@caupulican/pi-adaptative 0.81.26 → 0.81.29

package/dist/core/tool-gate-controller.js

@@ -38,23 +38,27 @@ export class ToolGateController {
             };
         }
         const runner = this.deps.getExtensionRunner();
-        if (!runner.hasHandlers("tool_call")) {
-            return undefined;
-        }
-        try {
-            return await runner.emitToolCall({
-                type: "tool_call",
-                toolName: toolCall.name,
-                toolCallId: toolCall.id,
-                input: args,
-            });
-        }
-        catch (err) {
-            if (err instanceof Error) {
-                throw err;
+        let extensionResult;
+        if (runner.hasHandlers("tool_call")) {
+            try {
+                extensionResult = await runner.emitToolCall({
+                    type: "tool_call",
+                    toolName: toolCall.name,
+                    toolCallId: toolCall.id,
+                    input: args,
+                });
+            }
+            catch (err) {
+                if (err instanceof Error) {
+                    throw err;
+                }
+                throw new Error(`Extension failed, blocking execution: ${String(err)}`);
             }
-            throw new Error(`Extension failed, blocking execution: ${String(err)}`);
         }
+        if (extensionResult)
+            return extensionResult;
+        this.deps.getToolSelectionController?.()?.begin(toolCall.id, toolCall.name, args);
+        return undefined;
     };
     afterToolCall = async ({ toolCall, args, result, isError }) => {
         const runner = this.deps.getExtensionRunner();
@@ -85,6 +89,7 @@ export class ToolGateController {
             const wrapped = content.map((block) => block.type === "text" ? { ...block, text: wrapUntrustedText(block.text, source) } : block);
             content = wrapped;
         }
+        this.deps.getToolSelectionController?.()?.complete(toolCall.id, !resolvedIsError, content);
         if (content === result.content && details === result.details && resolvedIsError === isError) {
             return undefined;
         }

