{verdict:"pass",issues:[]}:{verdict:o.some(e=>!e.recoverable)?"blocked":"continue_react",issues:o}}export function evaluateExecutionRules(e,t,n=defaultExecutionEvaluatorRules){return n.flatMap(n=>n(e,t))}function numberSet(e){const t=new Set;for(const n of e.matchAll(r)){const e=normalizeNumber(n[0]);e&&t.add(e)}return t}function normalizeNumber(e){const t=e.replace(/,/gu,"").replace(/^\+/u,"").replace(/[%kKmMbBtTxX]$/u,"").trim();if(t){if(/^\d+$/u.test(t)){const e=Number.parseInt(t,10);if(e>=1&&e<=20)return;return String(e)}return/^\d+\.\d+$/u.test(t)?t.replace(/0+$/u,"").replace(/\.$/u,""):void 0}}function roundingTolerance(e){return Math.abs(e)>=1e3?1:Math.abs(e)>=100?.1:Math.abs(e)>=10?.05:.005}

package/packages/core/dist/recovery/control-omission.d.ts

@@ -0,0 +1,6 @@
+import type { RuntimeEvent } from "../types.js";
+export declare function omittedControlRecoveryLines(input: {
+    output: string;
+    events: RuntimeEvent[];
+    availableToolIds?: string[];
+}): string[] | undefined;

package/packages/core/dist/recovery/control-omission.js

@@ -0,0 +1 @@
+import{omittedControlGaps as e}from"../quality/event-evidence.js";export function omittedControlRecoveryLines(t){const o=e(t.events,t.output).filter(e=>!function mentionsGap(e,t){const[o,r]=t.split(":"),n=e.toLowerCase(),i=function statusPhrases(e){if(!e)return[];const t=e.toLowerCase().replaceAll("_"," ");return"task_inventory_blocked"===e?[t,"workspace inventory","not in the workspace inventory"]:"repeated_tool_call_limit"===e?[t,"repeat limit"]:[t]}(r);return n.includes(t.toLowerCase())||Boolean(r&&n.includes(r.toLowerCase()))||i.some(e=>n.includes(e))||Boolean(o&&r&&n.includes(o.toLowerCase())&&n.includes(r.toLowerCase()))}(t.output,e));if(0===o.length)return;const r=function recentToolEvidence(e,t){return e.flatMap(e=>{const o="runtime.adapter.event"===e.type&&isRecord(e.event)?e.event:void 0;return function isToolResultEvent(e){return"deepagents.tool_execution.result"===e?.eventType||"agent.tool.result"===e?.phase}(o)&&"string"==typeof o.toolId?function isControlToolOutput(e){if("string"!=typeof e||!e.trim().startsWith("{"))return!1;try{const t=JSON.parse(e),o=isRecord(t)?t.status:void 0;return"duplicate_tool_call"===o||"repeated_tool_call_limit"===o||"tool_argument_error"===o||"task_inventory_blocked"===o}catch{return!1}}(o.output)?[]:[`- ${o.toolId}: ${formatToolEvidence(o,t)}`]:[]}).slice(-5)}(t.events,1e3);return["Stable runtime recovery: your previous final answer omitted unresolved runtime control gap(s) or blocker(s).",`Unresolved control gap(s): ${o.slice(0,8).join(", ")}`,"Continue the same user request from the completed evidence.","Use one remaining declared tool or subagent action only when it is required and available in the runtime inventory.",...t.availableToolIds?.length?[`Available configured tools: ${t.availableToolIds.join(", ")}`,"Do not invent, print, or call tools that are not in this list."]:[],"If no declared action can safely continue, return a grounded blocker that explicitly preserves the unresolved runtime control state.",...r.length>0?["","Recent executed tool evidence:",...r]:[],"","Previous invalid final answer:",t.output]}function formatToolEvidence(e,t=1e3){return"string"==typeof e.output&&e.output.trim()?e.output.slice(0,t):"string"==typeof e.error&&e.error.trim()?`error: ${e.error.slice(0,t)}`:isRecord(e.args)?`completed with args: ${JSON.stringify(e.args).slice(0,t)}`:"completed"}function isRecord(e){return"object"==typeof e&&null!==e&&!Array.isArray(e)}

package/packages/core/dist/recovery/tool-call.js

@@ -1 +1 @@
-import{controlGaps as e}from"../quality/event-evidence.js";import{containsProgressOnlyToolIntent as t,progressOnlyToolIntentMessage as o}from"./progress-intent.js";import{containsRawToolResultText as n}from"./tool-result.js";export function toolCallRecoveryEnabled(e){return!0===readToolCallRecovery(e).enabled}export function isRecoverableAdapterError(e,t){const o=readToolCallRecovery(t);if(!0!==o.enabled)return!1;const n=e instanceof Error?e.message:String(e);return readRegexps(o.adapterErrorPatterns,[/XML syntax error|tool.?call.*syntax|malformed.*(?:XML|tool)|Non string tool message content|repeat limit reached for tool/iu]).some(e=>e.test(n))}export function buildAdapterErrorRecoveryPrompt(e,t,o){const n=t instanceof Error?t.message:String(t),r=readToolCallRecovery(o).instruction;return recoverRequest(e,["Stable runtime recovery: the backend failed while parsing a tool call.",`Parser error: ${n}`,"string"==typeof r?r:"Continue the same user request using the backend's normal tool-calling mechanism, then return a final human-readable answer.","Do not print raw tool-call markup, JSON tool-call envelopes, or pseudo tool-call text in the final answer."])