stable-harness 0.0.35 → 0.0.36

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "stable-harness",
-  "version": "0.0.35",
+  "version": "0.0.36",
   "type": "module",
   "description": "Stable application runtime and operator control plane for agent workspaces.",
   "license": "Apache-2.0",

package/packages/adapter-deepagents/dist/src/internal/gateway-tools.js

@@ -1 +1 @@
-import{ToolMessage as e}from"@langchain/core/messages";import{tool as t}from"@langchain/core/tools";import{isSuccessfulEvidenceOutput as o,observedToolEvidence as r,recordObservedToolEvidence as n}from"./gateway/tool-evidence.js";import{emitStructuredToolFailure as s}from"./gateway/tool-failure-events.js";export function buildGatewayTools(o,n,a,i,l=createToolRepeatState(o.workspace.runtime.toolGateway)){return o.toolGateway?a.flatMap(a=>{const u=o.toolGateway?.get(a);if(!u)return[];const c=o.workspace.tools.get(a),d=c?.schema??u.schema;return[t(async t=>async function invokeGuardedGatewayTool(t){emitToolResult(t.input,t.agentId,t.toolId,void 0);const o=function missingRequiredPlanContent(e,t){const o=readRecord(e.agent.config.executionContract);if(!0!==o?.requiresPlan)return"";const n=readStringArray(o.planEvidenceTools);if(0===n.length||n.includes(t))return"";const s=new Set(r(e).map(e=>e.toolId));return n.some(e=>s.has(e))?"":["Status: plan_required",`Evidence tool: ${t}`,`Blocker: execution contract requires a planning checkpoint from one of: ${n.join(", ")} before evidence tools run.`,"Instruction: call the planning tool first, then retry this atomic evidence tool with repaired arguments."].join("\n")}(t.input,t.toolId);if(o)return emitToolResult(t.input,t.agentId,t.toolId,o),o;const n=function missingToolDependencyContent(e,t){const o=readRecord(e.agent.config.executionContract),n=readStringArray(readRecord(o.toolDependencies)[t]);if(0===n.length)return"";const s=new Set(r(e).map(e=>e.toolId)),a=n.filter(e=>!s.has(e));return 0===a.length?"":["Status: dependency_required",`Evidence tool: ${t}`,`Blocker: this atomic evidence tool requires completed dependency evidence from: ${a.join(", ")}.`,"Instruction: complete the dependency tool first, evaluate it, then retry this atomic evidence tool."].join("\n")}(t.input,t.toolId);if(n)return emitToolResult(t.input,t.agentId,t.toolId,n),n;const a=t.repeatState?beforeToolInvoke(t.toolId,t.args,t.repeatState):void 0;if(a)return emitToolResult(t.input,t.agentId,t.toolId,a.eventOutput),a.modelOutput;const i=await async function invokeGatewayTool(t,o,r,n,a){try{if(t.toolFailureTracker?.isCircuitOpen(r))throw new Error(`Tool circuit is open: ${r}`);const e=await t.toolGateway.invoke({toolId:r,args:n,repairModel:a,context:{workspaceRoot:t.workspace.root,requestId:t.requestId,sessionId:t.sessionId,agentId:o,requestInput:t.request.input,observedEvidence:formatObservedEvidenceForToolContext(t),approvalIds:readApprovalIds(t.request.metadata)}});return t.toolFailureTracker?.recordSuccess(r),e}catch(n){if(s(t,o,r,n),function isToolArgumentValidationError(e){return e instanceof Error&&"ToolArgumentValidationError"===e.name&&"string"==typeof e.toolId&&Array.isArray(e.issues)}(n))return new e({tool_call_id:`stable-harness-${r}-argument-guard`,name:r,status:"error",content:formatToolArgumentError(n)});if(t.workspace.runtime.retry?.tools?.enabled)throw n;return new e({tool_call_id:`stable-harness-${r}-execution-error`,name:r,status:"error",content:JSON.stringify({error:"tool_execution_failed",toolId:r,message:formatError(n),retry:"Use the error as evidence, adjust the tool arguments if possible, or return a final answer with the blocker."})})}}(t.input,t.agentId,t.toolId,t.args,t.repairModel),l=i instanceof e?String(i.content):stringifyDeepAgentResult(i.output),u=t.repeatState?afterToolInvoke(t.toolId,t.args,l,i,t.repeatState):{};return emitToolResult(t.input,t.agentId,t.toolId,u.eventOutput??l),void 0!==u.modelOutput?u.modelOutput:i instanceof e?i:l}({input:o,agentId:n,toolId:a,args:t,repairModel:i,repeatState:l}),{name:a,description:buildToolDescription(c?.description??u.description??a,d,o.workspace.runtime.toolGateway,a),schema:{type:"object",additionalProperties:!0}})]}):[]}export function createToolRepeatState(e){if(function repeatGuardEnabled(e){return!0===repeatGuardConfig(e).enabled}(e))return{successfulCalls:new Map,duplicateCallCounts:new Map,latestSuccessfulOutputByTool:new Map,successfulToolCounts:new Map,toolCallCounts:new Map,maxDuplicateCallsPerTool:readPositiveInteger(repeatGuardConfig(e).maxDuplicateCallsPerTool)??3,maxCallsPerTool:readPositiveInteger(repeatGuardConfig(e).maxCallsPerTool),maxSuccessfulCallsPerTool:readPositiveInteger(repeatGuardConfig(e).maxSuccessfulCallsPerTool),maxCallsByTool:readPositiveIntegerMap(repeatGuardConfig(e).maxCallsByTool),maxSuccessfulCallsByTool:readPositiveIntegerMap(repeatGuardConfig(e).maxSuccessfulCallsByTool),returnPreviousOutputOnRepeatLimit:!0===repeatGuardConfig(e).returnPreviousOutputOnRepeatLimit}}function repeatGuardConfig(e){return isRecord(e)&&isRecord(e.repeatGuard)?e.repeatGuard:{}}function readPositiveInteger(e){return"number"==typeof e&&Number.isInteger(e)&&e>0?e:void 0}function readPositiveIntegerMap(e){return isRecord(e)?new Map(Object.entries(e).map(([e,t])=>[e,readPositiveInteger(t)]).filter(e=>void 0!==e[1])):new Map}export function beforeToolInvoke(e,t,o){const r=o.toolCallCounts.get(e)??0;o.toolCallCounts.set(e,r+1);const n=o.maxCallsByTool.get(e)??o.maxCallsPerTool;if(void 0!==n&&r>=n){const t=o.latestSuccessfulOutputByTool.get(e),r=repeatedToolCallLimitContent(e,t);return{eventOutput:r,modelOutput:repeatLimitModelOutput(r,t,o)}}const s=o.maxSuccessfulCallsByTool.get(e)??o.maxSuccessfulCallsPerTool;if(void 0!==s&&(o.successfulToolCounts.get(e)??0)>=s){const t=o.latestSuccessfulOutputByTool.get(e),r=repeatedToolCallLimitContent(e,t);return{eventOutput:r,modelOutput:repeatLimitModelOutput(r,t,o)}}const a=stableToolCallKey(e,t),i=o.successfulCalls.get(a);if(void 0!==i){const t=o.duplicateCallCounts.get(a)??0;if(o.duplicateCallCounts.set(a,t+1),void 0!==o.maxDuplicateCallsPerTool&&t>=o.maxDuplicateCallsPerTool){const t=repeatedToolCallLimitContent(e);return{eventOutput:t,modelOutput:t}}const r=function duplicateToolCallContent(e,t){return JSON.stringify({status:"duplicate_tool_call",toolId:e,instruction:"This agent already completed an equivalent tool call. Use the prior evidence instead of calling the tool again.",previousOutput:t})}(e,i);return{eventOutput:r,modelOutput:i}}}function repeatLimitModelOutput(e,t,o){return o.returnPreviousOutputOnRepeatLimit&&void 0!==t&&0!==t.trim().length?t:e}export function isToolRepeatLimitReached(e,t){if(!t)return!1;const o=t.maxCallsByTool.get(e)??t.maxCallsPerTool;if(void 0!==o&&(t.toolCallCounts.get(e)??0)>=o)return!0;const r=t.maxSuccessfulCallsByTool.get(e)??t.maxSuccessfulCallsPerTool;return void 0!==r&&(t.successfulToolCounts.get(e)??0)>=r}export function afterToolInvoke(t,r,n,s,a){return s instanceof e&&"error"===s.status?{}:o(n)?(a.successfulCalls.set(stableToolCallKey(t,r),n),a.latestSuccessfulOutputByTool.set(t,n),a.successfulToolCounts.set(t,(a.successfulToolCounts.get(t)??0)+1),{}):{}}function emitToolResult(e,t,o,r){void 0!==r&&n(e,t,o,r),e.emit({type:"runtime.adapter.event",requestId:e.requestId,sessionId:e.sessionId,agentId:t,event:void 0===r?{adapter:"deepagents",phase:"agent.tool.start",toolId:o}:{adapter:"deepagents",phase:"agent.tool.result",toolId:o,output:previewToolOutput(r),...toolControlProjection(r)}})}function repeatedToolCallLimitContent(e,t){return JSON.stringify({status:"repeated_tool_call_limit",toolId:e,instruction:"This tool reached the configured repeat limit for this request. Do not call this tool or a substitute tool for the same evidence need again. Use previousOutput and the collected evidence to produce the final answer now, or report the remaining gap explicitly.",...void 0!==t?{previousOutput:t}:{}})}export function stringifyDeepAgentResult(t){if(t instanceof e)return function stringifyToolMessageContent(e){return"string"==typeof e?e:JSON.stringify(e)}(t.content);if("string"==typeof t)return t;if(isRecord(t)){const e=t.structuredResponse??t.structured_response;if(void 0!==e)return"string"==typeof e?e:JSON.stringify(e);const o=(Array.isArray(t.messages)?t.messages:[]).at(-1);if(isRecord(o)&&"string"==typeof o.content)return o.content;const r=(isRecord(t.update)&&Array.isArray(t.update.messages)?t.update.messages:[]).at(-1);if(isRecord(r)&&isRecord(r.kwargs)&&"string"==typeof r.kwargs.content)return r.kwargs.content;if(isRecord(r)&&"string"==typeof r.content)return r.content}return JSON.stringify(t)}function buildToolDescription(e,t,o,r){const n=function toolRepeatPolicyDescription(e,t){const o=repeatGuardConfig(e),r=readPositiveIntegerMap(o.maxSuccessfulCallsByTool).get(t)??readPositiveInteger(o.maxSuccessfulCallsPerTool);return void 0===r?"":`Stable runtime repeat policy: call this tool at most ${r} successful time(s) for this request. If more detail is needed, include the dimensions in the first call and synthesize after the result returns.`}(o,r),s=n?`${e}\n\n${n}`:e;return t?`${s}\n\nStable tool input schema:\n${previewToolOutput(JSON.stringify(t))}`:s}function previewToolOutput(e){const t=e.replace(/\s+/gu," ").trim();return t.length>500?`${t.slice(0,497)}...`:t}export function toolControlProjection(e){const t=function parseJsonRecord(e){try{const t=JSON.parse(e);return isRecord(t)?t:void 0}catch{return}}(e);if("string"==typeof t?.status)return{controlStatus:t.status};const o=function readTextStatus(e){return String(e).match(/^Status:\s*([A-Za-z0-9_-]+)/imu)?.[1]}(e);return o?{controlStatus:o}:"string"==typeof t?.error?{controlStatus:t.error}:e.startsWith("Task delegation target is not in the workspace inventory")?{controlStatus:"task_inventory_blocked"}:{}}function stableToolCallKey(e,t){return`${e}:${stableJson(t)}`}function stableJson(e){return Array.isArray(e)?`[${e.map(stableJson).join(",")}]`:isRecord(e)?`{${Object.keys(e).sort().map(t=>`${JSON.stringify(t)}:${stableJson(e[t])}`).join(",")}}`:JSON.stringify(e)}function readApprovalIds(e){const t=e?.approvalIds??e?.approvalId;return"string"==typeof t&&t.trim()?[t.trim()]:Array.isArray(t)?t.filter(e=>"string"==typeof e&&e.trim().length>0):void 0}function formatObservedEvidenceForToolContext(e){const t=r(e).map(e=>`Tool: ${e.toolId}\n${e.output}`).join("\n\n---\n\n");return t.length>12e3?`${t.slice(0,12e3)}\n[truncated]`:t}function formatToolArgumentError(e){return JSON.stringify({error:"tool_argument_validation_failed",toolId:e.toolId,issues:e.issues,retry:"Call the same tool again with arguments that satisfy the reported schema and semantic issues."})}function isRecord(e){return"object"==typeof e&&null!==e&&!Array.isArray(e)}function readRecord(e){return isRecord(e)?e:{}}function readStringArray(e){return Array.isArray(e)?e.filter(e=>"string"==typeof e&&e.length>0):[]}function formatError(e){return e instanceof Error?e.message:String(e)}