package/dist/core/tool-gate-controller.js.map

@@ -1 +1 @@
-{"version":3,"file":"tool-gate-controller.js","sourceRoot":"","sources":["../../src/core/tool-gate-controller.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AAIH,OAAO,EAAE,gBAAgB,EAAE,MAAM,qBAAqB,CAAC;AAEvD,OAAO,EAAE,iBAAiB,EAAE,iBAAiB,EAAE,MAAM,kCAAkC,CAAC;AAgBxF,MAAM,OAAO,kBAAkB;IACb,IAAI,CAAyB;IAE9C,YAAY,IAA4B,EAAE;QACzC,IAAI,CAAC,IAAI,GAAG,IAAI,CAAC;IAAA,CACjB;IAEQ,cAAc,GAAmB,KAAK,EAAE,EAAE,QAAQ,EAAE,IAAI,EAAE,EAAE,EAAE,CAAC;QACvE,MAAM,UAAU,GAAG,IAAI,CAAC,IAAI,CAAC,qBAAqB,CAAC,QAAQ,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC;QACxE,IAAI,UAAU,EAAE,CAAC;YAChB,OAAO,UAAU,CAAC;QACnB,CAAC;QAED,uBAAuB;QACvB,MAAM,QAAQ,GAAG,IAAI,CAAC,IAAI,CAAC,qBAAqB,EAAE,CAAC;QACnD,MAAM,UAAU,GAAG,gBAAgB,CAAC;YACnC,QAAQ,EAAE,QAAQ,CAAC,IAAI;YACvB,IAAI;YACJ,GAAG,EAAE,IAAI,CAAC,IAAI,CAAC,MAAM,EAAE;YACvB,QAAQ;SACR,CAAC,CAAC;QAEH,IAAI,QAAQ,EAAE,CAAC;YACd,IAAI,CAAC,IAAI,CAAC,iBAAiB,CAAC,UAAU,CAAC,CAAC;QACzC,CAAC;QAED,IAAI,UAAU,CAAC,OAAO,KAAK,OAAO,IAAI,UAAU,CAAC,OAAO,KAAK,UAAU,EAAE,CAAC;YACzE,OAAO;gBACN,KAAK,EAAE,IAAI;gBACX,MAAM,EAAE,4CAA4C,UAAU,CAAC,IAAI,MAAM,UAAU,CAAC,OAAO,KAAK,UAAU,CAAC,UAAU,GAAG;aACxH,CAAC;QACH,CAAC;QAED,MAAM,MAAM,GAAG,IAAI,CAAC,IAAI,CAAC,kBAAkB,EAAE,CAAC;QAC9C,IAAI,CAAC,MAAM,CAAC,WAAW,CAAC,WAAW,CAAC,EAAE,CAAC;YACtC,OAAO,SAAS,CAAC;QAClB,CAAC;QAED,IAAI,CAAC;YACJ,OAAO,MAAM,MAAM,CAAC,YAAY,CAAC;gBAChC,IAAI,EAAE,WAAW;gBACjB,QAAQ,EAAE,QAAQ,CAAC,IAAI;gBACvB,UAAU,EAAE,QAAQ,CAAC,EAAE;gBACvB,KAAK,EAAE,IAA+B;aACtC,CAAC,CAAC;QACJ,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACd,IAAI,GAAG,YAAY,KAAK,EAAE,CAAC;gBAC1B,MAAM,GAAG,CAAC;YACX,CAAC;YACD,MAAM,IAAI,KAAK,CAAC,yCAAyC,MAAM,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;QACzE,CAAC;IAAA,CACD,CAAC;IAEO,aAAa,GAAkB,KAAK,EAAE,EAAE,QAAQ,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,EAAE,EAAE,CAAC;QACtF,MAAM,MAAM,GAAG,IAAI,CAAC,IAAI,CAAC,kBAAkB,EAAE,CAAC;QAC9C,IAAI,OAAO,GAAG,MAAM,CAAC,OAAO,CAAC;QAC7B,IAAI,OAAO,GAAG,MAAM,CAAC,OAAO,CAAC;QAC7B,IAAI,eAAe,GAAG,OAAO,CAAC;QAE9B,IAAI,MAAM,CAAC,WAAW,CAAC,aAAa,CAAC,EAAE,CAAC;YACvC,MAAM,UAAU,GAAG,MAAM,MAAM,CAAC,cAAc,CAAC;gBAC9C,IAAI,EAAE,aAAa;gBACnB,QAAQ,EAAE,QAAQ,CAAC,IAAI;gBACvB,UAAU,EAAE,QAAQ,CAAC,EAAE;gBACvB,KAAK,EAAE,IAA+B;gBACtC,OAAO;gBACP,OAAO;gBACP,OAAO;aACP,CAAC,CAAC;YACH,IAAI,UAAU,EAAE,CAAC;gBAChB,OAAO,GAAG,UAAU,CAAC,OAAO,IAAI,OAAO,CAAC;gBACxC,OAAO,GAAG,UAAU,CAAC,OAAO,CAAC;gBAC7B,eAAe,GAAG,UAAU,CAAC,OAAO,IAAI,OAAO,CAAC;YACjD,CAAC;QACF,CAAC;QAED,2FAA2F;QAC3F,+FAA+F;QAC/F,gGAAgG;QAChG,IAAI,iBAAiB,CAAC,QAAQ,CAAC,IAAI,CAAC,KAAK,WAAW,EAAE,CAAC;YACtD,MAAM,MAAM,GAAG,QAAQ,QAAQ,CAAC,IAAI,EAAE,CAAC;YACvC,MAAM,OAAO,GAAG,OAAO,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CACrC,KAAK,CAAC,IAAI,KAAK,MAAM,CAAC,CAAC,CAAC,EAAE,GAAG,KAAK,EAAE,IAAI,EAAE,iBAAiB,CAAC,KAAK,CAAC,IAAI,EAAE,MAAM,CAAC,EAAE,CAAC,CAAC,CAAC,KAAK,CACzF,CAAC;YACF,OAAO,GAAG,OAAO,CAAC;QACnB,CAAC;QAED,IAAI,OAAO,KAAK,MAAM,CAAC,OAAO,IAAI,OAAO,KAAK,MAAM,CAAC,OAAO,IAAI,eAAe,KAAK,OAAO,EAAE,CAAC;YAC7F,OAAO,SAAS,CAAC;QAClB,CAAC;QACD,OAAO,EAAE,OAAO,EAAE,OAAO,EAAE,OAAO,EAAE,eAAe,EAAE,CAAC;IAAA,CACtD,CAAC;CACF","sourcesContent":["/**\n * Agent tool-call gate: model-router escalation, autonomy gating, extension tool hooks, and the\n * untrusted-content output boundary.\n *\n * Extracted verbatim from agent-session.ts (god-file decomposition). Provides the two agent-core\n * hooks (`beforeToolCall`/`afterToolCall`) as bound arrow-field members the session installs onto its\n * agent. `beforeToolCall` runs the router escalation, then the autonomy tool gate (recording the\n * outcome when a capability envelope is active), then extension `tool_call` handlers. `afterToolCall`\n * runs extension `tool_result` handlers and structurally fences output from untrusted-content sources.\n */\n\nimport type { Agent } from \"@caupulican/pi-agent-core\";\nimport type { CapabilityEnvelope, GateOutcome } from \"./autonomy/contracts.ts\";\nimport { evaluateToolGate } from \"./autonomy/gates.ts\";\nimport type { ExtensionRunner } from \"./extensions/index.ts\";\nimport { classifyToolTrust, wrapUntrustedText } from \"./security/untrusted-boundary.ts\";\n\ntype BeforeToolCall = NonNullable<Agent[\"beforeToolCall\"]>;\ntype AfterToolCall = NonNullable<Agent[\"afterToolCall\"]>;\n\nexport interface ToolGateControllerDeps {\n\t/** Router escalation: block a tool the active cheap route is not allowed to run. */\n\tmaybeEscalateToolCall(toolName: string, args: unknown): { block: true; reason: string } | undefined;\n\tgetCwd(): string;\n\t/** Active foreground capability envelope, if any — presence enables gate-outcome recording. */\n\tgetCapabilityEnvelope(): CapabilityEnvelope | undefined;\n\t/** Record an autonomy gate outcome (only when a capability envelope is active). */\n\trecordGateOutcome(outcome: GateOutcome): void;\n\tgetExtensionRunner(): ExtensionRunner;\n}\n\nexport class ToolGateController {\n\tprivate readonly deps: ToolGateControllerDeps;\n\n\tconstructor(deps: ToolGateControllerDeps) {\n\t\tthis.deps = deps;\n\t}\n\n\treadonly beforeToolCall: BeforeToolCall = async ({ toolCall, args }) => {\n\t\tconst escalation = this.deps.maybeEscalateToolCall(toolCall.name, args);\n\t\tif (escalation) {\n\t\t\treturn escalation;\n\t\t}\n\n\t\t// Autonomy tool gating\n\t\tconst envelope = this.deps.getCapabilityEnvelope();\n\t\tconst gateResult = evaluateToolGate({\n\t\t\ttoolName: toolCall.name,\n\t\t\targs,\n\t\t\tcwd: this.deps.getCwd(),\n\t\t\tenvelope,\n\t\t});\n\n\t\tif (envelope) {\n\t\t\tthis.deps.recordGateOutcome(gateResult);\n\t\t}\n\n\t\tif (gateResult.outcome === \"block\" || gateResult.outcome === \"ask-user\") {\n\t\t\treturn {\n\t\t\t\tblock: true,\n\t\t\t\treason: `Tool execution blocked by autonomy gate [${gateResult.gate}]: ${gateResult.message} (${gateResult.reasonCode})`,\n\t\t\t};\n\t\t}\n\n\t\tconst runner = this.deps.getExtensionRunner();\n\t\tif (!runner.hasHandlers(\"tool_call\")) {\n\t\t\treturn undefined;\n\t\t}\n\n\t\ttry {\n\t\t\treturn await runner.emitToolCall({\n\t\t\t\ttype: \"tool_call\",\n\t\t\t\ttoolName: toolCall.name,\n\t\t\t\ttoolCallId: toolCall.id,\n\t\t\t\tinput: args as Record<string, unknown>,\n\t\t\t});\n\t\t} catch (err) {\n\t\t\tif (err instanceof Error) {\n\t\t\t\tthrow err;\n\t\t\t}\n\t\t\tthrow new Error(`Extension failed, blocking execution: ${String(err)}`);\n\t\t}\n\t};\n\n\treadonly afterToolCall: AfterToolCall = async ({ toolCall, args, result, isError }) => {\n\t\tconst runner = this.deps.getExtensionRunner();\n\t\tlet content = result.content;\n\t\tlet details = result.details;\n\t\tlet resolvedIsError = isError;\n\n\t\tif (runner.hasHandlers(\"tool_result\")) {\n\t\t\tconst hookResult = await runner.emitToolResult({\n\t\t\t\ttype: \"tool_result\",\n\t\t\t\ttoolName: toolCall.name,\n\t\t\t\ttoolCallId: toolCall.id,\n\t\t\t\tinput: args as Record<string, unknown>,\n\t\t\t\tcontent,\n\t\t\t\tdetails,\n\t\t\t\tisError,\n\t\t\t});\n\t\t\tif (hookResult) {\n\t\t\t\tcontent = hookResult.content ?? content;\n\t\t\t\tdetails = hookResult.details;\n\t\t\t\tresolvedIsError = hookResult.isError ?? isError;\n\t\t\t}\n\t\t}\n\n\t\t// Untrusted-content boundary: structurally fence output from attacker-controllable sources\n\t\t// (web/search, subagents, recall, third-party tools) so injection payloads are framed as data.\n\t\t// First-party tools (read/grep/find/ls/edit/write/bash) are trusted and pass through unchanged.\n\t\tif (classifyToolTrust(toolCall.name) === \"untrusted\") {\n\t\t\tconst source = `tool:${toolCall.name}`;\n\t\t\tconst wrapped = content.map((block) =>\n\t\t\t\tblock.type === \"text\" ? { ...block, text: wrapUntrustedText(block.text, source) } : block,\n\t\t\t);\n\t\t\tcontent = wrapped;\n\t\t}\n\n\t\tif (content === result.content && details === result.details && resolvedIsError === isError) {\n\t\t\treturn undefined;\n\t\t}\n\t\treturn { content, details, isError: resolvedIsError };\n\t};\n}\n"]}
+{"version":3,"file":"tool-gate-controller.js","sourceRoot":"","sources":["../../src/core/tool-gate-controller.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AAIH,OAAO,EAAE,gBAAgB,EAAE,MAAM,qBAAqB,CAAC;AAEvD,OAAO,EAAE,iBAAiB,EAAE,iBAAiB,EAAE,MAAM,kCAAkC,CAAC;AAmBxF,MAAM,OAAO,kBAAkB;IACb,IAAI,CAAyB;IAE9C,YAAY,IAA4B,EAAE;QACzC,IAAI,CAAC,IAAI,GAAG,IAAI,CAAC;IAAA,CACjB;IAEQ,cAAc,GAAmB,KAAK,EAAE,EAAE,QAAQ,EAAE,IAAI,EAAE,EAAE,EAAE,CAAC;QACvE,MAAM,UAAU,GAAG,IAAI,CAAC,IAAI,CAAC,qBAAqB,CAAC,QAAQ,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC;QACxE,IAAI,UAAU,EAAE,CAAC;YAChB,OAAO,UAAU,CAAC;QACnB,CAAC;QAED,uBAAuB;QACvB,MAAM,QAAQ,GAAG,IAAI,CAAC,IAAI,CAAC,qBAAqB,EAAE,CAAC;QACnD,MAAM,UAAU,GAAG,gBAAgB,CAAC;YACnC,QAAQ,EAAE,QAAQ,CAAC,IAAI;YACvB,IAAI;YACJ,GAAG,EAAE,IAAI,CAAC,IAAI,CAAC,MAAM,EAAE;YACvB,QAAQ;SACR,CAAC,CAAC;QAEH,IAAI,QAAQ,EAAE,CAAC;YACd,IAAI,CAAC,IAAI,CAAC,iBAAiB,CAAC,UAAU,CAAC,CAAC;QACzC,CAAC;QAED,IAAI,UAAU,CAAC,OAAO,KAAK,OAAO,IAAI,UAAU,CAAC,OAAO,KAAK,UAAU,EAAE,CAAC;YACzE,OAAO;gBACN,KAAK,EAAE,IAAI;gBACX,MAAM,EAAE,4CAA4C,UAAU,CAAC,IAAI,MAAM,UAAU,CAAC,OAAO,KAAK,UAAU,CAAC,UAAU,GAAG;aACxH,CAAC;QACH,CAAC;QAED,MAAM,MAAM,GAAG,IAAI,CAAC,IAAI,CAAC,kBAAkB,EAAE,CAAC;QAC9C,IAAI,eAAiD,CAAC;QACtD,IAAI,MAAM,CAAC,WAAW,CAAC,WAAW,CAAC,EAAE,CAAC;YACrC,IAAI,CAAC;gBACJ,eAAe,GAAG,MAAM,MAAM,CAAC,YAAY,CAAC;oBAC3C,IAAI,EAAE,WAAW;oBACjB,QAAQ,EAAE,QAAQ,CAAC,IAAI;oBACvB,UAAU,EAAE,QAAQ,CAAC,EAAE;oBACvB,KAAK,EAAE,IAA+B;iBACtC,CAAC,CAAC;YACJ,CAAC;YAAC,OAAO,GAAG,EAAE,CAAC;gBACd,IAAI,GAAG,YAAY,KAAK,EAAE,CAAC;oBAC1B,MAAM,GAAG,CAAC;gBACX,CAAC;gBACD,MAAM,IAAI,KAAK,CAAC,yCAAyC,MAAM,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;YACzE,CAAC;QACF,CAAC;QACD,IAAI,eAAe;YAAE,OAAO,eAAe,CAAC;QAC5C,IAAI,CAAC,IAAI,CAAC,0BAA0B,EAAE,EAAE,EAAE,KAAK,CAAC,QAAQ,CAAC,EAAE,EAAE,QAAQ,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC;QAClF,OAAO,SAAS,CAAC;IAAA,CACjB,CAAC;IAEO,aAAa,GAAkB,KAAK,EAAE,EAAE,QAAQ,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,EAAE,EAAE,CAAC;QACtF,MAAM,MAAM,GAAG,IAAI,CAAC,IAAI,CAAC,kBAAkB,EAAE,CAAC;QAC9C,IAAI,OAAO,GAAG,MAAM,CAAC,OAAO,CAAC;QAC7B,IAAI,OAAO,GAAG,MAAM,CAAC,OAAO,CAAC;QAC7B,IAAI,eAAe,GAAG,OAAO,CAAC;QAE9B,IAAI,MAAM,CAAC,WAAW,CAAC,aAAa,CAAC,EAAE,CAAC;YACvC,MAAM,UAAU,GAAG,MAAM,MAAM,CAAC,cAAc,CAAC;gBAC9C,IAAI,EAAE,aAAa;gBACnB,QAAQ,EAAE,QAAQ,CAAC,IAAI;gBACvB,UAAU,EAAE,QAAQ,CAAC,EAAE;gBACvB,KAAK,EAAE,IAA+B;gBACtC,OAAO;gBACP,OAAO;gBACP,OAAO;aACP,CAAC,CAAC;YACH,IAAI,UAAU,EAAE,CAAC;gBAChB,OAAO,GAAG,UAAU,CAAC,OAAO,IAAI,OAAO,CAAC;gBACxC,OAAO,GAAG,UAAU,CAAC,OAAO,CAAC;gBAC7B,eAAe,GAAG,UAAU,CAAC,OAAO,IAAI,OAAO,CAAC;YACjD,CAAC;QACF,CAAC;QAED,2FAA2F;QAC3F,+FAA+F;QAC/F,gGAAgG;QAChG,IAAI,iBAAiB,CAAC,QAAQ,CAAC,IAAI,CAAC,KAAK,WAAW,EAAE,CAAC;YACtD,MAAM,MAAM,GAAG,QAAQ,QAAQ,CAAC,IAAI,EAAE,CAAC;YACvC,MAAM,OAAO,GAAG,OAAO,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CACrC,KAAK,CAAC,IAAI,KAAK,MAAM,CAAC,CAAC,CAAC,EAAE,GAAG,KAAK,EAAE,IAAI,EAAE,iBAAiB,CAAC,KAAK,CAAC,IAAI,EAAE,MAAM,CAAC,EAAE,CAAC,CAAC,CAAC,KAAK,CACzF,CAAC;YACF,OAAO,GAAG,OAAO,CAAC;QACnB,CAAC;QAED,IAAI,CAAC,IAAI,CAAC,0BAA0B,EAAE,EAAE,EAAE,QAAQ,CAAC,QAAQ,CAAC,EAAE,EAAE,CAAC,eAAe,EAAE,OAAO,CAAC,CAAC;QAC3F,IAAI,OAAO,KAAK,MAAM,CAAC,OAAO,IAAI,OAAO,KAAK,MAAM,CAAC,OAAO,IAAI,eAAe,KAAK,OAAO,EAAE,CAAC;YAC7F,OAAO,SAAS,CAAC;QAClB,CAAC;QACD,OAAO,EAAE,OAAO,EAAE,OAAO,EAAE,OAAO,EAAE,eAAe,EAAE,CAAC;IAAA,CACtD,CAAC;CACF","sourcesContent":["/**\n * Agent tool-call gate: model-router escalation, autonomy gating, extension tool hooks, and the\n * untrusted-content output boundary.\n *\n * Extracted verbatim from agent-session.ts (god-file decomposition). Provides the two agent-core\n * hooks (`beforeToolCall`/`afterToolCall`) as bound arrow-field members the session installs onto its\n * agent. `beforeToolCall` runs the router escalation, then the autonomy tool gate (recording the\n * outcome when a capability envelope is active), then extension `tool_call` handlers. `afterToolCall`\n * runs extension `tool_result` handlers and structurally fences output from untrusted-content sources.\n */\n\nimport type { Agent, BeforeToolCallResult } from \"@caupulican/pi-agent-core\";\nimport type { CapabilityEnvelope, GateOutcome } from \"./autonomy/contracts.ts\";\nimport { evaluateToolGate } from \"./autonomy/gates.ts\";\nimport type { ExtensionRunner } from \"./extensions/index.ts\";\nimport { classifyToolTrust, wrapUntrustedText } from \"./security/untrusted-boundary.ts\";\nimport type { ToolSelectionController } from \"./tool-selection/tool-selection-controller.ts\";\n\ntype BeforeToolCall = NonNullable<Agent[\"beforeToolCall\"]>;\ntype AfterToolCall = NonNullable<Agent[\"afterToolCall\"]>;\n\nexport interface ToolGateControllerDeps {\n\t/** Router escalation: block a tool the active cheap route is not allowed to run. */\n\tmaybeEscalateToolCall(toolName: string, args: unknown): { block: true; reason: string } | undefined;\n\tgetCwd(): string;\n\t/** Active foreground capability envelope, if any — presence enables gate-outcome recording. */\n\tgetCapabilityEnvelope(): CapabilityEnvelope | undefined;\n\t/** Record an autonomy gate outcome (only when a capability envelope is active). */\n\trecordGateOutcome(outcome: GateOutcome): void;\n\tgetExtensionRunner(): ExtensionRunner;\n\t/** Observe an execution only after all pre-execution gates and extension hooks allow it. */\n\tgetToolSelectionController?(): ToolSelectionController | undefined;\n}\n\nexport class ToolGateController {\n\tprivate readonly deps: ToolGateControllerDeps;\n\n\tconstructor(deps: ToolGateControllerDeps) {\n\t\tthis.deps = deps;\n\t}\n\n\treadonly beforeToolCall: BeforeToolCall = async ({ toolCall, args }) => {\n\t\tconst escalation = this.deps.maybeEscalateToolCall(toolCall.name, args);\n\t\tif (escalation) {\n\t\t\treturn escalation;\n\t\t}\n\n\t\t// Autonomy tool gating\n\t\tconst envelope = this.deps.getCapabilityEnvelope();\n\t\tconst gateResult = evaluateToolGate({\n\t\t\ttoolName: toolCall.name,\n\t\t\targs,\n\t\t\tcwd: this.deps.getCwd(),\n\t\t\tenvelope,\n\t\t});\n\n\t\tif (envelope) {\n\t\t\tthis.deps.recordGateOutcome(gateResult);\n\t\t}\n\n\t\tif (gateResult.outcome === \"block\" || gateResult.outcome === \"ask-user\") {\n\t\t\treturn {\n\t\t\t\tblock: true,\n\t\t\t\treason: `Tool execution blocked by autonomy gate [${gateResult.gate}]: ${gateResult.message} (${gateResult.reasonCode})`,\n\t\t\t};\n\t\t}\n\n\t\tconst runner = this.deps.getExtensionRunner();\n\t\tlet extensionResult: BeforeToolCallResult | undefined;\n\t\tif (runner.hasHandlers(\"tool_call\")) {\n\t\t\ttry {\n\t\t\t\textensionResult = await runner.emitToolCall({\n\t\t\t\t\ttype: \"tool_call\",\n\t\t\t\t\ttoolName: toolCall.name,\n\t\t\t\t\ttoolCallId: toolCall.id,\n\t\t\t\t\tinput: args as Record<string, unknown>,\n\t\t\t\t});\n\t\t\t} catch (err) {\n\t\t\t\tif (err instanceof Error) {\n\t\t\t\t\tthrow err;\n\t\t\t\t}\n\t\t\t\tthrow new Error(`Extension failed, blocking execution: ${String(err)}`);\n\t\t\t}\n\t\t}\n\t\tif (extensionResult) return extensionResult;\n\t\tthis.deps.getToolSelectionController?.()?.begin(toolCall.id, toolCall.name, args);\n\t\treturn undefined;\n\t};\n\n\treadonly afterToolCall: AfterToolCall = async ({ toolCall, args, result, isError }) => {\n\t\tconst runner = this.deps.getExtensionRunner();\n\t\tlet content = result.content;\n\t\tlet details = result.details;\n\t\tlet resolvedIsError = isError;\n\n\t\tif (runner.hasHandlers(\"tool_result\")) {\n\t\t\tconst hookResult = await runner.emitToolResult({\n\t\t\t\ttype: \"tool_result\",\n\t\t\t\ttoolName: toolCall.name,\n\t\t\t\ttoolCallId: toolCall.id,\n\t\t\t\tinput: args as Record<string, unknown>,\n\t\t\t\tcontent,\n\t\t\t\tdetails,\n\t\t\t\tisError,\n\t\t\t});\n\t\t\tif (hookResult) {\n\t\t\t\tcontent = hookResult.content ?? content;\n\t\t\t\tdetails = hookResult.details;\n\t\t\t\tresolvedIsError = hookResult.isError ?? isError;\n\t\t\t}\n\t\t}\n\n\t\t// Untrusted-content boundary: structurally fence output from attacker-controllable sources\n\t\t// (web/search, subagents, recall, third-party tools) so injection payloads are framed as data.\n\t\t// First-party tools (read/grep/find/ls/edit/write/bash) are trusted and pass through unchanged.\n\t\tif (classifyToolTrust(toolCall.name) === \"untrusted\") {\n\t\t\tconst source = `tool:${toolCall.name}`;\n\t\t\tconst wrapped = content.map((block) =>\n\t\t\t\tblock.type === \"text\" ? { ...block, text: wrapUntrustedText(block.text, source) } : block,\n\t\t\t);\n\t\t\tcontent = wrapped;\n\t\t}\n\n\t\tthis.deps.getToolSelectionController?.()?.complete(toolCall.id, !resolvedIsError, content);\n\t\tif (content === result.content && details === result.details && resolvedIsError === isError) {\n\t\t\treturn undefined;\n\t\t}\n\t\treturn { content, details, isError: resolvedIsError };\n\t};\n}\n"]}