}export function buildResultRecoveryRequest(e){const o=readToolCallRecovery(e.policy);if(!0!==o.enabled)return;if(containsRawToolCallText(e.output,o)){const t=recentToolEvidence(e.events,1e3);return recoverRequest(e.request,["Stable runtime recovery: your previous final answer printed raw tool-call markup instead of executing the tool.","Continue the same user request by calling the available upstream tool normally when more evidence is required.","If you call a tool, the next assistant action must be the backend's structured tool call itself, with no prose before it.",...e.availableToolIds?.length?[`Available configured tools: ${e.availableToolIds.join(", ")}`,"Do not invent, print, or call tools that are not in this list."]:[],"If the conversation context already contains enough evidence to answer, synthesize the final answer from that context instead.","Do not print XML, JSON, markdown fences, pseudo tool-call text, plans, or future-intent text such as saying you will call or wait for a tool.",...t.length>0?["","Recent executed tool evidence:",...t]:[],"","Previous invalid final answer:",e.output])}if(t(e.output,e.availableToolIds)){const t=recentToolEvidence(e.events,1e3);return recoverRequest(e.request,["Stable runtime recovery: your previous final answer was progress-only text that declared a future tool call instead of executing it.","Continue the same user request by calling the named available tool through the backend's normal structured mechanism.","If the conversation context already contains enough evidence to answer, synthesize the final answer from that context instead.",...e.availableToolIds?.length?[`Available configured tools: ${e.availableToolIds.join(", ")}`,"Do not invent, print, or call tools that are not in this list."]:[],"Do not narrate intended future tool calls as a final answer.",...t.length>0?["","Recent executed tool evidence:",...t]:[],"","Previous invalid final answer:",e.output])}if(n(e.output,e.events)){const t=recentToolEvidence(e.events,1e3);return recoverRequest(e.request,["Stable runtime recovery: your previous final answer copied an executed tool result JSON as the user-facing answer.","Continue the same user request from the existing tool evidence.","If more evidence is required, use one remaining declared tool or subagent action through the backend's normal structured mechanism.",...e.availableToolIds?.length?[`Available configured tools: ${e.availableToolIds.join(", ")}`,"Do not invent, print, or call tools that are not in this list."]:[],"Otherwise synthesize a human-readable final answer from the executed tool evidence.","Do not return the raw tool result JSON as the final answer.",...t.length>0?["","Recent executed tool evidence:",...t]:[],"","Previous invalid final answer:",e.output])}if(containsToolExecutionErrorText(e.output,o)){const t=recentToolEvidence(e.events,1e3);return recoverRequest(e.request,["Stable runtime recovery: your previous final answer exposed a backend tool execution error instead of handling it.","Continue the same user request using the backend's normal structured tool-calling mechanism.","Do not retry the same invalid tool arguments. If the failed tool is not required to answer the user, synthesize the final answer from the available context instead.",...e.availableToolIds?.length?[`Available configured tools: ${e.availableToolIds.join(", ")}`,"Do not invent, print, or call tools that are not in this list."]:[],"Do not print tool error stacks, schema validation diagnostics, raw tool-call markup, JSON tool-call envelopes, or pseudo tool-call text in the final answer.",...t.length>0?["","Recent executed tool evidence:",...t]:[],"","Previous invalid final answer:",e.output])}const r=function lastConfiguredEventHint(e,t){const o=function readEventRecoveryHints(e){return(Array.isArray(e)?e:[]).flatMap(e=>isRecord(e)&&"string"==typeof e.instruction?[{..."string"==typeof e.toolId?{toolId:e.toolId}:{},..."string"==typeof e.phase?{phase:e.phase}:{},..."string"==typeof e.outputIncludes?{outputIncludes:e.outputIncludes}:{},..."string"==typeof e.outputMatches?{outputMatches:e.outputMatches}:{},instruction:e.instruction}]:[])}(t.eventRecoveryHints);if(0!