+import{ToolMessage as e}from"@langchain/core/messages";import{tool as t}from"@langchain/core/tools";import{isSuccessfulEvidenceOutput as o,observedToolEvidence as r,recordObservedToolEvidence as n}from"./gateway/tool-evidence.js";import{emitStructuredToolFailure as s}from"./gateway/tool-failure-events.js";export function buildGatewayTools(o,n,a,i,l=createToolRepeatState(o.workspace.runtime.toolGateway)){return o.toolGateway?a.flatMap(a=>{const u=o.toolGateway?.get(a);if(!u)return[];const c=o.workspace.tools.get(a),d=c?.schema??u.schema;return[t(async t=>async function invokeGuardedGatewayTool(t){emitToolResult(t.input,t.agentId,t.toolId,void 0);const o=function missingRequiredPlanContent(e,t){const o=readRecord(e.agent.config.executionContract);if(!0!==o?.requiresPlan)return"";const n=readStringArray(o.planEvidenceTools);if(0===n.length||n.includes(t))return"";const s=new Set(r(e).map(e=>e.toolId));return n.some(e=>s.has(e))?"":["Status: plan_required",`Evidence tool: ${t}`,`Blocker: execution contract requires a planning checkpoint from one of: ${n.join(", ")} before evidence tools run.`,"Instruction: call the planning tool first, then retry this atomic evidence tool with repaired arguments."].join("\n")}(t.input,t.toolId);if(o)return emitToolResult(t.input,t.agentId,t.toolId,o),o;const n=function missingToolDependencyContent(e,t){const o=readRecord(e.agent.config.executionContract),n=readStringArray(readRecord(o.toolDependencies)[t]);if(0===n.length)return"";const s=new Set(r(e).map(e=>e.toolId)),a=n.filter(e=>!s.has(e));return 0===a.length?"":["Status: dependency_required",`Evidence tool: ${t}`,`Blocker: this atomic evidence tool requires completed dependency evidence from: ${a.join(", ")}.`,"Instruction: complete the dependency tool first, evaluate it, then retry this atomic evidence tool."].join("\n")}(t.input,t.toolId);if(n)return emitToolResult(t.input,t.agentId,t.toolId,n),n;const a=t.repeatState?beforeToolInvoke(t.toolId,t.args,t.repeatState):void 0;if(a)return emitToolResult(t.input,t.agentId,t.toolId,a.eventOutput),a.modelOutput;const i=await async function invokeGatewayTool(t,o,r,n,a){try{if(t.toolFailureTracker?.isCircuitOpen(r))throw new Error(`Tool circuit is open: ${r}`);const e=await t.toolGateway.invoke({toolId:r,args:n,repairModel:a,context:{workspaceRoot:t.workspace.root,requestId:t.requestId,sessionId:t.sessionId,agentId:o,requestInput:t.request.input,observedEvidence:formatObservedEvidenceForToolContext(t),approvalIds:readApprovalIds(t.request.metadata)}});return t.toolFailureTracker?.recordSuccess(r),e}catch(n){if(s(t,o,r,n),function isToolArgumentValidationError(e){return e instanceof Error&&"ToolArgumentValidationError"===e.name&&"string"==typeof e.toolId&&Array.isArray(e.issues)}(n))return new e({tool_call_id:`stable-harness-${r}-argument-guard`,name:r,status:"error",content:formatToolArgumentError(n)});if(t.workspace.runtime.retry?.tools?.enabled)throw n;return new e({tool_call_id:`stable-harness-${r}-execution-error`,name:r,status:"error",content:JSON.stringify({error:"tool_execution_failed",toolId:r,message:formatError(n),retry:"Use the error as evidence, adjust the tool arguments if possible, or return a final answer with the blocker."})})}}(t.input,t.agentId,t.toolId,t.args,t.repairModel),l=i instanceof e?String(i.content):stringifyDeepAgentResult(i.output),u=t.repeatState?afterToolInvoke(t.toolId,t.args,l,i,t.repeatState):{};return emitToolResult(t.input,t.agentId,t.toolId,u.eventOutput??l),void 0!==u.modelOutput?u.modelOutput:i instanceof e?i:l}({input:o,agentId:n,toolId:a,args:t,repairModel:i,repeatState:l}),{name:a,description:buildToolDescription(c?.description??u.description??a,d,o.workspace.runtime.toolGateway,a),schema:{type:"object",additionalProperties:!0}})]}):[]}export function createToolRepeatState(e){if(function repeatGuardEnabled(e){return!0===repeatGuardConfig(e).enabled}(e))return{successfulCalls:new Map,duplicateCallCounts:new Map,latestSuccessfulOutputByTool:new Map,successfulToolCounts:new Map,toolCallCounts:new Map,maxDuplicateCallsPerTool:readPositiveInteger(repeatGuardConfig(e).maxDuplicateCallsPerTool)??3,maxCallsPerTool:readPositiveInteger(repeatGuardConfig(e).maxCallsPerTool),maxSuccessfulCallsPerTool:readPositiveInteger(repeatGuardConfig(e).maxSuccessfulCallsPerTool),maxCallsByTool:readPositiveIntegerMap(repeatGuardConfig(e).maxCallsByTool),maxSuccessfulCallsByTool:readPositiveIntegerMap(repeatGuardConfig(e).maxSuccessfulCallsByTool),returnPreviousOutputOnRepeatLimit:!0===repeatGuardConfig(e).returnPreviousOutputOnRepeatLimit}}function repeatGuardConfig(e){return isRecord(e)&&isRecord(e.repeatGuard)?e.repeatGuard:{}}function readPositiveInteger(e){return"number"==typeof e&&Number.isInteger(e)&&e>0?e:void 0}function readPositiveIntegerMap(e){return isRecord(e)?new Map(Object.entries(e).map(([e,t])=>[e,readPositiveInteger(t)]).filter(e=>void 0!==e[1])):new Map}export function beforeToolInvoke(e,t,o){const r=o.toolCallCounts.get(e)??0;o.toolCallCounts.set(e,r+1);const n=o.maxCallsByTool.get(e)??o.maxCallsPerTool;if(void 0!==n&&r>=n){const t=o.latestSuccessfulOutputByTool.get(e),r=repeatedToolCallLimitContent(e,t);return{eventOutput:r,modelOutput:repeatLimitModelOutput(r,t,o)}}const s=o.maxSuccessfulCallsByTool.get(e)??o.maxSuccessfulCallsPerTool;if(void 0!==s&&(o.successfulToolCounts.get(e)??0)>=s){const t=o.latestSuccessfulOutputByTool.get(e),r=repeatedToolCallLimitContent(e,t);return{eventOutput:r,modelOutput:repeatLimitModelOutput(r,t,o)}}const a=stableToolCallKey(e,t),i=o.successfulCalls.get(a);if(void 0!==i){const t=o.duplicateCallCounts.get(a)??0;if(o.duplicateCallCounts.set(a,t+1),void 0!==o.maxDuplicateCallsPerTool&&t>=o.maxDuplicateCallsPerTool){const t=repeatedToolCallLimitContent(e);return{eventOutput:t,modelOutput:t}}const r=function duplicateToolCallContent(e,t){return JSON.stringify({status:"duplicate_tool_call",toolId:e,instruction:"This agent already completed an equivalent tool call. Use the prior evidence instead of calling the tool again.",previousOutput:t})}(e,i);return{eventOutput:r,modelOutput:i}}}function repeatLimitModelOutput(e,t,o){return o.returnPreviousOutputOnRepeatLimit&&void 0!==t&&0!==t.trim().length?t:e}export function isToolRepeatLimitReached(e,t){if(!t)return!1;const o=t.maxCallsByTool.get(e)??t.maxCallsPerTool;if(void 0!==o&&(t.toolCallCounts.get(e)??0)>=o)return!0;const r=t.maxSuccessfulCallsByTool.get(e)??t.maxSuccessfulCallsPerTool;return void 0!==r&&(t.successfulToolCounts.get(e)??0)>=r}export function afterToolInvoke(t,r,n,s,a){return s instanceof e&&"error"===s.status?{}:o(n)?(a.successfulCalls.set(stableToolCallKey(t,r),n),a.latestSuccessfulOutputByTool.set(t,n),a.successfulToolCounts.set(t,(a.successfulToolCounts.get(t)??0)+1),{}):{}}function emitToolResult(e,t,o,r){void 0!==r&&n(e,t,o,r),e.emit({type:"runtime.adapter.event",requestId:e.requestId,sessionId:e.sessionId,agentId:t,event:void 0===r?{adapter:"deepagents",phase:"agent.tool.start",toolId:o}:{adapter:"deepagents",phase:"agent.tool.result",toolId:o,output:previewToolOutput(r),evidenceOutput:r,...toolControlProjection(r)}})}function repeatedToolCallLimitContent(e,t){return JSON.stringify({status:"repeated_tool_call_limit",toolId:e,instruction:"This tool reached the configured repeat limit for this request. Do not call this tool or a substitute tool for the same evidence need again. Use previousOutput and the collected evidence to produce the final answer now, or report the remaining gap explicitly.",...void 0!==t?{previousOutput:t}:{}})}export function stringifyDeepAgentResult(t){if(t instanceof e)return function stringifyToolMessageContent(e){return"string"==typeof e?e:JSON.stringify(e)}(t.content);if("string"==typeof t)return t;if(isRecord(t)){const e=t.structuredResponse??t.structured_response;if(void 0!==e)return"string"==typeof e?e:JSON.stringify(e);const o=(Array.isArray(t.messages)?t.messages:[]).at(-1);if(isRecord(o)&&"string"==typeof o.content)return o.content;const r=(isRecord(t.update)&&Array.isArray(t.update.messages)?t.update.messages:[]).at(-1);if(isRecord(r)&&isRecord(r.kwargs)&&"string"==typeof r.kwargs.content)return r.kwargs.content;if(isRecord(r)&&"string"==typeof r.content)return r.content}return JSON.stringify(t)}function buildToolDescription(e,t,o,r){const n=function toolRepeatPolicyDescription(e,t){const o=repeatGuardConfig(e),r=readPositiveIntegerMap(o.maxSuccessfulCallsByTool).get(t)??readPositiveInteger(o.maxSuccessfulCallsPerTool);return void 0===r?"":`Stable runtime repeat policy: call this tool at most ${r} successful time(s) for this request. If more detail is needed, include the dimensions in the first call and synthesize after the result returns.`}(o,r),s=n?`${e}\n\n${n}`:e;return t?`${s}\n\nStable tool input schema:\n${previewToolOutput(JSON.stringify(t))}`:s}function previewToolOutput(e){const t=e.replace(/\s+/gu," ").trim();return t.length>500?`${t.slice(0,497)}...`:t}export function toolControlProjection(e){const t=function parseJsonRecord(e){try{const t=JSON.parse(e);return isRecord(t)?t:void 0}catch{return}}(e);if("string"==typeof t?.status)return{controlStatus:t.status};const o=function readTextStatus(e){return String(e).match(/^Status:\s*([A-Za-z0-9_-]+)/imu)?.[1]}(e);return o?{controlStatus:o}:"string"==typeof t?.error?{controlStatus:t.error}:e.startsWith("Task delegation target is not in the workspace inventory")?{controlStatus:"task_inventory_blocked"}:{}}function stableToolCallKey(e,t){return`${e}:${stableJson(t)}`}function stableJson(e){return Array.isArray(e)?`[${e.map(stableJson).join(",")}]`:isRecord(e)?`{${Object.keys(e).sort().map(t=>`${JSON.stringify(t)}:${stableJson(e[t])}`).join(",")}}`:JSON.stringify(e)}function readApprovalIds(e){const t=e?.approvalIds??e?.approvalId;return"string"==typeof t&&t.trim()?[t.trim()]:Array.isArray(t)?t.filter(e=>"string"==typeof e&&e.trim().length>0):void 0}function formatObservedEvidenceForToolContext(e){const t=r(e).map(e=>`Tool: ${e.toolId}\n${e.output}`).join("\n\n---\n\n");return t.length>12e3?`${t.slice(0,12e3)}\n[truncated]`:t}function formatToolArgumentError(e){return JSON.stringify({error:"tool_argument_validation_failed",toolId:e.toolId,issues:e.issues,retry:"Call the same tool again with arguments that satisfy the reported schema and semantic issues."})}function isRecord(e){return"object"==typeof e&&null!==e&&!Array.isArray(e)}function readRecord(e){return isRecord(e)?e:{}}function readStringArray(e){return Array.isArray(e)?e.filter(e=>"string"==typeof e&&e.length>0):[]}function formatError(e){return e instanceof Error?e.message:String(e)}