package/dist/core/tool-selection/expected-utility.d.ts

@@ -0,0 +1,51 @@
+export interface ExpectedUtilityWeights {
+    latency: number;
+    tokens: number;
+    risk: number;
+    context: number;
+    latencyScaleMs: number;
+    tokenScale: number;
+    minimumEvidence: number;
+    minimumMargin: number;
+    highEntropy: number;
+}
+export declare const DEFAULT_EXPECTED_UTILITY_WEIGHTS: ExpectedUtilityWeights;
+export interface ExpectedUtilityCandidate {
+    tool: string;
+    value: number;
+    alpha: number;
+    beta: number;
+    sampleCount: number;
+    latencyMs?: number;
+    tokenEstimate?: number;
+    riskCost?: number;
+    contextCost?: number;
+    deterministicMatch?: boolean;
+    /** Candidates with unresolved paths may be observed/shortlisted but never recommended. */
+    automaticEligible?: boolean;
+    /** `no_tool` uses a known probability rather than a learned posterior. */
+    successProbability?: number;
+}
+export interface RankedToolCandidate extends ExpectedUtilityCandidate {
+    successProbability: number;
+    latencyCost: number;
+    tokenCost: number;
+    riskCost: number;
+    contextCost: number;
+    utility: number;
+    probability: number;
+}
+export type ToolSelectionDisposition = "recommend" | "shortlist" | "abstain";
+export interface ToolSelectionDecision {
+    disposition: ToolSelectionDisposition;
+    recommendation?: string;
+    shortlist: string[];
+    entropy: number;
+    margin: number;
+    ranked: RankedToolCandidate[];
+}
+export declare function betaSuccessProbability(alpha: number, beta: number): number;
+export declare function normalizedEntropy(probabilities: readonly number[]): number;
+export declare function rankExpectedUtilityCandidates(candidates: readonly ExpectedUtilityCandidate[], weights?: ExpectedUtilityWeights): RankedToolCandidate[];
+export declare function decideExpectedUtility(candidates: readonly ExpectedUtilityCandidate[], weights?: ExpectedUtilityWeights): ToolSelectionDecision;
+//# sourceMappingURL=expected-utility.d.ts.map