==o.length)return e.flatMap(e=>function readMatchingHints(e,t){const o="runtime.adapter.event"===e.type&&isRecord(e.event)?e.event:void 0;return o?t.filter(e=>function eventMatchesHint(e,t){return(!t.toolId||e.toolId===t.toolId)&&(!t.phase||e.phase===t.phase)&&(t.outputIncludes?"string"==typeof e.output&&e.output.includes(t.outputIncludes):!t.outputMatches||"string"==typeof e.output&&new RegExp(t.outputMatches,"u").test(e.output))}(o,e)).map(e=>({output:"string"==typeof o.output?o.output:"Adapter event matched configured recovery hint.",instruction:e.instruction})):[]}(e,o)).at(-1)}(e.events,o);return r?recoverRequest(e.request,["Stable runtime recovery: a previous adapter event matched a configured recovery hint.",r.output,r.instruction]):void 0}export function buildExecutionContractRecoveryRequest(e){if(!0!==readToolCallRecovery(e.policy).enabled)return;const t=function lastMissingEvidenceTools(e){for(let t=e.length-1;t>=0;t-=1){const o=e[t];if("runtime.execution.contract.failed"===o?.type)return readStringArray(o.missingEvidenceTools)}return[]}(e.events);return 0!==t.length?recoverRequest(e.request,["Stable runtime recovery: the execution contract was not satisfied.",`Required evidence tool(s) were missing: ${t.join(", ")}`,"Continue the same user request by calling the missing required evidence tool(s) through the backend's normal structured tool-calling mechanism.","Do not produce a final answer until the required evidence tool call has executed and you have synthesized its result.","Do not print XML, JSON, markdown fences, pseudo tool-call text, plans, or future-intent text in the final answer."]):void 0}export function assertNoRawToolCallOutput(e,t){if(containsRawToolCallOutput(e,t))throw new Error(`Adapter returned raw tool-call text as the final answer after recovery. The backend must execute tools instead of printing tool-call markup. Output preview: ${previewOutput(e)}`)}export function containsRawToolCallOutput(e,t){const o=readToolCallRecovery(t);return!0===o.enabled&&containsRawToolCallText(e,o)}export function containsRecoverableResultOutput(e,t){const o=readToolCallRecovery(t);return!0===o.enabled&&containsRecoverableResultFailureText(e,o)}export function assertNoRawToolResultOutput(e,t,o){if(!0===readToolCallRecovery(o).enabled&&n(e,t))throw new Error(`Adapter returned raw tool result JSON as the final answer after recovery. The backend must synthesize a user-facing answer. Output preview: ${previewOutput(e)}`)}export function assertNoToolExecutionErrorOutput(e,t){const o=readToolCallRecovery(t);if(!0===o.enabled&&containsToolExecutionErrorText(e,o))throw new Error(`Adapter returned a tool execution error as the final answer after recovery. Output preview: ${previewOutput(e)}`)}export function assertNoProgressOnlyToolIntentOutput(e,n,r){if(!0===readToolCallRecovery(r).enabled&&t(e,n))throw new Error(o(e))}export function rawToolCallFailureMessage(){return["The model attempted to call a tool but returned the tool call as text instead of executing it.","Please retry the request or use a model/backend configuration with reliable tool calling for this workspace."].join(" ")}export function buildEvidenceSynthesisOutput(t){const o=readToolCallRecovery(t.policy);if(!0!==o.enabled||!1===o.synthesizeFromEvidenceOnFailure||!containsRecoverableResultFailureText(t.output,o))return;const n=function latestDelegatedTaskReport(e){return e.flatMap(e=>{const t="runtime.adapter.event"===e.type&&isRecord(e.event)?e.event:void 0;if(!isToolResultEvent(t)||"task"!==t.toolId)return[];const o="string"==typeof t.output?t.output.trim():"";return function looksLikeFinalReport(e){const t=e.trim();return!(t.length<80)&&(/^#{1,3}\s+\S/mu.test(t)||/\n#{1,3}\s+\S/mu.test(t)||/\n-{3,}\n/u.test(t))}(o)?[o]:[]}).at(-1)||void 0}(t.events);if(n)return n;const r=recentToolEvidence(t.events,6e3);if(0===r.length)return;const s=/\p{Script=Han}/u.test(t.request.input)?"zh":"en",a=e(t.events);return"zh"===s?function buildChineseEvidenceSynthesis(e,t,o){return["上游模型在已有工具证据后仍输出了伪工具调用；runtime 已拒绝该 raw 输出，并直接交付已执行工具返回的证据结果。","","已执行的工具证据：",...t,...o.length>0?["","未解决的证据缺口：",...o.map(e=>`- ${e}`)]:[],"",`被拒绝的最终输出预览：${previewRejectedOutput(e)}`].join("\n")}(t.output,r,a):function buildEnglishEvidenceSynthesis(e,t,o){return["The upstream model still returned pseudo tool-call text after tool evidence was available. The runtime rejected that raw output and is returning the executed tool evidence directly.","","Executed tool evidence:",...t,...o.length>0?["","Unresolved evidence gaps:",...o.map(e=>`- ${e}`)]:[],"",`Rejected final output preview: ${previewRejectedOutput(e)}`].join("\n")}(t.output,r,a)}function previewOutput(e){const t=e.replace(/\s+/gu," ").trim();return t.length>300?`${t.slice(0,297)}...