package/packages/core/dist/index.d.ts

@@ -4,6 +4,7 @@ export * from "./boundary-scan.js";
 export * from "./execution-contract.js";
 export * from "./recovery/tool-call.js";
 export * from "./runtime/persistence/inspection.js";
+export { createWorkspaceSandboxPolicy } from "./runtime/governance/sandbox.js";
 export * from "./memory-plugins.js";
 export * from "./runtime/persistence/queue.js";
 export * from "./runtime.js";

package/packages/core/dist/index.js

@@ -1 +1 @@
-export*from"./runtime/persistence/artifacts.js";export*from"./boundary-scan.js";export*from"./execution-contract.js";export*from"./recovery/tool-call.js";export*from"./runtime/persistence/inspection.js";export*from"./memory-plugins.js";export*from"./runtime/persistence/queue.js";export*from"./runtime.js";export*from"./runtime/selection-repair.js";export*from"./runtime/tool-failure.js";export*from"./runtime/persistence/stores.js";export*from"./trace.js";export*from"./types.js";export*from"./workspace/tool-quality.js";export*from"./evaluations/index.js";export*from"./quality/index.js";export*from"./spec-driven/index.js";export*from"./workflows/index.js";
+export*from"./runtime/persistence/artifacts.js";export*from"./boundary-scan.js";export*from"./execution-contract.js";export*from"./recovery/tool-call.js";export*from"./runtime/persistence/inspection.js";export{createWorkspaceSandboxPolicy}from"./runtime/governance/sandbox.js";export*from"./memory-plugins.js";export*from"./runtime/persistence/queue.js";export*from"./runtime.js";export*from"./runtime/selection-repair.js";export*from"./runtime/tool-failure.js";export*from"./runtime/persistence/stores.js";export*from"./trace.js";export*from"./types.js";export*from"./workspace/tool-quality.js";export*from"./evaluations/index.js";export*from"./quality/index.js";export*from"./spec-driven/index.js";export*from"./workflows/index.js";

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 e&&"agent.tool.result"===e.phase&&"string"==typeof e.toolId&&isSuccessfulEvidenceEvent(e)?[e.toolId]:[]});return[...new Set(e)]}export function successfulEvidenceOutputs(t){return successfulEvidenceItems(t).map(t=>t.output)}export function successfulEvidenceItems(t){return t.flatMap(t=>{if("runtime.tool.direct.completed"===t.type)return stringifyEvidence(t.output).map(e=>({source:t.toolId,output:e}));const e=readAdapterEvent(t);return e&&"agent.tool.result"===e.phase&&"string"==typeof e.toolId?!isSuccessfulEvidenceEvent(e)||function isPlanningTool(t){return/^(?:write_todos|read_todos)$/u.test(t)}(e.toolId)?[]:stringifyEvidence(e.output).filter(t=>function isUsableEvidenceOutput(t,e){return"task"!==t||!function looksLikeUnexecutedToolIntent(t){const e=t.trim();return!(e.length>4e3)&&[/\b(?:I need to|I will|I'll|I am going to|I'm going to|Let me)\s+(?:call|use|run|invoke|get|fetch|check)\b[\s\S]{0,1200}\b(?:tool|function|system_time|web_search|url_text_fetch|task)\b/iu,/(?:我需要|我要|我会|我将|让我|我来|接下来我(?:会|将)?)\s*(?:先)?(?:调用|使用|运行|执行|检查|获取|查看)\b[\s\S]{0,1200}\b(?:工具|函数|system_time|web_search|url_text_fetch|task)\b/iu,/```(?:json)?\s*(?:system_time|web_search|url_text_fetch|task)\s*```/iu].some(t=>t.test(e))}(e)&&!function looksLikeDelegatedCommentary(t){const e=t.trim();return!(e.length>8e3)&&[/(?:这篇文章|this article)[\s\S]{0,1200}(?:我认为|我认同|打动我|值得深入探讨|特别认同|what I find|I think|I agree)/iu,/(?:你还有什么|还想深入探讨|would you like|do you want).{0,120}[?？]\s*$/iu].some(t=>t.test(e))}(e)}(e.toolId,t)).map(t=>({source:e.toolId,output:t})):[]})}export function controlBlockers(t){const e=function successfulEventIndexesBySource(t){const e=new Map;return t.forEach((t,n)=>{const o=function successfulEventSource(t){if("runtime.tool.direct.completed"===t.type)return t.toolId;const e=readAdapterEvent(t);return e&&"agent.tool.result"===e.phase&&"string"==typeof e.toolId&&isSuccessfulEvidenceEvent(e)?e.toolId:void 0}(t);o&&e.set(o,[...e.get(o)??[],n])}),e}(t);return t.flatMap((t,n)=>{const o=readAdapterEvent(t),r=readString(o?.controlStatus)??readOutputStatus(o?.output),s=readString(o?.toolId)??"tool";return r&&isBlockerStatus(r)?isResolvedByLaterCompletion(r)&&function completedAfter(t,e,n){return(t.get(e)??[]).some(t=>t>n)}(e,s,n)?[]:[`${s}:${r}`]:[]})}export function controlGaps(t){const e=new Set(successfulEvidenceItems(t).map(t=>t.source));return t.flatMap(t=>{const n=readAdapterEvent(t),o=readString(n?.controlStatus)??readOutputStatus(n?.output),r=readString(n?.toolId)??"tool";return o&&isGapStatus(o)?e.has(r)&&isResolvedByLaterCompletion(o)?[]:[`${r}:${o}`]:[]})}function stringifyEvidence(t){return"string"==typeof t?t.trim()?[t]:[]:null==t?[]:[JSON.stringify(t)]}function readAdapterEvent(t){return"runtime.adapter.event"===t.type&&isRecord(t.event)?t.event:void 0}function isSuccessfulEvidenceEvent(t){const e=readString(t.controlStatus)??readOutputStatus(t.output);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)$/iu.test(t)}function isResolvedByLaterCompletion(t){return isGapStatus(t)||isBlockerStatus(t)}function readOutputStatus(t){if("string"!=typeof t)return;const e=function parseJsonRecord(t){try{const e=JSON.parse(t);return isRecord(e)?e:void 0}catch{return}}(t);return"string"==typeof e?.status?e.status:t.match(/^Status:\s*([A-Za-z0-9_-]+)/imu)?.[1]}function isRecord(t){return"object"==typeof t&&null!==t&&!Array.isArray(t)}function readString(t){return"string"==typeof t&&t.trim()?t:void 0}
+export function hasPlanningEvidence(t){return t.some(t=>{const e=readAdapterEvent(t);return!!e&&("plan"===e.traceType||String(e.traceLabel??"").startsWith("plan.")||"write_todos"===e.toolId&&String(e.phase??"").startsWith("agent.tool."))})}export function successfulEvidenceToolIds(t){const e=t.flatMap(t=>{if("runtime.tool.direct.completed"===t.type)return[t.toolId];const e=readAdapterEvent(t);return e&&"agent.tool.result"===e.phase&&"string"==typeof e.toolId&&isSuccessfulEvidenceEvent(e)?[e.toolId]:[]});return[...new Set(e)]}export function successfulEvidenceOutputs(t){return successfulEvidenceItems(t).map(t=>t.output)}export function successfulEvidenceItems(t){return t.flatMap(t=>{if("runtime.tool.direct.completed"===t.type)return stringifyEvidence(t.output).map(e=>({source:t.toolId,output:e}));const e=readAdapterEvent(t);return e&&"agent.tool.result"===e.phase&&"string"==typeof e.toolId?!isSuccessfulEvidenceEvent(e)||function isPlanningTool(t){return/^(?:write_todos|read_todos)$/u.test(t)}(e.toolId)?[]: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*$/iu].some(t=>t.test(e))}(e)}(e.toolId,t)).map(t=>({source:e.toolId,output:t})):[]})}export function controlBlockers(t){const e=function successfulEventIndexesBySource(t){const e=new Map;return t.forEach((t,n)=>{const o=function successfulEventSource(t){if("runtime.tool.direct.completed"===t.type)return t.toolId;const e=readAdapterEvent(t);return e&&"agent.tool.result"===e.phase&&"string"==typeof e.toolId&&isSuccessfulEvidenceEvent(e)?e.toolId:void 0}(t);o&&e.set(o,[...e.get(o)??[],n])}),e}(t);return t.flatMap((t,n)=>{const o=readAdapterEvent(t),r=readString(o?.controlStatus)??readOutputStatus(o?.output),s=readString(o?.toolId)??"tool";return r&&isBlockerStatus(r)?isResolvedByLaterCompletion(r)&&function completedAfter(t,e,n){return(t.get(e)??[]).some(t=>t>n)}(e,s,n)?[]:[`${s}:${r}`]:[]})}export function controlGaps(t){const e=new Set(successfulEvidenceItems(t).map(t=>t.source));return t.flatMap(t=>{const n=readAdapterEvent(t),o=readString(n?.controlStatus)??readOutputStatus(n?.output),r=readString(n?.toolId)??"tool";return o&&isGapStatus(o)?e.has(r)&&isResolvedByLaterCompletion(o)?[]:[`${r}:${o}`]:[]})}function stringifyEvidence(t){return"string"==typeof t?t.trim()?[t]:[]:null==t?[]:[JSON.stringify(t)]}function readAdapterEvent(t){return"runtime.adapter.event"===t.type&&isRecord(t.event)?t.event:void 0}function isSuccessfulEvidenceEvent(t){const e=readString(t.controlStatus)??readOutputStatus(t.output);return!e||/^(?:completed|success|ok|recorded)$/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 readOutputStatus(t){if("string"!=typeof t)return;const e=function parseJsonRecord(t){try{const e=JSON.parse(t);return isRecord(e)?e:void 0}catch{return}}(t);return"string"==typeof e?.status?e.status:t.match(/^Status:\s*([A-Za-z0-9_-]+)/imu)?.[1]}function isRecord(t){return"object"==typeof t&&null!==t&&!Array.isArray(t)}function readString(t){return"string"==typeof t&&t.trim()?t:void 0}