package/dist/core/tool-selection/expected-utility.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"expected-utility.d.ts","sourceRoot":"","sources":["../../../src/core/tool-selection/expected-utility.ts"],"names":[],"mappings":"AAAA,MAAM,WAAW,sBAAsB;IACtC,OAAO,EAAE,MAAM,CAAC;IAChB,MAAM,EAAE,MAAM,CAAC;IACf,IAAI,EAAE,MAAM,CAAC;IACb,OAAO,EAAE,MAAM,CAAC;IAChB,cAAc,EAAE,MAAM,CAAC;IACvB,UAAU,EAAE,MAAM,CAAC;IACnB,eAAe,EAAE,MAAM,CAAC;IACxB,aAAa,EAAE,MAAM,CAAC;IACtB,WAAW,EAAE,MAAM,CAAC;CACpB;AAED,eAAO,MAAM,gCAAgC,EAAE,sBAU9C,CAAC;AAEF,MAAM,WAAW,wBAAwB;IACxC,IAAI,EAAE,MAAM,CAAC;IACb,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;IACd,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,EAAE,MAAM,CAAC;IACpB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,kBAAkB,CAAC,EAAE,OAAO,CAAC;IAC7B,0FAA0F;IAC1F,iBAAiB,CAAC,EAAE,OAAO,CAAC;IAC5B,0EAA0E;IAC1E,kBAAkB,CAAC,EAAE,MAAM,CAAC;CAC5B;AAED,MAAM,WAAW,mBAAoB,SAAQ,wBAAwB;IACpE,kBAAkB,EAAE,MAAM,CAAC;IAC3B,WAAW,EAAE,MAAM,CAAC;IACpB,SAAS,EAAE,MAAM,CAAC;IAClB,QAAQ,EAAE,MAAM,CAAC;IACjB,WAAW,EAAE,MAAM,CAAC;IACpB,OAAO,EAAE,MAAM,CAAC;IAChB,WAAW,EAAE,MAAM,CAAC;CACpB;AAED,MAAM,MAAM,wBAAwB,GAAG,WAAW,GAAG,WAAW,GAAG,SAAS,CAAC;AAE7E,MAAM,WAAW,qBAAqB;IACrC,WAAW,EAAE,wBAAwB,CAAC;IACtC,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,SAAS,EAAE,MAAM,EAAE,CAAC;IACpB,OAAO,EAAE,MAAM,CAAC;IAChB,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,mBAAmB,EAAE,CAAC;CAC9B;AAUD,wBAAgB,sBAAsB,CAAC,KAAK,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,GAAG,MAAM,CAK1E;AAED,wBAAgB,iBAAiB,CAAC,aAAa,EAAE,SAAS,MAAM,EAAE,GAAG,MAAM,CAM1E;AAUD,wBAAgB,6BAA6B,CAC5C,UAAU,EAAE,SAAS,wBAAwB,EAAE,EAC/C,OAAO,GAAE,sBAAyD,GAChE,mBAAmB,EAAE,CAqBvB;AAED,wBAAgB,qBAAqB,CACpC,UAAU,EAAE,SAAS,wBAAwB,EAAE,EAC/C,OAAO,GAAE,sBAAyD,GAChE,qBAAqB,CAwBvB","sourcesContent":["export interface ExpectedUtilityWeights {\n\tlatency: number;\n\ttokens: number;\n\trisk: number;\n\tcontext: number;\n\tlatencyScaleMs: number;\n\ttokenScale: number;\n\tminimumEvidence: number;\n\tminimumMargin: number;\n\thighEntropy: number;\n}\n\nexport const DEFAULT_EXPECTED_UTILITY_WEIGHTS: ExpectedUtilityWeights = {\n\tlatency: 0.15,\n\ttokens: 0.1,\n\trisk: 0.2,\n\tcontext: 0.1,\n\tlatencyScaleMs: 5_000,\n\ttokenScale: 1_000,\n\tminimumEvidence: 3,\n\tminimumMargin: 0.1,\n\thighEntropy: 0.85,\n};\n\nexport interface ExpectedUtilityCandidate {\n\ttool: string;\n\tvalue: number;\n\talpha: number;\n\tbeta: number;\n\tsampleCount: number;\n\tlatencyMs?: number;\n\ttokenEstimate?: number;\n\triskCost?: number;\n\tcontextCost?: number;\n\tdeterministicMatch?: boolean;\n\t/** Candidates with unresolved paths may be observed/shortlisted but never recommended. */\n\tautomaticEligible?: boolean;\n\t/** `no_tool` uses a known probability rather than a learned posterior. */\n\tsuccessProbability?: number;\n}\n\nexport interface RankedToolCandidate extends ExpectedUtilityCandidate {\n\tsuccessProbability: number;\n\tlatencyCost: number;\n\ttokenCost: number;\n\triskCost: number;\n\tcontextCost: number;\n\tutility: number;\n\tprobability: number;\n}\n\nexport type ToolSelectionDisposition = \"recommend\" | \"shortlist\" | \"abstain\";\n\nexport interface ToolSelectionDecision {\n\tdisposition: ToolSelectionDisposition;\n\trecommendation?: string;\n\tshortlist: string[];\n\tentropy: number;\n\tmargin: number;\n\tranked: RankedToolCandidate[];\n}\n\nfunction clampUnit(value: number): number {\n\treturn Math.max(0, Math.min(1, value));\n}\n\nfunction finiteOr(value: number | undefined, fallback: number): number {\n\treturn typeof value === \"number\" && Number.isFinite(value) ? value : fallback;\n}\n\nexport function betaSuccessProbability(alpha: number, beta: number): number {\n\tconst positiveAlpha = Math.max(0, finiteOr(alpha, 0));\n\tconst positiveBeta = Math.max(0, finiteOr(beta, 0));\n\tconst total = positiveAlpha + positiveBeta;\n\treturn total > 0 ? positiveAlpha / total : 0.5;\n}\n\nexport function normalizedEntropy(probabilities: readonly number[]): number {\n\tif (probabilities.length <= 1) return 0;\n\tconst positive = probabilities.filter((probability) => probability > 0 && Number.isFinite(probability));\n\tif (positive.length === 0) return 0;\n\tconst entropy = -positive.reduce((total, probability) => total + probability * Math.log(probability), 0);\n\treturn entropy <= 0 ? 0 : entropy / Math.log(probabilities.length);\n}\n\nfunction softmax(utilities: readonly number[]): number[] {\n\tif (utilities.length === 0) return [];\n\tconst maximum = Math.max(...utilities);\n\tconst exponentials = utilities.map((utility) => Math.exp(Math.max(-60, utility - maximum)));\n\tconst total = exponentials.reduce((sum, value) => sum + value, 0);\n\treturn total > 0 ? exponentials.map((value) => value / total) : utilities.map(() => 1 / utilities.length);\n}\n\nexport function rankExpectedUtilityCandidates(\n\tcandidates: readonly ExpectedUtilityCandidate[],\n\tweights: ExpectedUtilityWeights = DEFAULT_EXPECTED_UTILITY_WEIGHTS,\n): RankedToolCandidate[] {\n\tconst unranked = candidates.map((candidate) => {\n\t\tconst successProbability = clampUnit(\n\t\t\tcandidate.successProbability ?? betaSuccessProbability(candidate.alpha, candidate.beta),\n\t\t);\n\t\tconst latencyCost = clampUnit(finiteOr(candidate.latencyMs, 0) / Math.max(1, weights.latencyScaleMs));\n\t\tconst tokenCost = clampUnit(finiteOr(candidate.tokenEstimate, 0) / Math.max(1, weights.tokenScale));\n\t\tconst riskCost = clampUnit(finiteOr(candidate.riskCost, 0));\n\t\tconst contextCost = clampUnit(finiteOr(candidate.contextCost, 0));\n\t\tconst utility =\n\t\t\tsuccessProbability * clampUnit(candidate.value) -\n\t\t\tweights.latency * latencyCost -\n\t\t\tweights.tokens * tokenCost -\n\t\t\tweights.risk * riskCost -\n\t\t\tweights.context * contextCost;\n\t\treturn { ...candidate, successProbability, latencyCost, tokenCost, riskCost, contextCost, utility };\n\t});\n\tconst probabilities = softmax(unranked.map((candidate) => candidate.utility));\n\treturn unranked\n\t\t.map((candidate, index) => ({ ...candidate, probability: probabilities[index] ?? 0 }))\n\t\t.sort((left, right) => right.utility - left.utility || left.tool.localeCompare(right.tool));\n}\n\nexport function decideExpectedUtility(\n\tcandidates: readonly ExpectedUtilityCandidate[],\n\tweights: ExpectedUtilityWeights = DEFAULT_EXPECTED_UTILITY_WEIGHTS,\n): ToolSelectionDecision {\n\tconst ranked = rankExpectedUtilityCandidates(candidates, weights);\n\tconst best = ranked[0];\n\tconst runnerUp = ranked[1];\n\tconst entropy = normalizedEntropy(ranked.map((candidate) => candidate.probability));\n\tconst margin = best && runnerUp ? best.utility - runnerUp.utility : (best?.utility ?? 0);\n\tconst shortlist = ranked.slice(0, 3).map((candidate) => candidate.tool);\n\tif (!best || best.utility <= 0) {\n\t\treturn { disposition: \"abstain\", shortlist: [], entropy, margin, ranked };\n\t}\n\tif (entropy >= weights.highEntropy || margin < weights.minimumMargin) {\n\t\treturn { disposition: \"shortlist\", shortlist, entropy, margin, ranked };\n\t}\n\tif (best.automaticEligible !== false && (best.deterministicMatch || best.sampleCount >= weights.minimumEvidence)) {\n\t\treturn {\n\t\t\tdisposition: \"recommend\",\n\t\t\trecommendation: best.tool,\n\t\t\tshortlist: [],\n\t\t\tentropy,\n\t\t\tmargin,\n\t\t\tranked,\n\t\t};\n\t}\n\treturn { disposition: \"abstain\", shortlist: [], entropy, margin, ranked };\n}\n"]}

package/dist/core/tool-selection/expected-utility.js

@@ -0,0 +1,85 @@
+export const DEFAULT_EXPECTED_UTILITY_WEIGHTS = {
+    latency: 0.15,
+    tokens: 0.1,
+    risk: 0.2,
+    context: 0.1,
+    latencyScaleMs: 5_000,
+    tokenScale: 1_000,
+    minimumEvidence: 3,
+    minimumMargin: 0.1,
+    highEntropy: 0.85,
+};
+function clampUnit(value) {
+    return Math.max(0, Math.min(1, value));
+}
+function finiteOr(value, fallback) {
+    return typeof value === "number" && Number.isFinite(value) ? value : fallback;
+}
+export function betaSuccessProbability(alpha, beta) {
+    const positiveAlpha = Math.max(0, finiteOr(alpha, 0));
+    const positiveBeta = Math.max(0, finiteOr(beta, 0));
+    const total = positiveAlpha + positiveBeta;
+    return total > 0 ? positiveAlpha / total : 0.5;
+}
+export function normalizedEntropy(probabilities) {
+    if (probabilities.length <= 1)
+        return 0;
+    const positive = probabilities.filter((probability) => probability > 0 && Number.isFinite(probability));
+    if (positive.length === 0)
+        return 0;
+    const entropy = -positive.reduce((total, probability) => total + probability * Math.log(probability), 0);
+    return entropy <= 0 ? 0 : entropy / Math.log(probabilities.length);
+}
+function softmax(utilities) {
+    if (utilities.length === 0)
+        return [];
+    const maximum = Math.max(...utilities);
+    const exponentials = utilities.map((utility) => Math.exp(Math.max(-60, utility - maximum)));
+    const total = exponentials.reduce((sum, value) => sum + value, 0);
+    return total > 0 ? exponentials.map((value) => value / total) : utilities.map(() => 1 / utilities.length);
+}
+export function rankExpectedUtilityCandidates(candidates, weights = DEFAULT_EXPECTED_UTILITY_WEIGHTS) {
+    const unranked = candidates.map((candidate) => {
+        const successProbability = clampUnit(candidate.successProbability ?? betaSuccessProbability(candidate.alpha, candidate.beta));
+        const latencyCost = clampUnit(finiteOr(candidate.latencyMs, 0) / Math.max(1, weights.latencyScaleMs));
+        const tokenCost = clampUnit(finiteOr(candidate.tokenEstimate, 0) / Math.max(1, weights.tokenScale));
+        const riskCost = clampUnit(finiteOr(candidate.riskCost, 0));
+        const contextCost = clampUnit(finiteOr(candidate.contextCost, 0));
+        const utility = successProbability * clampUnit(candidate.value) -
+            weights.latency * latencyCost -
+            weights.tokens * tokenCost -
+            weights.risk * riskCost -
+            weights.context * contextCost;
+        return { ...candidate, successProbability, latencyCost, tokenCost, riskCost, contextCost, utility };
+    });
+    const probabilities = softmax(unranked.map((candidate) => candidate.utility));
+    return unranked
+        .map((candidate, index) => ({ ...candidate, probability: probabilities[index] ?? 0 }))
+        .sort((left, right) => right.utility - left.utility || left.tool.localeCompare(right.tool));
+}
+export function decideExpectedUtility(candidates, weights = DEFAULT_EXPECTED_UTILITY_WEIGHTS) {
+    const ranked = rankExpectedUtilityCandidates(candidates, weights);
+    const best = ranked[0];
+    const runnerUp = ranked[1];
+    const entropy = normalizedEntropy(ranked.map((candidate) => candidate.probability));
+    const margin = best && runnerUp ? best.utility - runnerUp.utility : (best?.utility ?? 0);
+    const shortlist = ranked.slice(0, 3).map((candidate) => candidate.tool);
+    if (!best || best.utility <= 0) {
+        return { disposition: "abstain", shortlist: [], entropy, margin, ranked };
+    }
+    if (entropy >= weights.highEntropy || margin < weights.minimumMargin) {
+        return { disposition: "shortlist", shortlist, entropy, margin, ranked };
+    }
+    if (best.automaticEligible !== false && (best.deterministicMatch || best.sampleCount >= weights.minimumEvidence)) {
+        return {
+            disposition: "recommend",
+            recommendation: best.tool,
+            shortlist: [],
+            entropy,
+            margin,
+            ranked,
+        };
+    }
+    return { disposition: "abstain", shortlist: [], entropy, margin, ranked };
+}
+//# sourceMappingURL=expected-utility.js.map