`:t}function previewRejectedOutput(e){return previewOutput(e).replace(/[<>]/gu,"")}export function rawToolCallOutputPreview(e){return previewOutput(e)}function recoverRequest(e,t){return{...e,input:[e.input,"",...t].join("\n"),metadata:{...e.metadata,stableHarnessRecovery:"tool_call"}}}function containsRawToolCallText(e,t){const o=readRegexps(t.rawOutputPatterns,[/\{\s*"name"\s*:\s*"[^"]+"\s*,\s*"arguments"\s*:/iu,/\{\s*"tool_name"\s*:\s*"[^"]+"\s*,\s*"parameters"\s*:/iu,/\{\s*"type"\s*:\s*"[^"]+"\s*,\s*"args"\s*:/iu,/^\s*[A-Za-z_][\w.-]*\s*\([^)]{0,2000}\)\s*$/iu,/^\s*[A-Za-z_][\w.-]*(?:_command|_tool|_analysis|_investigate|task)\s*$/iu,/```(?:json)?[\s\S]{0,4000}"(?:tool_name|tool|name|subagent_type)"\s*:/iu,/```(?:json)?[\s\S]{0,4000}"(?:tool_name|tool|name|subagent_type)"\s*:[\s\S]{0,4000}"(?:arguments|parameters|task)"\s*:/iu,/```(?:json)?[\s\S]{0,2000}"query"\s*:[\s\S]{0,2000}"(?:max_results|count|freshness|market)"\s*:/iu]);return!![/<\s*(?:tool_call|task)\b[^>]*>/iu,/<\s*\/\s*(?:tool_call|task)\s*>/iu,/<\s*\/?\s*tool_code\b[^>]*>/iu,/<\s*[A-Za-z_][\w.-]*\s*\([^>]{0,2000}\)\s*>/iu,/<\s*\/?\s*[A-Za-z_][\w.-]*(?:_command|_tool|_analysis|_investigate|_todos|task)\b[^>]*>/iu].some(t=>t.test(e))||function looksLikeStandaloneRecoveryCandidate(e){const t=e.trim();return t.length<=6e3||/^\s*(?:```|\{|\[|[A-Za-z_][\w.-]*\s*\()/u.test(t)}(e)&&(o.some(t=>t.test(e))||[/^[\s\S]{0,2400}\b(?:I need to|I will|I'll|I am going to|I'm going to)\s+(?:call|use|invoke|delegate)\b[\s\S]{0,1200}\b(?:tool|function|specialist|subagent)\b[\s\S]{0,400}$/iu,/^[\s\S]{0,2400}\b(?:I will|I'll|I am going to|I'm going to)\s+(?:investigate|gather|check)\b[\s\S]{0,1200}\b(?:evidence|cluster|system|results?)\b[\s\S]{0,400}$/iu,/^[\s\S]{0,2400}\b(?:waiting for|wait for)\b[\s\S]{0,1200}\b(?:tool|function|specialist|subagent)\b[\s\S]{0,400}$/iu,/^[\s\S]{0,2400}\b(?:Would you like me to|Do you want me to|Should I|I can help with)\b[\s\S]{0,1200}\?[\s\S]{0,400}$/iu,/^[\s\S]{0,2400}\bCould you please provide\b[\s\S]{0,1200}\b(?:tool|function|specialist|subagent|task)\b[\s\S]{0,400}$/iu,/^[\s\S]{0,2400}\b(?:I don't|I do not) have enough information\b[\s\S]{0,1200}\b(?:tool|function|specialist|subagent|task|context)\b[\s\S]{0,400}$/iu,/^[\s\S]{0,2400}\bLet me\s+(?:call|use|invoke|delegate|check|run|verify|gather|inspect)\b[\s\S]{0,1200}\b(?:tool|function|specialist|subagent|results?|data|evidence|commands?)\b[\s\S]{0,400}$/iu,/^[\s\S]{0,2400}\b(?:Let me|I'll|I will|I am going to|I'm going to)\s+(?:start\s+by\s+)?(?:read(?:ing)?|access(?:ing)?|gather(?:ing)?|collect(?:ing)?|fetch(?:ing)?|check(?:ing)?|inspect(?:ing)?)\b[\s\S]{0,1200}\b(?:context|instructions?|workflow|pull request|PR|issue|data|evidence|details?)\b[\s\S]{0,400}$/iu,/^[\s\S]{0,2400}(?: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|读取|收集|调查|验证|查看)[\s\S]{0,1200}$/iu,/^[\s\S]{0,2400}(?:要不要|是否需要|需要我|你想让我)[\s\S]{0,1200}(?:继续|进一步|帮你|分析|检查)[\s\S]{0,1200}[?？][\s\S]{0,400}$/iu].some(t=>t.test(e)))}function containsRecoverableResultFailureText(e,t){return containsRawToolCallText(e,t)||containsToolExecutionErrorText(e,t)}function containsToolExecutionErrorText(e,t){return readRegexps(t.toolFailureOutputPatterns,[/^Error invoking tool ['"][^'"]+['"] with kwargs /iu,/Received tool input did not match expected schema/iu,/ToolMessage.*status.*error/iu]).some(t=>t.test(e))}function recentToolEvidence(e,t){return e.flatMap(e=>{const o="runtime.adapter.event"===e.type&&isRecord(e.event)?e.event:void 0;return isToolResultEvent(o)&&"string"==typeof o.toolId?function isControlToolOutput(e){if("string"!=typeof e||!e.trim().startsWith("{"))return!1;try{const t=JSON.parse(e),o=isRecord(t)?t.status:void 0;return"duplicate_tool_call"===o||"repeated_tool_call_limit"===o||"tool_argument_error"===o}catch{return!1}}(o.output)?[]:[`- ${o.toolId}: ${formatToolEvidence(o,t)}`]:[]}).slice(-5)}function isToolResultEvent(e){return"deepagents.tool_execution.result"===e?.eventType||"agent.tool.result"===e?.phase}function formatToolEvidence(e,t=1e3){return"string"==typeof e.output&&e.output.trim()?e.output.slice(0,t):"string"==typeof e.error&&e.error.trim()?`error: ${e.error.slice(0,t)}`:isRecord(e.args)?`completed with args: ${previewOutput(JSON.stringify(e.args))}`:"completed"}function readToolCallRecovery(e){if(!isRecord(e))return{};const t=isRecord(e.recovery)?e.recovery:{};return isRecord(t.toolCall)?t.toolCall:{}}function readRegexps(e,t){const o=(Array.isArray(e)?e:[]).filter(e=>"string"==typeof e&&e.length>0).map(e=>new RegExp(e,"iu"));return o.length>0?o:t}function readStringArray(e){return Array.isArray(e)?e.filter(e=>"string"==typeof e&&e.length>0):[]}function isRecord(e){return"object"==typeof e&&null!==e&&!Array.isArray(e)}