package/packages/core/dist/quality/synthesis/fields.js

@@ -1 +1 @@
-export function evidenceFields(e){const t=new Set,l=[];for(const a of e)for(const e of parseEvidenceFields(a)){const a=`${normalizeFieldLabel(e.label)}\0${e.value.toLowerCase()}`;t.has(a)||(t.add(a),l.push(e))}return l}export function selectSectionFields(e,t,l){return function withSourceCoverage(e,t){const l=[...e].sort((e,t)=>scoreField(t)-scoreField(e)),a=new Map;for(const e of[...new Set(l.map(e=>e.source))]){const t=l.find(t=>t.source===e);t&&a.set(fieldKey(t),t)}for(const e of l)if(a.set(fieldKey(e),e),a.size>=t)break;return[...a.values()].slice(0,t)}(e.filter(e=>e.kind===t),l).sort((e,t)=>scoreField(t)-scoreField(e)||e.label.localeCompare(t.label))}export function fieldTableLines(e,t){return 0===e.length?[]:[..."zh"===t?["| 项目 | 数值 | 来源 |","|---|---:|---|"]:["| Field | Value | Source |","|---|---:|---|"],...e.map(e=>`| ${escapeTableCell(e.label)} | ${escapeTableCell(e.value)} | ${escapeTableCell(function humanSourceLabel(e){const t=e.replace(/([a-z])([A-Z])/gu,"$1 $2").split(/[_:.\-\s]+/u).map(e=>e.trim()).filter(Boolean);return 0===t.length?e:t.map(e=>e.charAt(0).toUpperCase()+e.slice(1)).join(" ")}(e.source))} |`)]}export function hasExtractableFields(e){return extractKeyValueFields(e).some(e=>!isBoilerplateField(e.label)&&e.value.length>0)}export function classifyFact(e){return/(?:gap|blocked|missing|unavailable|unsupported|缺口|阻塞|缺失|不支持|无法|未提供)/iu.test(e)?"limit":/(?:news|headline|recent|latest|filing|event|公告|新闻|近期|最新|披露|\b20\d{2}[-/年])/iu.test(e)?"timeBound":/(?:[$€£¥%]|\b\d[\d,]*(?:\.\d+)?\b)/u.test(e)?"data":/(?:name|company|symbol|resolved|overview|query|名称|公司|代码|概览|识别)/iu.test(e)?"context":"other"}function parseEvidenceFields(e){return extractKeyValueFields(e.text).filter(e=>e.value.length>0&&!isBoilerplateField(e.label)).filter(e=>!function isLowValueField(e,t){return!!/(?:snapshot for|source type|loaded sources)/iu.test(e)||!!/^\d{1,2}\s+(?:jan|feb|mar|apr|may|jun|jul|aug|sep|oct|nov|dec)\b/iu.test(e)||scoreField({source:"",label:e,value:t,kind:"other"})<0}(e.label,e.value)).map(t=>{const l=classifyFact(`${t.label}: ${t.value}`);return{source:e.source,label:titleCaseLabel(t.label),value:formatEvidence(t.value),kind:"other"===l?e.kind:l}})}function scoreField(e){const t=normalizeFieldLabel(e.label);let l=e.value.length>0?10:0;return/(?:price|value|amount|total|ratio|rate|range|target|estimate|revenue|income|earnings|margin|volume|date|rating|status|count|metric|价格|收入|利润|目标|评级|数量|日期|范围|比率)/iu.test(t)&&(l+=8),/(?:name|company|symbol|title|headline|summary|名称|公司|代码|标题|摘要)/iu.test(t)&&(l+=6),/(?:unix|epoch|timestamp|raw|html|url|href|source|id|identifier|internal|debug)/iu.test(t)&&(l-=20),(/^https?:\/\//iu.test(e.value)||"/"===e.value)&&(l-=12),e.value.length>240&&(l-=5),l}function extractKeyValueFields(e){const t=[...e.matchAll(/(^|[\s;|])([\p{L}\p{N}][\p{L}\p{N} _./()%&+\-]{0,48})[:：](?!\/\/)\s*/gu)].filter(e=>void 0!==e.index).map((e,t,l)=>{l[t-1];const a=function cleanFieldLabel(e){const t=e.replace(/^(?:[-*]\s*)+/u,"").replace(/\s+/gu," ").trim(),l=t.match(/^.+\b(PE Ratio TTM)$/iu);if(l?.[1]&&l[1]!==t&&/(?:beta|\b20\d{2}|\b(?:jan|feb|mar|apr|may|jun|jul|aug|sep|oct|nov|dec)\b)/iu.test(t))return l[1];const a=t.split(" ").filter(Boolean);return a.length>1&&/^\d{4}$/u.test(a[0]??"")||a.length>1&&/^\d+(?:\.\d+)?$/u.test(a[0]??"")&&/^(?:\d+|beta|pe|p\/b)$/iu.test(a[1]??"")||a.length>1&&/^[A-Z]$/u.test(a[0]??"")||a.length>1&&/^(?:inc\.?|llc\.?|ltd\.?|corp\.?|corporation|co\.?)$/iu.test(a[0]??"")||a.length>1&&/^(?:usd|eur|gbp|jpy|cny|cad|aud)$/iu.test(a[0]??"")||a.length>1&&/^(?:gmt|utc)$/iu.test(a[0]??"")?a.slice(1).join(" "):function trailingKnownLabel(e){if(/\bPE Ratio TTM$/iu.test(e)&&!/^PE Ratio TTM$/iu.test(e))return"PE Ratio TTM";const t=e.match(/\b(?:Revenue|Net Income|Operating Income|Diluted EPS|EPS|Market Cap|Book Value|Target Estimate|Target Mean Price|Target Median Price|Target High Price|Target Low Price|PE Ratio TTM|P\/B|Beta(?: 5Y Monthly)?|Published|Updated|Date|Time|Summary|Title|Headline)$/iu);return t&&/(?:filed|\b20\d{2}|10-k|10-q|\bfy\b|\b(?:jan|feb|mar|apr|may|jun|jul|aug|sep|oct|nov|dec)\b|\s{3,}|[a-z].{8,}\s)/iu.test(e.slice(0,t.index))?t?.[0]??e:e}(t)}(e[2]??""),r=function droppedLabelPrefix(e,t){const l=e.trim();if(l===t)return"";const a=l.toLowerCase().lastIndexOf(t.toLowerCase());return a>0?l.slice(0,a):""}(e[2]??"",a);return{index:(e.index??0)+String(e[1]??"").length+r.length,start:(e.index??0)+e[0].length,label:a}}).filter(e=>!function isBoundaryNoiseLabel(e){return/^\d{1,2}\s+(?:jan|feb|mar|apr|may|jun|jul|aug|sep|oct|nov|dec)\b/iu.test(e)}(e.label));return t.map((l,a)=>{const r=t[a+1]?.index??e.length;return{label:l.label,value:cleanFieldValue(e.slice(l.start,r),l.label)}}).filter(e=>e.label.length>0)}function cleanFieldValue(e,t){const l=e.replace(/^[\s,;|.-]+/u,"").replace(/[\s,;|.-]+$/u,"").replace(/\s+/gu," ").trim(),a=normalizeFieldLabel(t);if(/\brange\b/iu.test(a))return l.replace(/\s+Beta\b.*$/iu,"").trim();if(/^(?:ticker|symbol|resolved symbol|resolved ticker|股票 resolved ticker|代码)$/iu.test(a)){const e=l.match(/^([A-Z0-9._-]{1,24})(?:\s+\d+[.)]\s+|$)/u);if(e?.[1])return e[1]}return l}function isBoilerplateField(e){return/^(?:status|control status|evidence tool|tool|source|来源)$/iu.test(e.trim())}export function normalizeFieldLabel(e){return e.toLowerCase().replace(/[^a-z0-9\p{L}\p{N}]+/giu," ").trim()}function titleCaseLabel(e){if(/\p{Script=Han}/u.test(e))return e;const t=e.split(/[^A-Za-z0-9]+/u).filter(Boolean);return normalizeFieldLabel(e).split(" ").filter(Boolean).map((e,l)=>{const a=t[l]??e;return/^[A-Z0-9]{2,}$/u.test(a)?a:e.charAt(0).toUpperCase()+e.slice(1)}).join(" ")}function formatEvidence(e){const t=e.replace(/\s+/gu," ").trim();return t.length>1200?`${t.slice(0,1197)}...`:t}function escapeTableCell(e){return e.replace(/\|/gu,"\\|").replace(/\r?\n/gu," ")}function fieldKey(e){return`${e.source}\0${normalizeFieldLabel(e.label)}\0${e.value.toLowerCase()}`}
+export function evidenceFields(e){const t=new Set,l=[];for(const a of e)for(const e of parseEvidenceFields(a)){const a=`${normalizeFieldLabel(e.label)}\0${e.value.toLowerCase()}`;t.has(a)||(t.add(a),l.push(e))}return l}export function selectSectionFields(e,t,l){return function withSourceCoverage(e,t){const l=[...e].sort((e,t)=>scoreField(t)-scoreField(e)),a=new Map;for(const e of[...new Set(l.map(e=>e.source))]){const t=l.find(t=>t.source===e);t&&a.set(fieldKey(t),t)}for(const e of l)if(a.set(fieldKey(e),e),a.size>=t)break;return[...a.values()].slice(0,t)}(e.filter(e=>e.kind===t),l).sort((e,t)=>scoreField(t)-scoreField(e)||e.label.localeCompare(t.label))}export function fieldTableLines(e,t){return 0===e.length?[]:[..."zh"===t?["| 项目 | 数值 | 来源 |","|---|---:|---|"]:["| Field | Value | Source |","|---|---:|---|"],...e.map(e=>`| ${escapeTableCell(e.label)} | ${escapeTableCell(e.value)} | ${escapeTableCell(function humanSourceLabel(e){const t=e.replace(/([a-z])([A-Z])/gu,"$1 $2").split(/[_:.\-\s]+/u).map(e=>e.trim()).filter(Boolean);return 0===t.length?e:t.map(e=>e.charAt(0).toUpperCase()+e.slice(1)).join(" ")}(e.source))} |`)]}export function hasExtractableFields(e){return extractKeyValueFields(e).some(e=>!isBoilerplateField(e.label)&&e.value.length>0)}export function classifyFact(e){return/(?:gap|blocked|missing|unavailable|unsupported|缺口|阻塞|缺失|不支持|无法|未提供)/iu.test(e)?"limit":/(?:news|headline|recent|latest|filing|event|公告|新闻|近期|最新|披露|\b20\d{2}[-/年])/iu.test(e)?"timeBound":/(?:[$€£¥%]|\b\d[\d,]*(?:\.\d+)?\b)/u.test(e)?"data":/(?:name|company|symbol|resolved|overview|query|名称|公司|代码|概览|识别)/iu.test(e)?"context":"other"}function parseEvidenceFields(e){return extractKeyValueFields(e.text).filter(e=>e.value.length>0&&!isBoilerplateField(e.label)).filter(e=>!function isLowValueField(e,t){return!!/(?:snapshot for|source type|loaded sources)/iu.test(e)||!!/^\d{1,2}\s+(?:jan|feb|mar|apr|may|jun|jul|aug|sep|oct|nov|dec)\b/iu.test(e)||scoreField({source:"",label:e,value:t,kind:"other"})<0}(e.label,e.value)).map(t=>{const l=classifyFact(`${t.label}: ${t.value}`);return{source:e.source,label:titleCaseLabel(t.label),value:formatEvidence(t.value),kind:"other"===l?e.kind:l}})}function scoreField(e){const t=normalizeFieldLabel(e.label);let l=e.value.length>0?10:0;return/(?:price|value|amount|total|ratio|rate|range|target|estimate|revenue|income|earnings|margin|volume|date|rating|status|count|metric|价格|收入|利润|目标|评级|数量|日期|范围|比率)/iu.test(t)&&(l+=8),/(?:name|company|symbol|title|headline|summary|名称|公司|代码|标题|摘要)/iu.test(t)&&(l+=6),/(?:unix|epoch|timestamp|raw|html|url|href|source|id|identifier|internal|debug)/iu.test(t)&&(l-=20),(/^https?:\/\//iu.test(e.value)||"/"===e.value)&&(l-=12),e.value.length>240&&(l-=5),l}function extractKeyValueFields(e){const t=[...e.matchAll(/(^|[\s;|])([\p{L}\p{N}][\p{L}\p{N} _./()%&+\-]{0,48})[:：](?!\/\/)\s*/gu)].filter(e=>void 0!==e.index).map((e,t,l)=>{l[t-1];const a=function cleanFieldLabel(e){const t=e.replace(/^(?:[-*]\s*)+/u,"").replace(/\s+/gu," ").trim(),l=t.match(/^.+\b(PE Ratio TTM)$/iu);if(l?.[1]&&l[1]!==t&&/(?:beta|\b20\d{2}|\b(?:jan|feb|mar|apr|may|jun|jul|aug|sep|oct|nov|dec)\b)/iu.test(t))return l[1];const a=t.split(" ").filter(Boolean);return a.length>1&&/^\d{4}$/u.test(a[0]??"")||a.length>1&&/^\d+(?:\.\d+)?$/u.test(a[0]??"")&&/^(?:\d+|beta|pe|p\/b)$/iu.test(a[1]??"")||a.length>1&&/^[A-Z]$/u.test(a[0]??"")||a.length>1&&/^(?:inc\.?|llc\.?|ltd\.?|corp\.?|corporation|co\.?)$/iu.test(a[0]??"")||a.length>1&&/^(?:usd|eur|gbp|jpy|cny|cad|aud)$/iu.test(a[0]??"")||a.length>1&&/^(?:gmt|utc)$/iu.test(a[0]??"")?a.slice(1).join(" "):function trailingKnownLabel(e){if(/\bPE Ratio TTM$/iu.test(e)&&!/^PE Ratio TTM$/iu.test(e))return"PE Ratio TTM";const t=e.match(/\b(?:Revenue|Net Income|Operating Income|Diluted EPS|EPS|Market Cap|Book Value|Target Estimate|Target Mean Price|Target Median Price|Target High Price|Target Low Price|PE Ratio TTM|P\/B|Beta(?: 5Y Monthly)?|Published|Updated|Date|Time|Summary|Title|Headline)$/iu);return t&&/(?:filed|\b20\d{2}|10-k|10-q|\bfy\b|\b(?:jan|feb|mar|apr|may|jun|jul|aug|sep|oct|nov|dec)\b|\s{3,}|[a-z].{8,}\s)/iu.test(e.slice(0,t.index))?t?.[0]??e:e}(t)}(e[2]??""),r=function droppedLabelPrefix(e,t){const l=e.trim();if(l===t)return"";const a=l.toLowerCase().lastIndexOf(t.toLowerCase());return a>0?l.slice(0,a):""}(e[2]??"",a);return{index:(e.index??0)+String(e[1]??"").length+r.length,start:(e.index??0)+e[0].length,label:a,rawLabel:e[2]??""}}).filter(e=>!isBoundaryNoiseLabel(e.label)&&!isBoundaryNoiseLabel(e.rawLabel));return t.map((l,a)=>{const r=t[a+1]?.index??e.length;return{label:l.label,value:cleanFieldValue(e.slice(l.start,r),l.label)}}).filter(e=>e.label.length>0)}function cleanFieldValue(e,t){const l=e.replace(/^[\s,;|.-]+/u,"").replace(/[\s,;|.-]+$/u,"").replace(/\s+/gu," ").trim(),a=normalizeFieldLabel(t);if(/\brange\b/iu.test(a))return l.replace(/\s+Beta\b.*$/iu,"").trim();if(/^(?:ticker|symbol|resolved symbol|resolved ticker|股票 resolved ticker|代码)$/iu.test(a)){const e=l.match(/^([A-Z0-9._-]{1,24})(?:\s+\d+[.)]\s+|$)/u);if(e?.[1])return e[1]}return l}function isBoilerplateField(e){return/^(?:status|control status|evidence tool|tool|source|来源)$/iu.test(e.trim())}function isBoundaryNoiseLabel(e){return/^\d{1,2}\s+(?:jan|feb|mar|apr|may|jun|jul|aug|sep|oct|nov|dec)\b/iu.test(e)||/^(?:jan|feb|mar|apr|may|jun|jul|aug|sep|oct|nov|dec)\s+\d{1,4}(?:\s+\d{1,2})?$/iu.test(e)||/^\d{4}\s+\d{1,2}$/u.test(e)}export function normalizeFieldLabel(e){return e.toLowerCase().replace(/[^a-z0-9\p{L}\p{N}]+/giu," ").trim()}function titleCaseLabel(e){if(/\p{Script=Han}/u.test(e))return e;const t=e.split(/[^A-Za-z0-9]+/u).filter(Boolean);return normalizeFieldLabel(e).split(" ").filter(Boolean).map((e,l)=>{const a=t[l]??e;return/^[A-Z0-9]{2,}$/u.test(a)?a:e.charAt(0).toUpperCase()+e.slice(1)}).join(" ")}function formatEvidence(e){const t=e.replace(/\s+/gu," ").trim();return t.length>1200?`${t.slice(0,1197)}...`:t}function escapeTableCell(e){return e.replace(/\|/gu,"\\|").replace(/\r?\n/gu," ")}function fieldKey(e){return`${e.source}\0${normalizeFieldLabel(e.label)}\0${e.value.toLowerCase()}`}