package/dist/core/tool-selection/expected-utility.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"expected-utility.js","sourceRoot":"","sources":["../../../src/core/tool-selection/expected-utility.ts"],"names":[],"mappings":"AAYA,MAAM,CAAC,MAAM,gCAAgC,GAA2B;IACvE,OAAO,EAAE,IAAI;IACb,MAAM,EAAE,GAAG;IACX,IAAI,EAAE,GAAG;IACT,OAAO,EAAE,GAAG;IACZ,cAAc,EAAE,KAAK;IACrB,UAAU,EAAE,KAAK;IACjB,eAAe,EAAE,CAAC;IAClB,aAAa,EAAE,GAAG;IAClB,WAAW,EAAE,IAAI;CACjB,CAAC;AAwCF,SAAS,SAAS,CAAC,KAAa,EAAU;IACzC,OAAO,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,KAAK,CAAC,CAAC,CAAC;AAAA,CACvC;AAED,SAAS,QAAQ,CAAC,KAAyB,EAAE,QAAgB,EAAU;IACtE,OAAO,OAAO,KAAK,KAAK,QAAQ,IAAI,MAAM,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,QAAQ,CAAC;AAAA,CAC9E;AAED,MAAM,UAAU,sBAAsB,CAAC,KAAa,EAAE,IAAY,EAAU;IAC3E,MAAM,aAAa,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,QAAQ,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC,CAAC;IACtD,MAAM,YAAY,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,QAAQ,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC;IACpD,MAAM,KAAK,GAAG,aAAa,GAAG,YAAY,CAAC;IAC3C,OAAO,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,aAAa,GAAG,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC;AAAA,CAC/C;AAED,MAAM,UAAU,iBAAiB,CAAC,aAAgC,EAAU;IAC3E,IAAI,aAAa,CAAC,MAAM,IAAI,CAAC;QAAE,OAAO,CAAC,CAAC;IACxC,MAAM,QAAQ,GAAG,aAAa,CAAC,MAAM,CAAC,CAAC,WAAW,EAAE,EAAE,CAAC,WAAW,GAAG,CAAC,IAAI,MAAM,CAAC,QAAQ,CAAC,WAAW,CAAC,CAAC,CAAC;IACxG,IAAI,QAAQ,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,CAAC,CAAC;IACpC,MAAM,OAAO,GAAG,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,WAAW,EAAE,EAAE,CAAC,KAAK,GAAG,WAAW,GAAG,IAAI,CAAC,GAAG,CAAC,WAAW,CAAC,EAAE,CAAC,CAAC,CAAC;IACzG,OAAO,OAAO,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,OAAO,GAAG,IAAI,CAAC,GAAG,CAAC,aAAa,CAAC,MAAM,CAAC,CAAC;AAAA,CACnE;AAED,SAAS,OAAO,CAAC,SAA4B,EAAY;IACxD,IAAI,SAAS,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,EAAE,CAAC;IACtC,MAAM,OAAO,GAAG,IAAI,CAAC,GAAG,CAAC,GAAG,SAAS,CAAC,CAAC;IACvC,MAAM,YAAY,GAAG,SAAS,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,EAAE,OAAO,GAAG,OAAO,CAAC,CAAC,CAAC,CAAC;IAC5F,MAAM,KAAK,GAAG,YAAY,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,KAAK,EAAE,EAAE,CAAC,GAAG,GAAG,KAAK,EAAE,CAAC,CAAC,CAAC;IAClE,OAAO,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,YAAY,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,GAAG,CAAC,GAAG,EAAE,CAAC,CAAC,GAAG,SAAS,CAAC,MAAM,CAAC,CAAC;AAAA,CAC1G;AAED,MAAM,UAAU,6BAA6B,CAC5C,UAA+C,EAC/C,OAAO,GAA2B,gCAAgC,EAC1C;IACxB,MAAM,QAAQ,GAAG,UAAU,CAAC,GAAG,CAAC,CAAC,SAAS,EAAE,EAAE,CAAC;QAC9C,MAAM,kBAAkB,GAAG,SAAS,CACnC,SAAS,CAAC,kBAAkB,IAAI,sBAAsB,CAAC,SAAS,CAAC,KAAK,EAAE,SAAS,CAAC,IAAI,CAAC,CACvF,CAAC;QACF,MAAM,WAAW,GAAG,SAAS,CAAC,QAAQ,CAAC,SAAS,CAAC,SAAS,EAAE,CAAC,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,OAAO,CAAC,cAAc,CAAC,CAAC,CAAC;QACtG,MAAM,SAAS,GAAG,SAAS,CAAC,QAAQ,CAAC,SAAS,CAAC,aAAa,EAAE,CAAC,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,OAAO,CAAC,UAAU,CAAC,CAAC,CAAC;QACpG,MAAM,QAAQ,GAAG,SAAS,CAAC,QAAQ,CAAC,SAAS,CAAC,QAAQ,EAAE,CAAC,CAAC,CAAC,CAAC;QAC5D,MAAM,WAAW,GAAG,SAAS,CAAC,QAAQ,CAAC,SAAS,CAAC,WAAW,EAAE,CAAC,CAAC,CAAC,CAAC;QAClE,MAAM,OAAO,GACZ,kBAAkB,GAAG,SAAS,CAAC,SAAS,CAAC,KAAK,CAAC;YAC/C,OAAO,CAAC,OAAO,GAAG,WAAW;YAC7B,OAAO,CAAC,MAAM,GAAG,SAAS;YAC1B,OAAO,CAAC,IAAI,GAAG,QAAQ;YACvB,OAAO,CAAC,OAAO,GAAG,WAAW,CAAC;QAC/B,OAAO,EAAE,GAAG,SAAS,EAAE,kBAAkB,EAAE,WAAW,EAAE,SAAS,EAAE,QAAQ,EAAE,WAAW,EAAE,OAAO,EAAE,CAAC;IAAA,CACpG,CAAC,CAAC;IACH,MAAM,aAAa,GAAG,OAAO,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,SAAS,EAAE,EAAE,CAAC,SAAS,CAAC,OAAO,CAAC,CAAC,CAAC;IAC9E,OAAO,QAAQ;SACb,GAAG,CAAC,CAAC,SAAS,EAAE,KAAK,EAAE,EAAE,CAAC,CAAC,EAAE,GAAG,SAAS,EAAE,WAAW,EAAE,aAAa,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;SACrF,IAAI,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,OAAO,GAAG,IAAI,CAAC,OAAO,IAAI,IAAI,CAAC,IAAI,CAAC,aAAa,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC;AAAA,CAC7F;AAED,MAAM,UAAU,qBAAqB,CACpC,UAA+C,EAC/C,OAAO,GAA2B,gCAAgC,EAC1C;IACxB,MAAM,MAAM,GAAG,6BAA6B,CAAC,UAAU,EAAE,OAAO,CAAC,CAAC;IAClE,MAAM,IAAI,GAAG,MAAM,CAAC,CAAC,CAAC,CAAC;IACvB,MAAM,QAAQ,GAAG,MAAM,CAAC,CAAC,CAAC,CAAC;IAC3B,MAAM,OAAO,GAAG,iBAAiB,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,SAAS,EAAE,EAAE,CAAC,SAAS,CAAC,WAAW,CAAC,CAAC,CAAC;IACpF,MAAM,MAAM,GAAG,IAAI,IAAI,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,OAAO,GAAG,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,OAAO,IAAI,CAAC,CAAC,CAAC;IACzF,MAAM,SAAS,GAAG,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,SAAS,EAAE,EAAE,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC;IACxE,IAAI,CAAC,IAAI,IAAI,IAAI,CAAC,OAAO,IAAI,CAAC,EAAE,CAAC;QAChC,OAAO,EAAE,WAAW,EAAE,SAAS,EAAE,SAAS,EAAE,EAAE,EAAE,OAAO,EAAE,MAAM,EAAE,MAAM,EAAE,CAAC;IAC3E,CAAC;IACD,IAAI,OAAO,IAAI,OAAO,CAAC,WAAW,IAAI,MAAM,GAAG,OAAO,CAAC,aAAa,EAAE,CAAC;QACtE,OAAO,EAAE,WAAW,EAAE,WAAW,EAAE,SAAS,EAAE,OAAO,EAAE,MAAM,EAAE,MAAM,EAAE,CAAC;IACzE,CAAC;IACD,IAAI,IAAI,CAAC,iBAAiB,KAAK,KAAK,IAAI,CAAC,IAAI,CAAC,kBAAkB,IAAI,IAAI,CAAC,WAAW,IAAI,OAAO,CAAC,eAAe,CAAC,EAAE,CAAC;QAClH,OAAO;YACN,WAAW,EAAE,WAAW;YACxB,cAAc,EAAE,IAAI,CAAC,IAAI;YACzB,SAAS,EAAE,EAAE;YACb,OAAO;YACP,MAAM;YACN,MAAM;SACN,CAAC;IACH,CAAC;IACD,OAAO,EAAE,WAAW,EAAE,SAAS,EAAE,SAAS,EAAE,EAAE,EAAE,OAAO,EAAE,MAAM,EAAE,MAAM,EAAE,CAAC;AAAA,CAC1E","sourcesContent":["export interface ExpectedUtilityWeights {\n\tlatency: number;\n\ttokens: number;\n\trisk: number;\n\tcontext: number;\n\tlatencyScaleMs: number;\n\ttokenScale: number;\n\tminimumEvidence: number;\n\tminimumMargin: number;\n\thighEntropy: number;\n}\n\nexport const DEFAULT_EXPECTED_UTILITY_WEIGHTS: ExpectedUtilityWeights = {\n\tlatency: 0.15,\n\ttokens: 0.1,\n\trisk: 0.2,\n\tcontext: 0.1,\n\tlatencyScaleMs: 5_000,\n\ttokenScale: 1_000,\n\tminimumEvidence: 3,\n\tminimumMargin: 0.1,\n\thighEntropy: 0.85,\n};\n\nexport interface ExpectedUtilityCandidate {\n\ttool: string;\n\tvalue: number;\n\talpha: number;\n\tbeta: number;\n\tsampleCount: number;\n\tlatencyMs?: number;\n\ttokenEstimate?: number;\n\triskCost?: number;\n\tcontextCost?: number;\n\tdeterministicMatch?: boolean;\n\t/** Candidates with unresolved paths may be observed/shortlisted but never recommended. */\n\tautomaticEligible?: boolean;\n\t/** `no_tool` uses a known probability rather than a learned posterior. */\n\tsuccessProbability?: number;\n}\n\nexport interface RankedToolCandidate extends ExpectedUtilityCandidate {\n\tsuccessProbability: number;\n\tlatencyCost: number;\n\ttokenCost: number;\n\triskCost: number;\n\tcontextCost: number;\n\tutility: number;\n\tprobability: number;\n}\n\nexport type ToolSelectionDisposition = \"recommend\" | \"shortlist\" | \"abstain\";\n\nexport interface ToolSelectionDecision {\n\tdisposition: ToolSelectionDisposition;\n\trecommendation?: string;\n\tshortlist: string[];\n\tentropy: number;\n\tmargin: number;\n\tranked: RankedToolCandidate[];\n}\n\nfunction clampUnit(value: number): number {\n\treturn Math.max(0, Math.min(1, value));\n}\n\nfunction finiteOr(value: number | undefined, fallback: number): number {\n\treturn typeof value === \"number\" && Number.isFinite(value) ? value : fallback;\n}\n\nexport function betaSuccessProbability(alpha: number, beta: number): number {\n\tconst positiveAlpha = Math.max(0, finiteOr(alpha, 0));\n\tconst positiveBeta = Math.max(0, finiteOr(beta, 0));\n\tconst total = positiveAlpha + positiveBeta;\n\treturn total > 0 ? positiveAlpha / total : 0.5;\n}\n\nexport function normalizedEntropy(probabilities: readonly number[]): number {\n\tif (probabilities.length <= 1) return 0;\n\tconst positive = probabilities.filter((probability) => probability > 0 && Number.isFinite(probability));\n\tif (positive.length === 0) return 0;\n\tconst entropy = -positive.reduce((total, probability) => total + probability * Math.log(probability), 0);\n\treturn entropy <= 0 ? 0 : entropy / Math.log(probabilities.length);\n}\n\nfunction softmax(utilities: readonly number[]): number[] {\n\tif (utilities.length === 0) return [];\n\tconst maximum = Math.max(...utilities);\n\tconst exponentials = utilities.map((utility) => Math.exp(Math.max(-60, utility - maximum)));\n\tconst total = exponentials.reduce((sum, value) => sum + value, 0);\n\treturn total > 0 ? exponentials.map((value) => value / total) : utilities.map(() => 1 / utilities.length);\n}\n\nexport function rankExpectedUtilityCandidates(\n\tcandidates: readonly ExpectedUtilityCandidate[],\n\tweights: ExpectedUtilityWeights = DEFAULT_EXPECTED_UTILITY_WEIGHTS,\n): RankedToolCandidate[] {\n\tconst unranked = candidates.map((candidate) => {\n\t\tconst successProbability = clampUnit(\n\t\t\tcandidate.successProbability ?? betaSuccessProbability(candidate.alpha, candidate.beta),\n\t\t);\n\t\tconst latencyCost = clampUnit(finiteOr(candidate.latencyMs, 0) / Math.max(1, weights.latencyScaleMs));\n\t\tconst tokenCost = clampUnit(finiteOr(candidate.tokenEstimate, 0) / Math.max(1, weights.tokenScale));\n\t\tconst riskCost = clampUnit(finiteOr(candidate.riskCost, 0));\n\t\tconst contextCost = clampUnit(finiteOr(candidate.contextCost, 0));\n\t\tconst utility =\n\t\t\tsuccessProbability * clampUnit(candidate.value) -\n\t\t\tweights.latency * latencyCost -\n\t\t\tweights.tokens * tokenCost -\n\t\t\tweights.risk * riskCost -\n\t\t\tweights.context * contextCost;\n\t\treturn { ...candidate, successProbability, latencyCost, tokenCost, riskCost, contextCost, utility };\n\t});\n\tconst probabilities = softmax(unranked.map((candidate) => candidate.utility));\n\treturn unranked\n\t\t.map((candidate, index) => ({ ...candidate, probability: probabilities[index] ?? 0 }))\n\t\t.sort((left, right) => right.utility - left.utility || left.tool.localeCompare(right.tool));\n}\n\nexport function decideExpectedUtility(\n\tcandidates: readonly ExpectedUtilityCandidate[],\n\tweights: ExpectedUtilityWeights = DEFAULT_EXPECTED_UTILITY_WEIGHTS,\n): ToolSelectionDecision {\n\tconst ranked = rankExpectedUtilityCandidates(candidates, weights);\n\tconst best = ranked[0];\n\tconst runnerUp = ranked[1];\n\tconst entropy = normalizedEntropy(ranked.map((candidate) => candidate.probability));\n\tconst margin = best && runnerUp ? best.utility - runnerUp.utility : (best?.utility ?? 0);\n\tconst shortlist = ranked.slice(0, 3).map((candidate) => candidate.tool);\n\tif (!best || best.utility <= 0) {\n\t\treturn { disposition: \"abstain\", shortlist: [], entropy, margin, ranked };\n\t}\n\tif (entropy >= weights.highEntropy || margin < weights.minimumMargin) {\n\t\treturn { disposition: \"shortlist\", shortlist, entropy, margin, ranked };\n\t}\n\tif (best.automaticEligible !== false && (best.deterministicMatch || best.sampleCount >= weights.minimumEvidence)) {\n\t\treturn {\n\t\t\tdisposition: \"recommend\",\n\t\t\trecommendation: best.tool,\n\t\t\tshortlist: [],\n\t\t\tentropy,\n\t\t\tmargin,\n\t\t\tranked,\n\t\t};\n\t}\n\treturn { disposition: \"abstain\", shortlist: [], entropy, margin, ranked };\n}\n"]}