+import{controlGaps as e}from"../quality/event-evidence.js";import{omittedControlRecoveryLines as t}from"./control-omission.js";import{containsProgressOnlyToolIntent as o,progressOnlyToolIntentMessage as n}from"./progress-intent.js";import{containsRawToolResultText as r}from"./tool-result.js";export function toolCallRecoveryEnabled(e){return!0===readToolCallRecovery(e).enabled}export function isRecoverableAdapterError(e,t){const o=readToolCallRecovery(t);if(!0!==o.enabled)return!1;const n=e instanceof Error?e.message:String(e);return readRegexps(o.adapterErrorPatterns,[/XML syntax error|tool.?call.*syntax|malformed.*(?:XML|tool)|Non string tool message content|repeat limit reached for tool/iu]).some(e=>e.test(n))}export function buildAdapterErrorRecoveryPrompt(e,t,o){const n=t instanceof Error?t.message:String(t),r=readToolCallRecovery(o).instruction;return recoverRequest(e,["Stable runtime recovery: the backend failed while parsing a tool call.",`Parser error: ${n}`,"string"==typeof r?r:"Continue the same user request using the backend's normal tool-calling mechanism, then return a final human-readable answer.","Do not print raw tool-call markup, JSON tool-call envelopes, or pseudo tool-call text in the final answer."])}export function buildResultRecoveryRequest(e){const n=readToolCallRecovery(e.policy);if(!0!==n.enabled)return;if(containsRawToolCallText(e.output,n)){const t=recentToolEvidence(e.events,1e3);return recoverRequest(e.request,["Stable runtime recovery: your previous final answer printed raw tool-call markup instead of executing the tool.","Continue the same user request by calling the available upstream tool normally when more evidence is required.","If you call a tool, the next assistant action must be the backend's structured tool call itself, with no prose before it.",...e.availableToolIds?.length?[`Available configured tools: ${e.availableToolIds.join(", ")}`,"Do not invent, print, or call tools that are not in this list."]:[],"If the conversation context already contains enough evidence to answer, synthesize the final answer from that context instead.","Do not print XML, JSON, markdown fences, pseudo tool-call text, plans, or future-intent text such as saying you will call or wait for a tool.",...t.length>0?["","Recent executed tool evidence:",...t]:[],"","Previous invalid final answer:",e.output])}if(o(e.output,e.availableToolIds)){const t=recentToolEvidence(e.events,1e3);return recoverRequest(e.request,["Stable runtime recovery: your previous final answer was progress-only text that declared a future tool call instead of executing it.","Continue the same user request by calling the named available tool through the backend's normal structured mechanism.","If the conversation context already contains enough evidence to answer, synthesize the final answer from that context instead.",...e.availableToolIds?.length?[`Available configured tools: ${e.availableToolIds.join(", ")}`,"Do not invent, print, or call tools that are not in this list."]:[],"Do not narrate intended future tool calls as a final answer.",...t.length>0?["","Recent executed tool evidence:",...t]:[],"","Previous invalid final answer:",e.output])}if(r(e.output,e.events)){const t=recentToolEvidence(e.events,1e3);return recoverRequest(e.request,["Stable runtime recovery: your previous final answer copied an executed tool result JSON as the user-facing answer.","Continue the same user request from the existing tool evidence.","If more evidence is required, use one remaining declared tool or subagent action through the backend's normal structured mechanism.",...e.availableToolIds?.length?[`Available configured tools: ${e.availableToolIds.join(", ")}`,"Do not invent, print, or call tools that are not in this list."]:[],"Otherwise synthesize a human-readable final answer from the executed tool evidence.","Do not return the raw tool result JSON as the final answer.",...t.length>0?["","Recent executed tool evidence:",...t]:[],"","Previous invalid final answer:",e.output])}if(containsToolExecutionErrorText(e.output,n)){const t=recentToolEvidence(e.events,1e3);return recoverRequest(e.request,["Stable runtime recovery: your previous final answer exposed a backend tool execution error instead of handling it.","Continue the same user request using the backend's normal structured tool-calling mechanism.","Do not retry the same invalid tool arguments. If the failed tool is not required to answer the user, synthesize the final answer from the available context instead.",...e.availableToolIds?.length?