package/packages/core/dist/runtime/events.d.ts

@@ -1,6 +1,7 @@
 import type { MemoryCandidate, MemoryDecision, MemoryRecord } from "@stable-harness/memory";
 import type { ApprovalRequest } from "@stable-harness/governance";
 import type { RuntimeArtifact } from "../types.js";
+import type { RuntimeSandboxDecision } from "./tool-gateway.js";
 import type { RuntimeToolFailureClassification, RuntimeToolFailureReason } from "./tool-failure.js";
 export type RuntimeInventoryRepairLayer = "agent" | "workflow_route" | "workflow" | "tool" | "task" | "skill";
 export type RuntimeInventoryRepairStatus = "repaired" | "blocked";
@@ -71,6 +72,13 @@ export type RuntimeEvent = {
     agentId: string;
     toolId: string;
     output: unknown;
+} | {
+    type: "runtime.sandbox.decision";
+    requestId: string;
+    sessionId: string;
+    agentId: string;
+    toolId: string;
+    decision: RuntimeSandboxDecision;
 } | {
     type: "runtime.approval.requested";
     requestId: string;

package/packages/core/dist/runtime/governance/sandbox.d.ts

@@ -0,0 +1,9 @@
+import type { CompiledWorkspace, RuntimeEvent, RuntimeToolGateway } from "../../types.js";
+import type { RuntimeSandboxPolicy } from "../tool-gateway.js";
+export declare function createSandboxedToolGateway(input: {
+    gateway?: RuntimeToolGateway;
+    workspace: CompiledWorkspace;
+    sandbox?: RuntimeSandboxPolicy | false;
+    emit: (event: RuntimeEvent) => void;
+}): RuntimeToolGateway | undefined;
+export declare function createWorkspaceSandboxPolicy(config: unknown): RuntimeSandboxPolicy | undefined;

package/packages/core/dist/runtime/governance/sandbox.js

@@ -0,0 +1 @@
+export function createSandboxedToolGateway(e){const o=!1===e.sandbox?void 0:e.sandbox??createWorkspaceSandboxPolicy(e.workspace.runtime.sandbox);return e.gateway&&o?{get:o=>e.gateway.get(o),repairToolCall:e.gateway.repairToolCall?o=>e.gateway.repairToolCall(o):void 0,async invoke(t){const n=await o.decide({toolId:t.toolId,args:t.args,context:t.context});return function emitDecision(e,o,t){e({type:"runtime.sandbox.decision",requestId:o.context.requestId,sessionId:o.context.sessionId,agentId:o.context.agentId,toolId:o.toolId,decision:t})}(e.emit,t,n),"deny"===n.action?{toolId:t.toolId,output:sandboxDeniedOutput(t.toolId,n)}:e.gateway.invoke({...t,context:{...t.context,sandbox:n}})}}:e.gateway}export function createWorkspaceSandboxPolicy(e){const o=readRecord(e);if(!o||!1===o.enabled)return;const t=function readRuleMap(e){const o=readRecord(e);return o?new Map(Object.entries(o).map(([e,o])=>[e,readRule(o)])):new Map}(o.byToolId??o.sandboxByToolId??o.profilesByToolId),n=new Set(function readStringArray(e){return Array.isArray(e)?e.filter(e=>"string"==typeof e&&e.trim().length>0):[]}(o.denyToolIds)),a=function fallbackDecision(e){const o=readAction(e.defaultAction)??"allow",t=readString(e.defaultProfile)??readString(e.profile);return{action:o,...t?{profile:t}:{},reason:"allow"===o?"No sandbox rule matched; default allow applied.":"No sandbox rule matched; default deny applied."}}(o);return{decide:e=>n.has(e.toolId)?{action:"deny",reason:"Tool is denied by sandbox policy."}:t.get(e.toolId)??a}}function sandboxDeniedOutput(e,o){return{status:"sandbox_denied",toolId:e,reason:o.reason??"Tool execution was denied by sandbox policy.",...o.profile?{profile:o.profile}:{},...o.policyId?{policyId:o.policyId}:{}}}function readRule(e){if("string"==typeof e&&e.trim())return{action:"allow",profile:e.trim()};const o=readRecord(e);if(!o)return{action:"allow"};const t=readAction(o.action)??(!0===o.deny?"deny":"allow"),n=readString(o.profile),a=readString(o.reason),r=readString(o.policyId),d=readRecord(o.metadata);return{action:t,...n?{profile:n}:{},...a?{reason:a}:{},...r?{policyId:r}:{},...d?{metadata:d}:{}}}function readAction(e){return"allow"===e||"deny"===e?e:void 0}function readString(e){return"string"==typeof e&&e.trim()?e.trim():void 0}function readRecord(e){return"object"!=typeof e||null===e||Array.isArray(e)?void 0:e}