package/dist/core/tool-selection/tool-performance-store.d.ts

@@ -0,0 +1,82 @@
+import { type HostFingerprint } from "../models/fitness-store.ts";
+export type ToolSelectionIntentClass = "read" | "search" | "execute" | "write" | "retrieve" | "explain" | "other";
+export interface ToolPerformanceKey {
+    modelRef: string;
+    intentClass: ToolSelectionIntentClass;
+    tool: string;
+}
+export interface ToolPerformanceStats extends ToolPerformanceKey {
+    alpha: number;
+    beta: number;
+    sampleCount: number;
+    latencyEwmaMs?: number;
+    latencyDeviationEwmaMs?: number;
+    inputTokenEstimateEwma?: number;
+    outputTokenEstimateEwma?: number;
+    repairCount: number;
+    bounceCount: number;
+    failureCount: number;
+    lastUsedAt: string;
+}
+export interface ToolSelectionObservation {
+    at: string;
+    modelRef: string;
+    intentClass: ToolSelectionIntentClass;
+    actualTool: string;
+    firstTool: boolean;
+    succeeded: boolean;
+    disposition: "recommend" | "shortlist" | "abstain";
+    recommendation?: string;
+    shortlist: string[];
+    entropy: number;
+    margin: number;
+    /** Redacted ranking: names and numeric scores only, never prompts, arguments, paths, or output. */
+    ranked: Array<{
+        tool: string;
+        utility: number;
+        probability: number;
+    }>;
+    latencyMs?: number;
+    inputTokenEstimate?: number;
+    outputTokenEstimate?: number;
+}
+export interface ToolSelectionMetrics {
+    firstToolAttempts: number;
+    firstToolSuccesses: number;
+    wrongToolOrFailureCount: number;
+    recommendationCount: number;
+    recommendationMatchedCount: number;
+    shortlistCount: number;
+    abstentionCount: number;
+    averageLatencyMs?: number;
+    averageInputTokenEstimate?: number;
+    averageOutputTokenEstimate?: number;
+}
+export interface ToolExecutionObservation {
+    key: ToolPerformanceKey;
+    success: boolean;
+    latencyMs: number;
+    inputTokenEstimate?: number;
+    outputTokenEstimate?: number;
+    selection: Omit<ToolSelectionObservation, "at" | "modelRef" | "intentClass" | "actualTool" | "succeeded">;
+    at?: string;
+}
+export declare class ToolPerformanceStore {
+    private readonly filePath;
+    private readonly fingerprint;
+    constructor(filePath: string, options?: {
+        fingerprint?: () => HostFingerprint;
+    });
+    static forAgentDir(agentDir: string, options?: {
+        fingerprint?: () => HostFingerprint;
+    }): ToolPerformanceStore;
+    private load;
+    private save;
+    private hostData;
+    get(key: ToolPerformanceKey): ToolPerformanceStats;
+    recordValidation(key: ToolPerformanceKey, outcome: "repaired" | "bounced", at?: string): ToolPerformanceStats;
+    recordExecution(observation: ToolExecutionObservation): ToolPerformanceStats;
+    getMetrics(modelRef?: string): ToolSelectionMetrics;
+    getObservations(modelRef?: string): ToolSelectionObservation[];
+}
+//# sourceMappingURL=tool-performance-store.d.ts.map