[`Available configured tools: ${e.availableToolIds.join(", ")}`,"Do not invent, print, or call tools that are not in this list."]:[],"Do not print tool error stacks, schema validation diagnostics, raw tool-call markup, JSON tool-call envelopes, or pseudo tool-call text in the final answer.",...t.length>0?["","Recent executed tool evidence:",...t]:[],"","Previous invalid final answer:",e.output])}const s=t(e);if(s)return recoverRequest(e.request,s);const a=function lastConfiguredEventHint(e,t){const o=function readEventRecoveryHints(e){return(Array.isArray(e)?e:[]).flatMap(e=>isRecord(e)&&"string"==typeof e.instruction?[{..."string"==typeof e.toolId?{toolId:e.toolId}:{},..."string"==typeof e.phase?{phase:e.phase}:{},..."string"==typeof e.outputIncludes?{outputIncludes:e.outputIncludes}:{},..."string"==typeof e.outputMatches?{outputMatches:e.outputMatches}:{},instruction:e.instruction}]:[])}(t.eventRecoveryHints);if(0!==o.length)return e.flatMap(e=>function readMatchingHints(e,t){const o="runtime.adapter.event"===e.type&&isRecord(e.event)?e.event:void 0;return o?t.filter(e=>function eventMatchesHint(e,t){return(!t.toolId||e.toolId===t.toolId)&&(!t.phase||e.phase===t.phase)&&(t.outputIncludes?"string"==typeof e.output&&e.output.includes(t.outputIncludes):!t.outputMatches||"string"==typeof e.output&&new RegExp(t.outputMatches,"u").test(e.output))}(o,e)).map(e=>({output:"string"==typeof o.output?o.output:"Adapter event matched configured recovery hint.",instruction:e.instruction})):[]}(e,o)).at(-1)}(e.events,n);return a?recoverRequest(e.request,["Stable runtime recovery: a previous adapter event matched a configured recovery hint.",a.output,a.instruction]):void 0}export function buildExecutionContractRecoveryRequest(e){if(!0!==readToolCallRecovery(e.policy).enabled)return;const t=function lastMissingEvidenceTools(e){for(let t=e.length-1;t>=0;t-=1){const o=e[t];if("runtime.execution.contract.failed"===o?.type)return readStringArray(o.missingEvidenceTools)}return[]}(e.events);return 0!==t.length?recoverRequest(e.request,["Stable runtime recovery: the execution contract was not satisfied.",`Required evidence tool(s) were missing: ${t.join(", ")}`,"Continue the same user request by calling the missing required evidence tool(s) through the backend's normal structured tool-calling mechanism.","Do not produce a final answer until the required evidence tool call has executed and you have synthesized its result.","Do not print XML, JSON, markdown fences, pseudo tool-call text, plans, or future-intent text in the final answer."]):void 0}export function assertNoRawToolCallOutput(e,t){if(containsRawToolCallOutput(e,t))throw new Error(`Adapter returned raw tool-call text as the final answer after recovery. The backend must execute tools instead of printing tool-call markup. Output preview: ${previewOutput(e)}`)}export function containsRawToolCallOutput(e,t){const o=readToolCallRecovery(t);return!0===o.enabled&&containsRawToolCallText(e,o)}export function containsRecoverableResultOutput(e,t){const o=readToolCallRecovery(t);return!0===o.enabled&&containsRecoverableResultFailureText(e,o)}export function assertNoRawToolResultOutput(e,t,o){if(!0===readToolCallRecovery(o).enabled&&r(e,t))throw new Error(`Adapter returned raw tool result JSON as the final answer after recovery. The backend must synthesize a user-facing answer. Output preview: ${previewOutput(e)}`)}export function assertNoToolExecutionErrorOutput(e,t){const o=readToolCallRecovery(t);if(!0===o.enabled&&containsToolExecutionErrorText(e,o))throw new Error(`Adapter returned a tool execution error as the final answer after recovery. Output preview: ${previewOutput(e)}`)}export function assertNoProgressOnlyToolIntentOutput(e,t,r){if(!0===readToolCallRecovery(r).enabled&&o(e,t))throw new Error(n(e))}export function rawToolCallFailureMessage(){return["The model attempted to call a tool but returned the tool call as text instead of executing it.","Please retry the request or use a model/backend configuration with reliable tool calling for this workspace."].join(" ")}export function buildEvidenceSynthesisOutput(t){const o=readToolCallRecovery(t.policy);if(!0!==o.enabled||!1===o.synthesizeFromEvidenceOnFailure||!containsRecoverableResultFailureText(t.output,o))return;const n=function latestDelegatedTaskReport(e){return e.flatMap(e=>{const t="runtime.adapter.event"===e.type&&isRecord(e.event)?e.event:void 0;if(!isToolResultEvent(t)||"task"!==t.toolId)return[];const o="string"==typeof t.output?t.output.trim():"";return function looksLikeFinalReport(e){const t=e.trim();return!