package/packages/core/dist/runtime/tool-gateway.d.ts

@@ -10,6 +10,22 @@ export type RuntimeToolGatewayContext = {
     requestInput?: string;
     observedEvidence?: string;
     approvalIds?: string[];
+    sandbox?: RuntimeSandboxDecision;
+};
+export type RuntimeSandboxDecision = {
+    action: "allow" | "deny";
+    profile?: string;
+    reason?: string;
+    policyId?: string;
+    metadata?: Record<string, unknown>;
+};
+export type RuntimeSandboxPolicyInput = {
+    toolId: string;
+    args?: unknown;
+    context: RuntimeToolGatewayContext;
+};
+export type RuntimeSandboxPolicy = {
+    decide(input: RuntimeSandboxPolicyInput): Promise<RuntimeSandboxDecision> | RuntimeSandboxDecision;
 };
 export type RuntimeToolGatewayTool = {
     id: string;

package/packages/core/dist/runtime.d.ts

@@ -1,7 +1,7 @@
 import type { ApprovalQueue } from "@stable-harness/governance";
 import type { MemoryProvider, RuntimeMemoryStore } from "@stable-harness/memory";
 import type { QualityReviewModel } from "./quality/index.js";
-import type { CompiledWorkspace, RuntimeCapabilityModule, RuntimeToolGateway, RuntimeAdapter, RuntimeArtifactStore, RuntimeStore, RuntimeProgressNarrationOptions, RuntimeWorkflowAdapter, StableHarnessRuntime } from "./types.js";
+import type { CompiledWorkspace, RuntimeCapabilityModule, RuntimeToolGateway, RuntimeAdapter, RuntimeArtifactStore, RuntimeSandboxPolicy, RuntimeStore, RuntimeProgressNarrationOptions, RuntimeWorkflowAdapter, StableHarnessRuntime } from "./types.js";
 type RuntimeFactoryInput = {
     workspace: CompiledWorkspace;
     adapters: RuntimeAdapter[];
@@ -10,6 +10,7 @@ type RuntimeFactoryInput = {
     memoryProviders?: MemoryProvider[];
     approvals?: ApprovalQueue;
     toolGateway?: RuntimeToolGateway;
+    sandbox?: RuntimeSandboxPolicy | false;
     store?: RuntimeStore;
     artifacts?: RuntimeArtifactStore;
     progressNarration?: RuntimeProgressNarrationOptions | false;