package/dist/core/tool-selection/tool-performance-store.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"tool-performance-store.d.ts","sourceRoot":"","sources":["../../../src/core/tool-selection/tool-performance-store.ts"],"names":[],"mappings":"AAEA,OAAO,EAA0B,KAAK,eAAe,EAAE,MAAM,4BAA4B,CAAC;AAO1F,MAAM,MAAM,wBAAwB,GAAG,MAAM,GAAG,QAAQ,GAAG,SAAS,GAAG,OAAO,GAAG,UAAU,GAAG,SAAS,GAAG,OAAO,CAAC;AAElH,MAAM,WAAW,kBAAkB;IAClC,QAAQ,EAAE,MAAM,CAAC;IACjB,WAAW,EAAE,wBAAwB,CAAC;IACtC,IAAI,EAAE,MAAM,CAAC;CACb;AAED,MAAM,WAAW,oBAAqB,SAAQ,kBAAkB;IAC/D,KAAK,EAAE,MAAM,CAAC;IACd,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,EAAE,MAAM,CAAC;IACpB,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,sBAAsB,CAAC,EAAE,MAAM,CAAC;IAChC,sBAAsB,CAAC,EAAE,MAAM,CAAC;IAChC,uBAAuB,CAAC,EAAE,MAAM,CAAC;IACjC,WAAW,EAAE,MAAM,CAAC;IACpB,WAAW,EAAE,MAAM,CAAC;IACpB,YAAY,EAAE,MAAM,CAAC;IACrB,UAAU,EAAE,MAAM,CAAC;CACnB;AAED,MAAM,WAAW,wBAAwB;IACxC,EAAE,EAAE,MAAM,CAAC;IACX,QAAQ,EAAE,MAAM,CAAC;IACjB,WAAW,EAAE,wBAAwB,CAAC;IACtC,UAAU,EAAE,MAAM,CAAC;IACnB,SAAS,EAAE,OAAO,CAAC;IACnB,SAAS,EAAE,OAAO,CAAC;IACnB,WAAW,EAAE,WAAW,GAAG,WAAW,GAAG,SAAS,CAAC;IACnD,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,SAAS,EAAE,MAAM,EAAE,CAAC;IACpB,OAAO,EAAE,MAAM,CAAC;IAChB,MAAM,EAAE,MAAM,CAAC;IACf,mGAAmG;IACnG,MAAM,EAAE,KAAK,CAAC;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,OAAO,EAAE,MAAM,CAAC;QAAC,WAAW,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IACtE,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,kBAAkB,CAAC,EAAE,MAAM,CAAC;IAC5B,mBAAmB,CAAC,EAAE,MAAM,CAAC;CAC7B;AAED,MAAM,WAAW,oBAAoB;IACpC,iBAAiB,EAAE,MAAM,CAAC;IAC1B,kBAAkB,EAAE,MAAM,CAAC;IAC3B,uBAAuB,EAAE,MAAM,CAAC;IAChC,mBAAmB,EAAE,MAAM,CAAC;IAC5B,0BAA0B,EAAE,MAAM,CAAC;IACnC,cAAc,EAAE,MAAM,CAAC;IACvB,eAAe,EAAE,MAAM,CAAC;IACxB,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,yBAAyB,CAAC,EAAE,MAAM,CAAC;IACnC,0BAA0B,CAAC,EAAE,MAAM,CAAC;CACpC;AAaD,MAAM,WAAW,wBAAwB;IACxC,GAAG,EAAE,kBAAkB,CAAC;IACxB,OAAO,EAAE,OAAO,CAAC;IACjB,SAAS,EAAE,MAAM,CAAC;IAClB,kBAAkB,CAAC,EAAE,MAAM,CAAC;IAC5B,mBAAmB,CAAC,EAAE,MAAM,CAAC;IAC7B,SAAS,EAAE,IAAI,CAAC,wBAAwB,EAAE,IAAI,GAAG,UAAU,GAAG,aAAa,GAAG,YAAY,GAAG,WAAW,CAAC,CAAC;IAC1G,EAAE,CAAC,EAAE,MAAM,CAAC;CACZ;AA8ID,qBAAa,oBAAoB;IAChC,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAS;IAClC,OAAO,CAAC,QAAQ,CAAC,WAAW,CAAwB;IAEpD,YAAY,QAAQ,EAAE,MAAM,EAAE,OAAO,GAAE;QAAE,WAAW,CAAC,EAAE,MAAM,eAAe,CAAA;KAAO,EAGlF;IAED,MAAM,CAAC,WAAW,CAAC,QAAQ,EAAE,MAAM,EAAE,OAAO,GAAE;QAAE,WAAW,CAAC,EAAE,MAAM,eAAe,CAAA;KAAO,GAAG,oBAAoB,CAEhH;IAED,OAAO,CAAC,IAAI;IASZ,OAAO,CAAC,IAAI;IAKZ,OAAO,CAAC,QAAQ;IAShB,GAAG,CAAC,GAAG,EAAE,kBAAkB,GAAG,oBAAoB,CAKjD;IAED,gBAAgB,CACf,GAAG,EAAE,kBAAkB,EACvB,OAAO,EAAE,UAAU,GAAG,SAAS,EAC/B,EAAE,SAA2B,GAC3B,oBAAoB,CAetB;IAED,eAAe,CAAC,WAAW,EAAE,wBAAwB,GAAG,oBAAoB,CAyC3E;IAED,UAAU,CAAC,QAAQ,CAAC,EAAE,MAAM,GAAG,oBAAoB,CA6BlD;IAED,eAAe,CAAC,QAAQ,CAAC,EAAE,MAAM,GAAG,wBAAwB,EAAE,CAS7D;CACD","sourcesContent":["import { existsSync, mkdirSync, readFileSync, writeFileSync } from \"node:fs\";\nimport { dirname, join } from \"node:path\";\nimport { currentHostFingerprint, type HostFingerprint } from \"../models/fitness-store.ts\";\n\nconst STORE_VERSION = 1;\nconst MAX_STATS_PER_HOST = 500;\nconst MAX_OBSERVATIONS_PER_HOST = 1_000;\nconst EWMA_ALPHA = 0.25;\n\nexport type ToolSelectionIntentClass = \"read\" | \"search\" | \"execute\" | \"write\" | \"retrieve\" | \"explain\" | \"other\";\n\nexport interface ToolPerformanceKey {\n\tmodelRef: string;\n\tintentClass: ToolSelectionIntentClass;\n\ttool: string;\n}\n\nexport interface ToolPerformanceStats extends ToolPerformanceKey {\n\talpha: number;\n\tbeta: number;\n\tsampleCount: number;\n\tlatencyEwmaMs?: number;\n\tlatencyDeviationEwmaMs?: number;\n\tinputTokenEstimateEwma?: number;\n\toutputTokenEstimateEwma?: number;\n\trepairCount: number;\n\tbounceCount: number;\n\tfailureCount: number;\n\tlastUsedAt: string;\n}\n\nexport interface ToolSelectionObservation {\n\tat: string;\n\tmodelRef: string;\n\tintentClass: ToolSelectionIntentClass;\n\tactualTool: string;\n\tfirstTool: boolean;\n\tsucceeded: boolean;\n\tdisposition: \"recommend\" | \"shortlist\" | \"abstain\";\n\trecommendation?: string;\n\tshortlist: string[];\n\tentropy: number;\n\tmargin: number;\n\t/** Redacted ranking: names and numeric scores only, never prompts, arguments, paths, or output. */\n\tranked: Array<{ tool: string; utility: number; probability: number }>;\n\tlatencyMs?: number;\n\tinputTokenEstimate?: number;\n\toutputTokenEstimate?: number;\n}\n\nexport interface ToolSelectionMetrics {\n\tfirstToolAttempts: number;\n\tfirstToolSuccesses: number;\n\twrongToolOrFailureCount: number;\n\trecommendationCount: number;\n\trecommendationMatchedCount: number;\n\tshortlistCount: number;\n\tabstentionCount: number;\n\taverageLatencyMs?: number;\n\taverageInputTokenEstimate?: number;\n\taverageOutputTokenEstimate?: number;\n}\n\ninterface HostToolPerformanceData {\n\thost: HostFingerprint;\n\tstats: Record<string, ToolPerformanceStats>;\n\tobservations: ToolSelectionObservation[];\n}\n\ninterface ToolPerformanceStoreFile {\n\tversion: 1;\n\thosts: Record<string, HostToolPerformanceData>;\n}\n\nexport interface ToolExecutionObservation {\n\tkey: ToolPerformanceKey;\n\tsuccess: boolean;\n\tlatencyMs: number;\n\tinputTokenEstimate?: number;\n\toutputTokenEstimate?: number;\n\tselection: Omit<ToolSelectionObservation, \"at\" | \"modelRef\" | \"intentClass\" | \"actualTool\" | \"succeeded\">;\n\tat?: string;\n}\n\nfunction statKey(key: ToolPerformanceKey): string {\n\treturn `${key.modelRef}\\0${key.intentClass}\\0${key.tool}`;\n}\n\nfunction isRecord(value: unknown): value is Record<string, unknown> {\n\treturn typeof value === \"object\" && value !== null && !Array.isArray(value);\n}\n\nfunction isIntentClass(value: unknown): value is ToolSelectionIntentClass {\n\treturn (\n\t\tvalue === \"read\" ||\n\t\tvalue === \"search\" ||\n\t\tvalue === \"execute\" ||\n\t\tvalue === \"write\" ||\n\t\tvalue === \"retrieve\" ||\n\t\tvalue === \"explain\" ||\n\t\tvalue === \"other\"\n\t);\n}\n\nfunction isStats(value: unknown): value is ToolPerformanceStats {\n\treturn (\n\t\tisRecord(value) &&\n\t\ttypeof value.modelRef === \"string\" &&\n\t\tisIntentClass(value.intentClass) &&\n\t\ttypeof value.tool === \"string\" &&\n\t\ttypeof value.alpha === \"number\" &&\n\t\ttypeof value.beta === \"number\" &&\n\t\ttypeof value.sampleCount === \"number\" &&\n\t\ttypeof value.repairCount === \"number\" &&\n\t\ttypeof value.bounceCount === \"number\" &&\n\t\ttypeof value.failureCount === \"number\" &&\n\t\ttypeof value.lastUsedAt === \"string\"\n\t);\n}\n\nfunction isObservation(value: unknown): value is ToolSelectionObservation {\n\treturn (\n\t\tisRecord(value) &&\n\t\ttypeof value.at === \"string\" &&\n\t\ttypeof value.modelRef === \"string\" &&\n\t\tisIntentClass(value.intentClass) &&\n\t\ttypeof value.actualTool === \"string\" &&\n\t\ttypeof value.firstTool === \"boolean\" &&\n\t\ttypeof value.succeeded === \"boolean\" &&\n\t\t(value.disposition === \"recommend\" || value.disposition === \"shortlist\" || value.disposition === \"abstain\") &&\n\t\tArray.isArray(value.shortlist) &&\n\t\tvalue.shortlist.every((tool) => typeof tool === \"string\") &&\n\t\ttypeof value.entropy === \"number\" &&\n\t\ttypeof value.margin === \"number\" &&\n\t\tArray.isArray(value.ranked) &&\n\t\tvalue.ranked.every(\n\t\t\t(entry) =>\n\t\t\t\tisRecord(entry) &&\n\t\t\t\ttypeof entry.tool === \"string\" &&\n\t\t\t\ttypeof entry.utility === \"number\" &&\n\t\t\t\ttypeof entry.probability === \"number\",\n\t\t) &&\n\t\t(value.latencyMs === undefined || typeof value.latencyMs === \"number\") &&\n\t\t(value.inputTokenEstimate === undefined || typeof value.inputTokenEstimate === \"number\") &&\n\t\t(value.outputTokenEstimate === undefined || typeof value.outputTokenEstimate === \"number\")\n\t);\n}\n\nfunction emptyStats(key: ToolPerformanceKey, at: string): ToolPerformanceStats {\n\treturn {\n\t\t...key,\n\t\talpha: 1,\n\t\tbeta: 1,\n\t\tsampleCount: 0,\n\t\trepairCount: 0,\n\t\tbounceCount: 0,\n\t\tfailureCount: 0,\n\t\tlastUsedAt: at,\n\t};\n}\n\nfunction finiteNonNegative(value: number | undefined): number | undefined {\n\treturn typeof value === \"number\" && Number.isFinite(value) && value >= 0 ? value : undefined;\n}\n\nfunction updateEwma(previous: number | undefined, next: number | undefined): number | undefined {\n\tif (next === undefined) return previous;\n\treturn previous === undefined ? next : previous + EWMA_ALPHA * (next - previous);\n}\n\nfunction updateDeviation(\n\tpreviousValue: number | undefined,\n\tpreviousDeviation: number | undefined,\n\tnext: number | undefined,\n): number | undefined {\n\tif (next === undefined) return previousDeviation;\n\tconst deviation = previousValue === undefined ? 0 : Math.abs(next - previousValue);\n\treturn updateEwma(previousDeviation, deviation);\n}\n\nfunction parseHost(value: unknown): HostToolPerformanceData | undefined {\n\tif (!isRecord(value) || !isRecord(value.host) || !isRecord(value.stats) || !Array.isArray(value.observations))\n\t\treturn undefined;\n\tconst host = value.host;\n\tif (\n\t\ttypeof host.id !== \"string\" ||\n\t\ttypeof host.cpu !== \"string\" ||\n\t\ttypeof host.cores !== \"number\" ||\n\t\ttypeof host.totalMemGb !== \"number\"\n\t)\n\t\treturn undefined;\n\treturn {\n\t\thost: { id: host.id, cpu: host.cpu, cores: host.cores, totalMemGb: host.totalMemGb },\n\t\tstats: Object.fromEntries(\n\t\t\tObject.entries(value.stats).filter((entry): entry is [string, ToolPerformanceStats] => isStats(entry[1])),\n\t\t),\n\t\tobservations: value.observations.filter(isObservation),\n\t};\n}\n\nfunction parseFile(value: unknown): ToolPerformanceStoreFile {\n\tif (!isRecord(value) || value.version !== STORE_VERSION || !isRecord(value.hosts)) {\n\t\treturn { version: STORE_VERSION, hosts: {} };\n\t}\n\treturn {\n\t\tversion: STORE_VERSION,\n\t\thosts: Object.fromEntries(\n\t\t\tObject.entries(value.hosts)\n\t\t\t\t.map(([hostId, host]) => [hostId, parseHost(host)] as const)\n\t\t\t\t.filter((entry): entry is readonly [string, HostToolPerformanceData] => entry[1] !== undefined),\n\t\t),\n\t};\n}\n\nfunction trimStats(stats: Record<string, ToolPerformanceStats>): Record<string, ToolPerformanceStats> {\n\tconst entries = Object.entries(stats);\n\tif (entries.length <= MAX_STATS_PER_HOST) return stats;\n\treturn Object.fromEntries(\n\t\tentries\n\t\t\t.sort(([, left], [, right]) => Date.parse(right.lastUsedAt) - Date.parse(left.lastUsedAt))\n\t\t\t.slice(0, MAX_STATS_PER_HOST),\n\t);\n}\n\nexport class ToolPerformanceStore {\n\tprivate readonly filePath: string;\n\tprivate readonly fingerprint: () => HostFingerprint;\n\n\tconstructor(filePath: string, options: { fingerprint?: () => HostFingerprint } = {}) {\n\t\tthis.filePath = filePath;\n\t\tthis.fingerprint = options.fingerprint ?? currentHostFingerprint;\n\t}\n\n\tstatic forAgentDir(agentDir: string, options: { fingerprint?: () => HostFingerprint } = {}): ToolPerformanceStore {\n\t\treturn new ToolPerformanceStore(join(agentDir, \"state\", \"tool-performance.json\"), options);\n\t}\n\n\tprivate load(): ToolPerformanceStoreFile {\n\t\ttry {\n\t\t\tif (!existsSync(this.filePath)) return { version: STORE_VERSION, hosts: {} };\n\t\t\treturn parseFile(JSON.parse(readFileSync(this.filePath, \"utf8\")));\n\t\t} catch {\n\t\t\treturn { version: STORE_VERSION, hosts: {} };\n\t\t}\n\t}\n\n\tprivate save(file: ToolPerformanceStoreFile): void {\n\t\tmkdirSync(dirname(this.filePath), { recursive: true });\n\t\twriteFileSync(this.filePath, `${JSON.stringify(file, null, \"\\t\")}\\n`, \"utf8\");\n\t}\n\n\tprivate hostData(file: ToolPerformanceStoreFile): HostToolPerformanceData {\n\t\tconst fingerprint = this.fingerprint();\n\t\tconst existing = file.hosts[fingerprint.id];\n\t\tif (existing) return existing;\n\t\tconst created: HostToolPerformanceData = { host: fingerprint, stats: {}, observations: [] };\n\t\tfile.hosts[fingerprint.id] = created;\n\t\treturn created;\n\t}\n\n\tget(key: ToolPerformanceKey): ToolPerformanceStats {\n\t\tconst file = this.load();\n\t\tconst host = file.hosts[this.fingerprint().id];\n\t\tconst stats = host?.stats[statKey(key)];\n\t\treturn stats ? { ...stats } : emptyStats(key, new Date(0).toISOString());\n\t}\n\n\trecordValidation(\n\t\tkey: ToolPerformanceKey,\n\t\toutcome: \"repaired\" | \"bounced\",\n\t\tat = new Date().toISOString(),\n\t): ToolPerformanceStats {\n\t\tconst file = this.load();\n\t\tconst host = this.hostData(file);\n\t\tconst storageKey = statKey(key);\n\t\tconst current = host.stats[storageKey] ?? emptyStats(key, at);\n\t\tconst next: ToolPerformanceStats = {\n\t\t\t...current,\n\t\t\trepairCount: current.repairCount + (outcome === \"repaired\" ? 1 : 0),\n\t\t\tbounceCount: current.bounceCount + (outcome === \"bounced\" ? 1 : 0),\n\t\t\tlastUsedAt: at,\n\t\t};\n\t\thost.stats[storageKey] = next;\n\t\thost.stats = trimStats(host.stats);\n\t\tthis.save(file);\n\t\treturn { ...next };\n\t}\n\n\trecordExecution(observation: ToolExecutionObservation): ToolPerformanceStats {\n\t\tconst at = observation.at ?? new Date().toISOString();\n\t\tconst file = this.load();\n\t\tconst host = this.hostData(file);\n\t\tconst storageKey = statKey(observation.key);\n\t\tconst current = host.stats[storageKey] ?? emptyStats(observation.key, at);\n\t\tconst latencyMs = finiteNonNegative(observation.latencyMs);\n\t\tconst inputTokenEstimate = finiteNonNegative(observation.inputTokenEstimate);\n\t\tconst outputTokenEstimate = finiteNonNegative(observation.outputTokenEstimate);\n\t\tconst next: ToolPerformanceStats = {\n\t\t\t...current,\n\t\t\talpha: current.alpha + (observation.success ? 1 : 0),\n\t\t\tbeta: current.beta + (observation.success ? 0 : 1),\n\t\t\tsampleCount: current.sampleCount + 1,\n\t\t\tlatencyEwmaMs: updateEwma(current.latencyEwmaMs, latencyMs),\n\t\t\tlatencyDeviationEwmaMs: updateDeviation(current.latencyEwmaMs, current.latencyDeviationEwmaMs, latencyMs),\n\t\t\tinputTokenEstimateEwma: updateEwma(current.inputTokenEstimateEwma, inputTokenEstimate),\n\t\t\toutputTokenEstimateEwma: updateEwma(current.outputTokenEstimateEwma, outputTokenEstimate),\n\t\t\tfailureCount: current.failureCount + (observation.success ? 0 : 1),\n\t\t\tlastUsedAt: at,\n\t\t};\n\t\thost.stats[storageKey] = next;\n\t\thost.stats = trimStats(host.stats);\n\t\thost.observations.push({\n\t\t\t...observation.selection,\n\t\t\tat,\n\t\t\tmodelRef: observation.key.modelRef,\n\t\t\tintentClass: observation.key.intentClass,\n\t\t\tactualTool: observation.key.tool,\n\t\t\tsucceeded: observation.success,\n\t\t\tranked: observation.selection.ranked.slice(0, 6),\n\t\t\tshortlist: observation.selection.shortlist.slice(0, 3),\n\t\t\tlatencyMs,\n\t\t\tinputTokenEstimate,\n\t\t\toutputTokenEstimate,\n\t\t});\n\t\tif (host.observations.length > MAX_OBSERVATIONS_PER_HOST) {\n\t\t\thost.observations = host.observations.slice(-MAX_OBSERVATIONS_PER_HOST);\n\t\t}\n\t\tthis.save(file);\n\t\treturn { ...next };\n\t}\n\n\tgetMetrics(modelRef?: string): ToolSelectionMetrics {\n\t\tconst host = this.load().hosts[this.fingerprint().id];\n\t\tconst observations = (host?.observations ?? []).filter(\n\t\t\t(observation) => modelRef === undefined || observation.modelRef === modelRef,\n\t\t);\n\t\tconst firstTools = observations.filter((observation) => observation.firstTool);\n\t\tconst recommended = observations.filter((observation) => observation.disposition === \"recommend\");\n\t\tconst average = (values: Array<number | undefined>): number | undefined => {\n\t\t\tconst present = values.filter((value): value is number => value !== undefined);\n\t\t\treturn present.length > 0 ? present.reduce((sum, value) => sum + value, 0) / present.length : undefined;\n\t\t};\n\t\treturn {\n\t\t\tfirstToolAttempts: firstTools.length,\n\t\t\tfirstToolSuccesses: firstTools.filter((observation) => observation.succeeded).length,\n\t\t\twrongToolOrFailureCount: observations.filter(\n\t\t\t\t(observation) =>\n\t\t\t\t\t!observation.succeeded ||\n\t\t\t\t\t(observation.disposition === \"recommend\" && observation.recommendation !== observation.actualTool),\n\t\t\t).length,\n\t\t\trecommendationCount: recommended.length,\n\t\t\trecommendationMatchedCount: recommended.filter(\n\t\t\t\t(observation) => observation.recommendation === observation.actualTool,\n\t\t\t).length,\n\t\t\tshortlistCount: observations.filter((observation) => observation.disposition === \"shortlist\").length,\n\t\t\tabstentionCount: observations.filter((observation) => observation.disposition === \"abstain\").length,\n\t\t\taverageLatencyMs: average(observations.map((observation) => observation.latencyMs)),\n\t\t\taverageInputTokenEstimate: average(observations.map((observation) => observation.inputTokenEstimate)),\n\t\t\taverageOutputTokenEstimate: average(observations.map((observation) => observation.outputTokenEstimate)),\n\t\t};\n\t}\n\n\tgetObservations(modelRef?: string): ToolSelectionObservation[] {\n\t\tconst host = this.load().hosts[this.fingerprint().id];\n\t\treturn (host?.observations ?? [])\n\t\t\t.filter((observation) => modelRef === undefined || observation.modelRef === modelRef)\n\t\t\t.map((observation) => ({\n\t\t\t\t...observation,\n\t\t\t\tshortlist: [...observation.shortlist],\n\t\t\t\tranked: observation.ranked.map((candidate) => ({ ...candidate })),\n\t\t\t}));\n\t}\n}\n"]}