(t.length<80)&&(/^#{1,3}\s+\S/mu.test(t)||/\n#{1,3}\s+\S/mu.test(t)||/\n-{3,}\n/u.test(t))}(o)?[o]:[]}).at(-1)||void 0}(t.events);if(n)return n;const r=recentToolEvidence(t.events,6e3);if(0===r.length)return;const s=/\p{Script=Han}/u.test(t.request.input)?"zh":"en",a=e(t.events);return"zh"===s?function buildChineseEvidenceSynthesis(e,t,o){return["上游模型在已有工具证据后仍输出了伪工具调用；runtime 已拒绝该 raw 输出，并直接交付已执行工具返回的证据结果。","","已执行的工具证据：",...t,...o.length>0?["","未解决的证据缺口：",...o.map(e=>`- ${e}`)]:[],"",`被拒绝的最终输出预览：${previewRejectedOutput(e)}`].join("\n")}(t.output,r,a):function buildEnglishEvidenceSynthesis(e,t,o){return["The upstream model still returned pseudo tool-call text after tool evidence was available. The runtime rejected that raw output and is returning the executed tool evidence directly.","","Executed tool evidence:",...t,...o.length>0?["","Unresolved evidence gaps:",...o.map(e=>`- ${e}`)]:[],"",`Rejected final output preview: ${previewRejectedOutput(e)}`].join("\n")}(t.output,r,a)}function previewOutput(e){const t=e.replace(/\s+/gu," ").trim();return t.length>300?`${t.slice(0,297)}...`:t}function previewRejectedOutput(e){return previewOutput(e).replace(/[<>]/gu,"")}export function rawToolCallOutputPreview(e){return previewOutput(e)}function recoverRequest(e,t){return{...e,input:[e.input,"",...t].join("\n"),metadata:{...e.metadata,stableHarnessRecovery:"tool_call"}}}function containsRawToolCallText(e,t){const o=readRegexps(t.rawOutputPatterns,[/\{\s*"name"\s*:\s*"[^"]+"\s*,\s*"arguments"\s*:/iu,/\{\s*"tool_name"\s*:\s*"[^"]+"\s*,\s*"parameters"\s*:/iu,/\{\s*"type"\s*:\s*"[^"]+"\s*,\s*"args"\s*:/iu,/^\s*[A-Za-z_][\w.-]*\s*\([^)]{0,2000}\)\s*$/iu,/^\s*[A-Za-z_][\w.-]*(?:_command|_tool|_analysis|_investigate|task)\s*$/iu,/```(?:json)?[\s\S]{0,4000}"(?:tool_name|tool|name|subagent_type)"\s*:/iu,/```(?:json)?[\s\S]{0,4000}"(?:tool_name|tool|name|subagent_type)"\s*:[\s\S]{0,4000}"(?:arguments|parameters|task)"\s*:/iu,/```(?:json)?[\s\S]{0,2000}"query"\s*:[\s\S]{0,2000}"(?:max_results|count|freshness|market)"\s*:/iu]);return!![/<\s*(?:tool_call|task)\b[^>]*>/iu,/<\s*\/\s*(?:tool_call|task)\s*>/iu,/<\s*\/?\s*tool_code\b[^>]*>/iu,/<\s*[A-Za-z_][\w.-]*\s*\([^>]{0,2000}\)\s*>/iu,/<\s*\/?\s*[A-Za-z_][\w.-]*(?:_command|_tool|_analysis|_investigate|_todos|task)\b[^>]*>/iu].some(t=>t.test(e))||function looksLikeStandaloneRecoveryCandidate(e){const t=e.trim();return t.length<=6e3||/^\s*(?:```|\{|\[|[A-Za-z_][\w.-]*\s*\()/u.test(t)}(e)&&(o.some(t=>t.test(e))||[/^[\s\S]{0,2400}\b(?:I need to|I will|I'll|I am going to|I'm going to)\s+(?:call|use|invoke|delegate)\b[\s\S]{0,1200}\b(?:tool|function|specialist|subagent)\b[\s\S]{0,400}$/iu,/^[\s\S]{0,2400}\b(?:I will|I'll|I am going to|I'm going to)\s+(?:investigate|gather|check)\b[\s\S]{0,1200}\b(?:evidence|cluster|system|results?)\b[\s\S]{0,400}$/iu,/^[\s\S]{0,2400}\b(?:waiting for|wait for)\b[\s\S]{0,1200}\b(?:tool|function|specialist|subagent)\b[\s\S]{0,400}$/iu,/^[\s\S]{0,2400}\b(?:Would you like me to|Do you want me to|Should I|I can help with)\b[\s\S]{0,1200}\?[\s\S]{0,400}$/iu,/^[\s\S]{0,2400}\bCould you please provide\b[\s\S]{0,1200}\b(?:tool|function|specialist|subagent|task)\b[\s\S]{0,400}$/iu,/^[\s\S]{0,2400}\b(?:I don't|I do not) have enough information\b[\s\S]{0,1200}\b(?:tool|function|specialist|subagent|task|context)\b[\s\S]{0,400}$/iu,/^[\s\S]{0,2400}\bLet me\s+(?:call|use|invoke|delegate|check|run|verify|gather|inspect)\b[\s\S]{0,1200}\b(?:tool|function|specialist|subagent|results?|data|evidence|commands?)\b[\s\S]{0,400}$/iu,/^[\s\S]{0,2400}\b(?:Let me|I'll|I will|I am going to|I'm going to)\s+(?:start\s+by\s+)?(?:read(?:ing)?|access(?:ing)?|gather(?:ing)?|collect(?:ing)?|fetch(?:ing)?|check(?:ing)?|inspect(?:ing)?)\b[\s\S]{0,1200}\b(?:context|instructions?|workflow|pull request|PR|issue|data|evidence|details?)\b[\s\S]{0,400}$/iu,/^[\s\S]{0,2400}(?: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|读取|收集|调查|验证|查看)[\s\S]{0,1200}$/iu,/^[\s\S]{0,2400}(?:要不要|是否需要|需要我|你想让我)[\s\S]{0,1200}(?:继续|进一步|帮你|分析|检查)[\s\S]{0,1200}[?？][\s\S]{0,400}$/iu].some(t=>t.test(e)))}function containsRecoverableResultFailureText(e,t){return containsRawToolCallText(e,t)||containsToolExecutionErrorText(e,t)}function containsToolExecutionErrorText(e,t){return readRegexps(t.toolFailureOutputPatterns,[/^Error invoking tool ['"][^'"]+['"] with kwargs /iu,/Received tool input did not match expected schema/iu,/ToolMessage.*status.*error/iu]).some(t=>t.test(e))}function recentToolEvidence(e,t){return e.flatMap(e=>{const o="runtime.adapter.event"===e.type&&isRecord(e.event)?e.event:void 0;return isToolResultEvent(o)&&"string"==typeof o.toolId?function isControlToolOutput(e){if("string"!