package/packages/core/dist/runtime.js

@@ -1 +1 @@
-import{randomUUID as e}from"node:crypto";import{assertExecutionContract as t}from"./execution-contract.js";import{assertNoRawToolCallOutput as r,assertNoToolExecutionErrorOutput as a,buildAdapterErrorRecoveryPrompt as s,buildEvidenceSynthesisOutput as o,buildExecutionContractRecoveryRequest as n,buildResultRecoveryRequest as i,containsRawToolCallOutput as u,isRecoverableAdapterError as c,rawToolCallFailureMessage as l,toolCallRecoveryEnabled as p}from"./recovery/tool-call.js";import{recoverQualityReview as d,resolveQualityPolicy as m}from"./quality/index.js";import{completeRun as y,failRun as w}from"./runtime/completion.js";import{runDirectToolCall as f}from"./runtime/direct-tool-call.js";import{createApprovalGatedToolGateway as I}from"./runtime/governance/approval-gate.js";import{createRuntimeInspectionMethods as g}from"./runtime/inspection/methods.js";import{createRuntimeCapabilityRegistry as q,normalizeAdapterResult as R}from"./runtime/capabilities.js";import{createMemoryRuntimeCapability as v}from"./runtime/memory.js";import{createInMemoryRuntimeStore as k}from"./runtime/persistence/stores.js";import{createProgressNarrationCapability as b}from"./runtime/progress-narration.js";import{repairRuntimeSelection as A}from"./runtime/selection-repair.js";import{createToolFailureTracker as C}from"./runtime/tool-failure.js";import{runWorkflowRequest as x}from"./workflows/runtime.js";export function createStableHarnessRuntime(t){const r=new Set,a=t.store??k(),o=q([v(t),b({options:t.progressNarration,policy:t.workspace.runtime}),...t.capabilities??[]]),emitBase=e=>{a.appendEvent(e);for(const t of r)t(e)},emit=e=>{emitBase(e),o.emitSideEffects(e,emitBase)},i={...t,toolGateway:I({gateway:t.toolGateway,approvals:t.approvals,workspace:t.workspace,emit:emit})},u=C(function readToolFailurePolicy(e){if("object"!=typeof e||null===e||Array.isArray(e))return;const t=e.failurePolicy;return"object"!=typeof t||null===t||Array.isArray(t)?void 0:t}(t.workspace.runtime.toolGateway));return{request:async t=>async function runRuntimeRequest(t){const r=t.request.requestId??e(),a=t.request.sessionId??e(),o=[],{agent:i,adapter:u}=await async function resolveExecution(e,t,r){const a=t.agentId?await async function resolveRequestedAgentId(e,t,r){if(e.agents.has(t))return t;const a=await A({id:t,candidates:[...e.agents.values()].map(e=>({id:e.id,description:e.description})),trace:{...r,agentId:t,layer:"agent",owner:"stable_runtime_policy"}});return a.ok?a.id:t}(e.workspace,t.agentId,r):function resolveRoutedAgentId(e,t){for(const r of e.runtime.routes??[])if(routeMatches(r,t))return r.agentId;return e.runtime.defaultAgentId}(e.workspace,t.input),s=e.workspace.agents.get(a);if(!s)throw new Error(`Agent ${a} is not defined in the workspace`);if(t.toolCall||t.workflow)return{agent:s,adapter:void 0};const o=e.adapters.find(e=>e.canRun(s));if(!o)throw new Error(`No runtime adapter can run backend ${s.backend} for agent ${s.id}`);return{agent:s,adapter:o}}(t.input,t.request,{requestId:r,sessionId:a,emit:e=>o.push(e)});t.store.createRun(function createRunRecord(e,t,r,a){return{requestId:t,sessionId:r,agentId:a.id,input:e.input,state:"running",parentRunId:e.parentRunId,metadata:e.metadata,artifacts:[],startedAt:(new Date).toISOString(),events:[]}}(t.request,r,a,i)),o.forEach(t.emit),t.emit({type:"runtime.request.started",requestId:r,sessionId:a,agentId:i.id,input:t.request.input});try{if(t.request.workflow){const e=await x({workspace:t.input.workspace,adapters:t.input.workflowAdapters??[],toolGateway:t.input.toolGateway,request:{input:t.request.input,...t.request.workflow},requestId:r,sessionId:a,agentId:i.id,emit:t.emit});return y({store:t.store,emit:t.emit,requestId:r,sessionId:a,agent:i,result:e,artifacts:t.input.artifacts})}if(t.request.toolCall){const e=await f({gateway:t.input.toolGateway,workspace:t.input.workspace,emit:t.emit,request:t.request,requestId:r,sessionId:a,agent:i,toolFailureTracker:t.toolFailureTracker});return y({store:t.store,emit:t.emit,requestId:r,sessionId:a,agent:i,result:e,artifacts:t.input.artifacts})}return await async function runAdapterRequest(e){if(!e.adapter)throw new Error(`No runtime adapter can run backend ${e.agent.backend} for agent ${e.agent.id}`);const t=e.adapter,r=await e.capabilities.beforeAdapterRun(createCapabilityContext(e)),a=r.memory,o=r.pluginMemories??[],i=e.input.workspace.runtime,u=m(e.input.workspace.runtime,e.agent),l=new Map;let p;try{p=await runAdapterOnce(e,t,e.request,a,o,l)}catch(r){if(!c(r,i))throw r;p=await runAdapterOnce(e,t,s(e.request,r,i),a,o,l)}p=await recoverAdapterResultOutput(e,t,e.request,p,a,o,i,l),p=await d(createQualityRuntimeInput(e,a,o,l),e.request,p,u),await e.capabilities.beforeAdapterResultContract({...createCapabilityContext(e),result:p});try{assertRequestExecutionContract(e)}catch(r){const s=n({request:e.request,events:e.store.getRun(e.requestId)?.events??[],policy:i});if(!s)throw r;p=await runAdapterOnce(e,t,s,a,o,l),p=await recoverAdapterResultOutput(e,t,s,p,a,o,i,l),p=await d(createQualityRuntimeInput(e,a,o,l),s,p,u),assertRequestExecutionContract(e)}const w=y({store:e.store,emit:e.emit,requestId:e.requestId,sessionId:e.sessionId,agent:e.agent,result:p,artifacts:e.input.artifacts});return await e.capabilities.afterAdapterResponse({...createCapabilityContext(e),result:p,response:w}),w}({...t,adapter:u,requestId:r,sessionId:a,agent:i})}catch(e){return w({store:t.store,emit:t.emit,requestId:r,sessionId:a,agent:i,error:e})}}({input:i,capabilities:o,store:a,emit:emit,request:t,toolFailureTracker:u}),subscribe:e=>(r.add(e),()=>r.delete(e)),...g({workspace:t.workspace,store:a,artifacts:t.artifacts,approvals:t.approvals,emit:emit}),cancel(e,t){const r=a.getRun(e);r&&"running"===r.state&&(a.updateRun(e,{state:"cancelled",completedAt:(new Date).toISOString()}),emit({type:"runtime.request.cancelled",requestId:e,sessionId:r.sessionId,agentId:r.agentId,reason:t}))},async stop(){await o.stop(),r.clear()}}}function createCapabilityContext(e){return{workspace:e.input.workspace,store:e.store,emit:e.emit,request:e.request,requestId:e.requestId,sessionId:e.sessionId,agent:e.agent}}function createQualityRuntimeInput(e,t,r,a){return{workspace:e.input.workspace,agent:e.agent,request:e.request,requestId:e.requestId,sessionId:e.sessionId,events:e.store.getRun(e.requestId)?.events??[],emit:e.emit,getEvents:()=>e.store.getRun(e.requestId)?.events??[],runAdapter:s=>runAdapterOnce(e,e.adapter,s,t,r,a),reviewModel:e.input.qualityReviewModel,memory:t,pluginMemories:r}}async function recoverAdapterResultOutput(e,t,s,n,c,d,m,y){let w=s;const f=function resultRecoveryAttempts(e){const t="object"!=typeof e||null===e||Array.isArray(e)?void 0:e.recovery,r="object"!=typeof t||null===t||Array.isArray(t)?void 0:t.toolCall,a="object"!=typeof r||null===r||Array.isArray(r)?void 0:r.maxResultRecoveryAttempts;return"number"==typeof a&&Number.isInteger(a)&&a>0?a:3}(m);let I=0;for(let r=0;r<f;r+=1){const r=e.store.getRun(e.requestId)?.events??[],a=i({request:w,output:n.text,events:r.slice(I),availableToolIds:e.agent.tools,policy:m});if(!a)break;w=a,I=e.store.getRun(e.requestId)?.events.length??0,n=await runAdapterOnce(e,t,a,c,d,y)}if(p(m)){let t=!1;u(n.text,m)&&function rawToolCallFailureReturnsMessage(e){return"message"===("object"!=typeof e?.toolCallRecovery||null===e.toolCallRecovery||Array.isArray(e.toolCallRecovery)?{}:e.toolCallRecovery).onFailure}(s.metadata)&&(n={...n,text:l(),metadata:{...n.metadata,toolCallRecovery:{failed:!0,reason:"raw_tool_call_output"}}});const i=o({request:s,output:n.text,events:e.store.getRun(e.requestId)?.events??[],policy:m});i&&(t=!0,n={...n,text:i,metadata:{...n.metadata,toolCallRecovery:{synthesized:!0,reason:"raw_tool_call_output_with_evidence"}}}),t||(r(n.text,m),a(n.text,m))}return n}function assertRequestExecutionContract(e){t({store:e.store,emit:e.emit,requestId:e.requestId,sessionId:e.sessionId,agent:e.agent,metadata:e.request.metadata})}async function runAdapterOnce(e,t,r,a,s,o){return R(await t.run({workspace:e.input.workspace,agent:e.agent,request:r,requestId:e.requestId,sessionId:e.sessionId,memory:a,pluginMemories:s,toolGateway:e.input.toolGateway,toolFailureTracker:e.input.toolFailureTracker,requestState:o,emit:e.emit}))}function routeMatches(e,t){if(e.pattern)try{if(new RegExp(e.pattern,"iu").test(t))return!0}catch{return!1}const r=t.toLowerCase();return(e.keywords??[]).some(e=>r.includes(e.toLowerCase()))}
+import{randomUUID as e}from"node:crypto";import{assertExecutionContract as t}from"./execution-contract.js";import{assertNoRawToolCallOutput as r,assertNoToolExecutionErrorOutput as a,buildAdapterErrorRecoveryPrompt as o,buildEvidenceSynthesisOutput as s,buildExecutionContractRecoveryRequest as n,buildResultRecoveryRequest as i,containsRawToolCallOutput as u,isRecoverableAdapterError as c,rawToolCallFailureMessage as l,toolCallRecoveryEnabled as p}from"./recovery/tool-call.js";import{recoverQualityReview as d,resolveQualityPolicy as m}from"./quality/index.js";import{completeRun as w,failRun as y}from"./runtime/completion.js";import{runDirectToolCall as f}from"./runtime/direct-tool-call.js";import{createApprovalGatedToolGateway as g}from"./runtime/governance/approval-gate.js";import{createSandboxedToolGateway as I}from"./runtime/governance/sandbox.js";import{createRuntimeInspectionMethods as q}from"./runtime/inspection/methods.js";import{createRuntimeCapabilityRegistry as R,normalizeAdapterResult as v}from"./runtime/capabilities.js";import{createMemoryRuntimeCapability as b}from"./runtime/memory.js";import{createInMemoryRuntimeStore as k}from"./runtime/persistence/stores.js";import{createProgressNarrationCapability as A}from"./runtime/progress-narration.js";import{repairRuntimeSelection as C}from"./runtime/selection-repair.js";import{createToolFailureTracker as x}from"./runtime/tool-failure.js";import{runWorkflowRequest as h}from"./workflows/runtime.js";export function createStableHarnessRuntime(t){const r=new Set,a=t.store??k(),s=R([b(t),A({options:t.progressNarration,policy:t.workspace.runtime}),...t.capabilities??[]]),emitBase=e=>{a.appendEvent(e);for(const t of r)t(e)},emit=e=>{emitBase(e),s.emitSideEffects(e,emitBase)},i=I({gateway:g({gateway:t.toolGateway,approvals:t.approvals,workspace:t.workspace,emit:emit}),workspace:t.workspace,sandbox:t.sandbox,emit:emit}),u={...t,toolGateway:i},l=x(function readToolFailurePolicy(e){if("object"!=typeof e||null===e||Array.isArray(e))return;const t=e.failurePolicy;return"object"!=typeof t||null===t||Array.isArray(t)?void 0:t}(t.workspace.runtime.toolGateway));return{request:async t=>async function runRuntimeRequest(t){const r=t.request.requestId??e(),a=t.request.sessionId??e(),s=[],{agent:i,adapter:u}=await async function resolveExecution(e,t,r){const a=t.agentId?await async function resolveRequestedAgentId(e,t,r){if(e.agents.has(t))return t;const a=await C({id:t,candidates:[...e.agents.values()].map(e=>({id:e.id,description:e.description})),trace:{...r,agentId:t,layer:"agent",owner:"stable_runtime_policy"}});return a.ok?a.id:t}(e.workspace,t.agentId,r):function resolveRoutedAgentId(e,t){for(const r of e.runtime.routes??[])if(routeMatches(r,t))return r.agentId;return e.runtime.defaultAgentId}(e.workspace,t.input),o=e.workspace.agents.get(a);if(!o)throw new Error(`Agent ${a} is not defined in the workspace`);if(t.toolCall||t.workflow)return{agent:o,adapter:void 0};const s=e.adapters.find(e=>e.canRun(o));if(!s)throw new Error(`No runtime adapter can run backend ${o.backend} for agent ${o.id}`);return{agent:o,adapter:s}}(t.input,t.request,{requestId:r,sessionId:a,emit:e=>s.push(e)});t.store.createRun(function createRunRecord(e,t,r,a){return{requestId:t,sessionId:r,agentId:a.id,input:e.input,state:"running",parentRunId:e.parentRunId,metadata:e.metadata,artifacts:[],startedAt:(new Date).toISOString(),events:[]}}(t.request,r,a,i)),s.forEach(t.emit),t.emit({type:"runtime.request.started",requestId:r,sessionId:a,agentId:i.id,input:t.request.input});try{if(t.request.workflow){const e=await h({workspace:t.input.workspace,adapters:t.input.workflowAdapters??[],toolGateway:t.input.toolGateway,request:{input:t.request.input,...t.request.workflow},requestId:r,sessionId:a,agentId:i.id,emit:t.emit});return w({store:t.store,emit:t.emit,requestId:r,sessionId:a,agent:i,result:e,artifacts:t.input.artifacts})}if(t.request.toolCall){const e=await f({gateway:t.input.toolGateway,workspace:t.input.workspace,emit:t.emit,request:t.request,requestId:r,sessionId:a,agent:i,toolFailureTracker:t.toolFailureTracker});return w({store:t.store,emit:t.emit,requestId:r,sessionId:a,agent:i,result:e,artifacts:t.input.artifacts})}return await async function runAdapterRequest(e){if(!e.adapter)throw new Error(`No runtime adapter can run backend ${e.agent.backend} for agent ${e.agent.id}`);const t=e.adapter,r=await e.capabilities.beforeAdapterRun(createCapabilityContext(e)),a=r.memory,s=r.pluginMemories??[],i=e.input.workspace.runtime,u=m(e.input.workspace.runtime,e.agent),l=new Map;let p;try{p=await runAdapterOnce(e,t,e.request,a,s,l)}catch(r){if(!c(r,i))throw r;p=await runAdapterOnce(e,t,o(e.request,r,i),a,s,l)}p=await recoverAdapterResultOutput(e,t,e.request,p,a,s,i,l),p=await d(createQualityRuntimeInput(e,a,s,l),e.request,p,u),await e.capabilities.beforeAdapterResultContract({...createCapabilityContext(e),result:p});try{assertRequestExecutionContract(e)}catch(r){const o=n({request:e.request,events:e.store.getRun(e.requestId)?.events??[],policy:i});if(!o)throw r;p=await runAdapterOnce(e,t,o,a,s,l),p=await recoverAdapterResultOutput(e,t,o,p,a,s,i,l),p=await d(createQualityRuntimeInput(e,a,s,l),o,p,u),assertRequestExecutionContract(e)}const y=w({store:e.store,emit:e.emit,requestId:e.requestId,sessionId:e.sessionId,agent:e.agent,result:p,artifacts:e.input.artifacts});return await e.capabilities.afterAdapterResponse({...createCapabilityContext(e),result:p,response:y}),y}({...t,adapter:u,requestId:r,sessionId:a,agent:i})}catch(e){return y({store:t.store,emit:t.emit,requestId:r,sessionId:a,agent:i,error:e})}}({input:u,capabilities:s,store:a,emit:emit,request:t,toolFailureTracker:l}),subscribe:e=>(r.add(e),()=>r.delete(e)),...q({workspace:t.workspace,store:a,artifacts:t.artifacts,approvals:t.approvals,emit:emit}),cancel(e,t){const r=a.getRun(e);r&&"running"===r.state&&(a.updateRun(e,{state:"cancelled",completedAt:(new Date).toISOString()}),emit({type:"runtime.request.cancelled",requestId:e,sessionId:r.sessionId,agentId:r.agentId,reason:t}))},async stop(){await s.stop(),r.clear()}}}function createCapabilityContext(e){return{workspace:e.input.workspace,store:e.store,emit:e.emit,request:e.request,requestId:e.requestId,sessionId:e.sessionId,agent:e.agent}}function createQualityRuntimeInput(e,t,r,a){return{workspace:e.input.workspace,agent:e.agent,request:e.request,requestId:e.requestId,sessionId:e.sessionId,events:e.store.getRun(e.requestId)?.events??[],emit:e.emit,getEvents:()=>e.store.getRun(e.requestId)?.events??[],runAdapter:o=>runAdapterOnce(e,e.adapter,o,t,r,a),reviewModel:e.input.qualityReviewModel,memory:t,pluginMemories:r}}async function recoverAdapterResultOutput(e,t,o,n,c,d,m,w){let y=o;const f=function resultRecoveryAttempts(e){const t="object"!=typeof e||null===e||Array.isArray(e)?void 0:e.recovery,r="object"!=typeof t||null===t||Array.isArray(t)?void 0:t.toolCall,a="object"!=typeof r||null===r||Array.isArray(r)?void 0:r.maxResultRecoveryAttempts;return"number"==typeof a&&Number.isInteger(a)&&a>0?a:3}(m);let g=0;for(let r=0;r<f;r+=1){const r=e.store.getRun(e.requestId)?.events??[],a=i({request:y,output:n.text,events:r.slice(g),availableToolIds:e.agent.tools,policy:m});if(!a)break;y=a,g=e.store.getRun(e.requestId)?.events.length??0,n=await runAdapterOnce(e,t,a,c,d,w)}if(p(m)){let t=!1;u(n.text,m)&&function rawToolCallFailureReturnsMessage(e){return"message"===("object"!=typeof e?.toolCallRecovery||null===e.toolCallRecovery||Array.isArray(e.toolCallRecovery)?{}:e.toolCallRecovery).onFailure}(o.metadata)&&(n={...n,text:l(),metadata:{...n.metadata,toolCallRecovery:{failed:!0,reason:"raw_tool_call_output"}}});const i=s({request:o,output:n.text,events:e.store.getRun(e.requestId)?.events??[],policy:m});i&&(t=!0,n={...n,text:i,metadata:{...n.metadata,toolCallRecovery:{synthesized:!0,reason:"raw_tool_call_output_with_evidence"}}}),t||(r(n.text,m),a(n.text,m))}return n}function assertRequestExecutionContract(e){t({store:e.store,emit:e.emit,requestId:e.requestId,sessionId:e.sessionId,agent:e.agent,metadata:e.request.metadata})}async function runAdapterOnce(e,t,r,a,o,s){return v(await t.run({workspace:e.input.workspace,agent:e.agent,request:r,requestId:e.requestId,sessionId:e.sessionId,memory:a,pluginMemories:o,toolGateway:e.input.toolGateway,toolFailureTracker:e.input.toolFailureTracker,requestState:s,emit:e.emit}))}function routeMatches(e,t){if(e.pattern)try{if(new RegExp(e.pattern,"iu").test(t))return!0}catch{return!1}const r=t.toLowerCase();return(e.keywords??[]).some(e=>r.includes(e.toLowerCase()))}

package/packages/core/dist/trace.js

@@ -1 +1 @@
-export function projectRuntimeTrace(e){return e.events.map(projectEvent).filter(isTraceEntry)}export function projectEvent(e){return"runtime.request.started"===e.type||"runtime.request.completed"===e.type||"runtime.request.failed"===e.type?base(e,"request",e.type):"runtime.request.cancelled"===e.type?base(e,"request",e.type,{reason:e.reason}):"runtime.execution.contract.failed"===e.type?base(e,"request",e.type,{reason:e.reason,missingEvidenceTools:e.missingEvidenceTools}):"runtime.inventory.repair"===e.type?base(e,"adapter","runtime.inventory.repair",{status:e.status,...e.diagnostic}):"runtime.tool.direct.started"===e.type?base(e,"tool","runtime.tool.direct.started",{toolId:e.toolId}):"runtime.tool.direct.completed"===e.type?base(e,"tool","runtime.tool.direct.completed",{toolId:e.toolId}):"runtime.tool.failure"===e.type?base(e,"tool","runtime.tool.failure",{toolId:e.toolId,...e.failure}):"runtime.tool.circuit.opened"===e.type?base(e,"tool","runtime.tool.circuit.opened",{toolId:e.toolId,reason:e.reason}):"runtime.workflow.started"===e.type||"runtime.workflow.completed"===e.type?base(e,"workflow",e.type,{workflowId:e.workflowId,adapter:e.adapter}):function isSpecDrivenPhaseEvent(e){return e.type.startsWith("runtime.specDriven.phase.")}(e)?base(e,"spec",e.type,{phaseId:e.phaseId,..."workflowId"in e&&e.workflowId?{workflowId:e.workflowId}:{},..."reason"in e?{reason:e.reason}:{},..."artifact"in e&&e.artifact?{artifact:e.artifact}:{}}):"runtime.adapter.event"===e.type?function adapterTrace(e){const t=e.event;if(isRecord(t)&&"string"==typeof t.phase){const r=function semanticAdapterTrace(e,t){if("inventory.repair"===t.phase)return base(e,"adapter","runtime.inventory.repair",{status:t.status,...isRecord(t.diagnostic)?t.diagnostic:{}});const r=function readTraceType(e){const t=readString(e);return"request"===t||"tool"===t||"workflow"===t||"spec"===t||"adapter"===t||"memory"===t||"artifact"===t||"progress"===t||"plan"===t||"delegation"===t?t:void 0}(t.traceType),o=readString(t.traceLabel);return r&&o?base(e,r,o,t):void 0}(e,t);return r||base(e,"adapter",t.phase.startsWith("agent.")?t.phase:`adapter.${t.phase}`,t)}return base(e,"adapter","runtime.adapter.event",{event:t})}(e):"runtime.artifact.created"===e.type?base(e,"artifact","runtime.artifact.created",{artifact:e.artifact}):e.type.startsWith("runtime.memory.")?base(e,"memory",e.type):"runtime.progress.narration"===e.type?base(e,"progress",e.type,{message:e.message,provider:e.provider,sourceEventTypes:e.sourceEventTypes}):void 0}export function readPlanTodos(e){const t=function readPlanRecord(e){if(isRecord(e))return e;if("string"==typeof e)try{const t=JSON.parse(e);return isRecord(t)?t:void 0}catch{return}}(e),r=function readTodosArray(e){const t=isRecord(e?.args)?e.args:void 0;return Array.isArray(e?.todos)?e.todos:Array.isArray(t?.todos)?t.todos:[]}(t);return r.map(readTodo).filter(isPlanTodoItem)}function readTodo(e){if(isRecord(e)&&"string"==typeof e.content)return{content:e.content,status:"string"==typeof e.status?e.status:"pending"}}function isPlanTodoItem(e){return void 0!==e}function base(e,t,r,o){return{type:t,label:r,agentId:e.agentId,requestId:e.requestId,detail:o}}function isTraceEntry(e){return void 0!==e}function isRecord(e){return"object"==typeof e&&null!==e&&!Array.isArray(e)}function readString(e){return"string"==typeof e&&e.trim()?e:void 0}
+export function projectRuntimeTrace(e){return e.events.map(projectEvent).filter(isTraceEntry)}export function projectEvent(e){return"runtime.request.started"===e.type||"runtime.request.completed"===e.type||"runtime.request.failed"===e.type?base(e,"request",e.type):"runtime.request.cancelled"===e.type?base(e,"request",e.type,{reason:e.reason}):"runtime.execution.contract.failed"===e.type?base(e,"request",e.type,{reason:e.reason,missingEvidenceTools:e.missingEvidenceTools}):"runtime.inventory.repair"===e.type?base(e,"adapter","runtime.inventory.repair",{status:e.status,...e.diagnostic}):"runtime.tool.direct.started"===e.type?base(e,"tool","runtime.tool.direct.started",{toolId:e.toolId}):"runtime.tool.direct.completed"===e.type?base(e,"tool","runtime.tool.direct.completed",{toolId:e.toolId}):"runtime.sandbox.decision"===e.type?base(e,"tool","runtime.sandbox.decision",{toolId:e.toolId,...e.decision}):"runtime.tool.failure"===e.type?base(e,"tool","runtime.tool.failure",{toolId:e.toolId,...e.failure}):"runtime.tool.circuit.opened"===e.type?base(e,"tool","runtime.tool.circuit.opened",{toolId:e.toolId,reason:e.reason}):"runtime.workflow.started"===e.type||"runtime.workflow.completed"===e.type?base(e,"workflow",e.type,{workflowId:e.workflowId,adapter:e.adapter}):function isSpecDrivenPhaseEvent(e){return e.type.startsWith("runtime.specDriven.phase.")}(e)?base(e,"spec",e.type,{phaseId:e.phaseId,..."workflowId"in e&&e.workflowId?{workflowId:e.workflowId}:{},..."reason"in e?{reason:e.reason}:{},..."artifact"in e&&e.artifact?{artifact:e.artifact}:{}}):"runtime.adapter.event"===e.type?function adapterTrace(e){const t=e.event;if(isRecord(t)&&"string"==typeof t.phase){const r=function semanticAdapterTrace(e,t){if("inventory.repair"===t.phase)return base(e,"adapter","runtime.inventory.repair",{status:t.status,...isRecord(t.diagnostic)?t.diagnostic:{}});const r=function readTraceType(e){const t=readString(e);return"request"===t||"tool"===t||"workflow"===t||"spec"===t||"adapter"===t||"memory"===t||"artifact"===t||"progress"===t||"plan"===t||"delegation"===t?t:void 0}(t.traceType),o=readString(t.traceLabel);return r&&o?base(e,r,o,t):void 0}(e,t);return r||base(e,"adapter",t.phase.startsWith("agent.")?t.phase:`adapter.${t.phase}`,t)}return base(e,"adapter","runtime.adapter.event",{event:t})}(e):"runtime.artifact.created"===e.type?base(e,"artifact","runtime.artifact.created",{artifact:e.artifact}):e.type.startsWith("runtime.memory.")?base(e,"memory",e.type):"runtime.progress.narration"===e.type?base(e,"progress",e.type,{message:e.message,provider:e.provider,sourceEventTypes:e.sourceEventTypes}):void 0}export function readPlanTodos(e){const t=function readPlanRecord(e){if(isRecord(e))return e;if("string"==typeof e)try{const t=JSON.parse(e);return isRecord(t)?t:void 0}catch{return}}(e),r=function readTodosArray(e){const t=isRecord(e?.args)?e.args:void 0;return Array.isArray(e?.todos)?e.todos:Array.isArray(t?.todos)?t.todos:[]}(t);return r.map(readTodo).filter(isPlanTodoItem)}function readTodo(e){if(isRecord(e)&&"string"==typeof e.content)return{content:e.content,status:"string"==typeof e.status?e.status:"pending"}}function isPlanTodoItem(e){return void 0!==e}function base(e,t,r,o){return{type:t,label:r,agentId:e.agentId,requestId:e.requestId,detail:o}}function isTraceEntry(e){return void 0!==e}function isRecord(e){return"object"==typeof e&&null!==e&&!Array.isArray(e)}function readString(e){return"string"==typeof e&&e.trim()?e:void 0}

package/packages/core/dist/types.d.ts

@@ -16,7 +16,7 @@ export type { CompiledWorkspace, WorkspaceAdapterPolicy, WorkspaceAgent, Workspa
 export type { WorkspaceToolQualityDiagnostic, WorkspaceToolQualityDiagnosticCode, WorkspaceToolQualityPolicy, } from "./workspace/tool-quality.js";
 export type { WorkspaceEvaluation, WorkspaceEvaluationCase, } from "./evaluations/index.js";
 export type { SpecDrivenPhaseId, SpecDrivenPhaseRecord, SpecDrivenPhaseStatus, SpecDrivenPhaseTransition, SpecDrivenWorkflowState, WorkspaceSpecDrivenPhase, WorkspaceSpecDrivenWorkflowPolicy, } from "./spec-driven/index.js";
-export type { RuntimeToolCallRequest, RuntimeToolGateway, RuntimeToolGatewayContext, RuntimeToolRepairModel, RuntimeToolGatewayTool, } from "./runtime/tool-gateway.js";
+export type { RuntimeToolCallRequest, RuntimeToolGateway, RuntimeToolGatewayContext, RuntimeToolRepairModel, RuntimeToolGatewayTool, RuntimeSandboxDecision, RuntimeSandboxPolicy, RuntimeSandboxPolicyInput, } from "./runtime/tool-gateway.js";
 export type { RuntimeToolFailureClassification, RuntimeToolFailureKind, RuntimeToolFailurePolicy, RuntimeToolFailureReason, RuntimeToolFailureTracker, } from "./runtime/tool-failure.js";
 export type RuntimeAdapterContext = {
     workspace: CompiledWorkspace;

package/packages/core/dist/workspace/types.d.ts

@@ -63,6 +63,7 @@ export type WorkspaceRuntimePolicy = {
     recovery?: Record<string, unknown>;
     retry?: WorkspaceRetryPolicy;
     toolGateway?: Record<string, unknown>;
+    sandbox?: Record<string, unknown>;
     memory?: Record<string, unknown>;
     protocols?: Record<string, unknown>;
     progress?: Record<string, unknown>;

package/packages/tool-gateway/dist/src/types.d.ts

@@ -6,6 +6,14 @@ export type ToolGatewayContext = {
     requestInput?: string;
     observedEvidence?: string;
     approvalIds?: string[];
+    sandbox?: ToolGatewaySandboxDecision;
+};
+export type ToolGatewaySandboxDecision = {
+    action: "allow" | "deny";
+    profile?: string;
+    reason?: string;
+    policyId?: string;
+    metadata?: Record<string, unknown>;
 };
 export type ToolGatewayTool = {
     id: string;