=typeof e||!e.trim().startsWith("{"))return!1;try{const t=JSON.parse(e),o=isRecord(t)?t.status:void 0;return"duplicate_tool_call"===o||"repeated_tool_call_limit"===o||"tool_argument_error"===o}catch{return!1}}(o.output)?[]:[`- ${o.toolId}: ${formatToolEvidence(o,t)}`]:[]}).slice(-5)}function isToolResultEvent(e){return"deepagents.tool_execution.result"===e?.eventType||"agent.tool.result"===e?.phase}function formatToolEvidence(e,t=1e3){return"string"==typeof e.output&&e.output.trim()?e.output.slice(0,t):"string"==typeof e.error&&e.error.trim()?`error: ${e.error.slice(0,t)}`:isRecord(e.args)?`completed with args: ${previewOutput(JSON.stringify(e.args))}`:"completed"}function readToolCallRecovery(e){if(!isRecord(e))return{};const t=isRecord(e.recovery)?e.recovery:{};return isRecord(t.toolCall)?t.toolCall:{}}function readRegexps(e,t){const o=(Array.isArray(e)?e:[]).filter(e=>"string"==typeof e&&e.length>0).map(e=>new RegExp(e,"iu"));return o.length>0?o:t}function readStringArray(e){return Array.isArray(e)?e.filter(e=>"string"==typeof e&&e.length>0):[]}function isRecord(e){return"object"==typeof e&&null!==e&&!Array.isArray(e)}

package/packages/core/dist/runtime/completion.js

@@ -1 +1 @@
-import{omittedControlGaps 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){const s=t(r,e.text).filter(t=>!function mentionsGap(t,e){const[r,s]=e.split(":"),n=t.toLowerCase(),o=function statusPhrases(t){if(!t)return[];const e=t.toLowerCase().replaceAll("_"," ");return"repeated_tool_call_limit"===t?[e,"repeat limit"]:[e]}(s);return n.includes(e.toLowerCase())||Boolean(s&&n.includes(s.toLowerCase()))||o.some(t=>n.includes(t))||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}:{}}}
+import{omittedControlGaps 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){const s=t(r,e.text).filter(t=>!function mentionsGap(t,e){const[r,s]=e.split(":"),n=t.toLowerCase(),o=function statusPhrases(t){if(!t)return[];const e=t.toLowerCase().replaceAll("_"," ");return"task_inventory_blocked"===t?[e,"workspace inventory","not in the workspace inventory"]:"repeated_tool_call_limit"===t?[e,"repeat limit"]:[e]}(s);return n.includes(e.toLowerCase())||Boolean(s&&n.includes(s.toLowerCase()))||o.some(t=>n.includes(t))||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/packages/core/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@stable-harness/core",
-  "version": "0.0.88",
+  "version": "0.0.90",
   "type": "module",
   "files": [
     "dist/**/*.js",
@@ -11,7 +11,7 @@
     ".": "./dist/index.js"
   },
   "peerDependencies": {
-    "@stable-harness/governance": "0.0.88",
-    "@stable-harness/memory": "0.0.88"
+    "@stable-harness/governance": "0.0.90",
+    "@stable-harness/memory": "0.0.90"
   }
 }

package/packages/evaluation/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@stable-harness/evaluation",
-  "version": "0.0.88",
+  "version": "0.0.90",
   "type": "module",
   "files": [
     "dist/**/*.js",
@@ -10,6 +10,6 @@
   "main": "dist/src/index.js",
   "types": "dist/src/index.d.ts",
   "peerDependencies": {
-    "@stable-harness/core": "0.0.88"
+    "@stable-harness/core": "0.0.90"
   }
 }

package/packages/governance/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@stable-harness/governance",
-  "version": "0.0.88",
+  "version": "0.0.90",
   "type": "module",
   "files": [
     "dist/**/*.js",

package/packages/memory/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@stable-harness/memory",
-  "version": "0.0.88",
+  "version": "0.0.90",
   "type": "module",
   "files": [
     "dist/**/*.js",

package/packages/protocols/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@stable-harness/protocols",
-  "version": "0.0.88",
+  "version": "0.0.90",
   "type": "module",
   "files": [
     "dist/**/*.js",
@@ -10,6 +10,6 @@
   "main": "dist/src/index.js",
   "types": "dist/src/index.d.ts",
   "peerDependencies": {
-    "@stable-harness/core": "0.0.88"
+    "@stable-harness/core": "0.0.90"
   }
 }

package/packages/tool-gateway/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@stable-harness/tool-gateway",
-  "version": "0.0.88",
+  "version": "0.0.90",
   "type": "module",
   "files": [
     "dist/**/*.js",

package/packages/workspace-yaml/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@stable-harness/workspace-yaml",
-  "version": "0.0.88",
+  "version": "0.0.90",
   "type": "module",
   "files": [
     "dist/**/*.js",
@@ -11,6 +11,6 @@
     ".": "./dist/index.js"
   },
   "peerDependencies": {
-    "@stable-harness/core": "0.0.88"
+    "@stable-harness/core": "0.0.90"
   }
 }