package/dist/core/tool-selection/tool-performance-store.js

@@ -0,0 +1,249 @@
+import { existsSync, mkdirSync, readFileSync, writeFileSync } from "node:fs";
+import { dirname, join } from "node:path";
+import { currentHostFingerprint } from "../models/fitness-store.js";
+const STORE_VERSION = 1;
+const MAX_STATS_PER_HOST = 500;
+const MAX_OBSERVATIONS_PER_HOST = 1_000;
+const EWMA_ALPHA = 0.25;
+function statKey(key) {
+    return `${key.modelRef}\0${key.intentClass}\0${key.tool}`;
+}
+function isRecord(value) {
+    return typeof value === "object" && value !== null && !Array.isArray(value);
+}
+function isIntentClass(value) {
+    return (value === "read" ||
+        value === "search" ||
+        value === "execute" ||
+        value === "write" ||
+        value === "retrieve" ||
+        value === "explain" ||
+        value === "other");
+}
+function isStats(value) {
+    return (isRecord(value) &&
+        typeof value.modelRef === "string" &&
+        isIntentClass(value.intentClass) &&
+        typeof value.tool === "string" &&
+        typeof value.alpha === "number" &&
+        typeof value.beta === "number" &&
+        typeof value.sampleCount === "number" &&
+        typeof value.repairCount === "number" &&
+        typeof value.bounceCount === "number" &&
+        typeof value.failureCount === "number" &&
+        typeof value.lastUsedAt === "string");
+}
+function isObservation(value) {
+    return (isRecord(value) &&
+        typeof value.at === "string" &&
+        typeof value.modelRef === "string" &&
+        isIntentClass(value.intentClass) &&
+        typeof value.actualTool === "string" &&
+        typeof value.firstTool === "boolean" &&
+        typeof value.succeeded === "boolean" &&
+        (value.disposition === "recommend" || value.disposition === "shortlist" || value.disposition === "abstain") &&
+        Array.isArray(value.shortlist) &&
+        value.shortlist.every((tool) => typeof tool === "string") &&
+        typeof value.entropy === "number" &&
+        typeof value.margin === "number" &&
+        Array.isArray(value.ranked) &&
+        value.ranked.every((entry) => isRecord(entry) &&
+            typeof entry.tool === "string" &&
+            typeof entry.utility === "number" &&
+            typeof entry.probability === "number") &&
+        (value.latencyMs === undefined || typeof value.latencyMs === "number") &&
+        (value.inputTokenEstimate === undefined || typeof value.inputTokenEstimate === "number") &&
+        (value.outputTokenEstimate === undefined || typeof value.outputTokenEstimate === "number"));
+}
+function emptyStats(key, at) {
+    return {
+        ...key,
+        alpha: 1,
+        beta: 1,
+        sampleCount: 0,
+        repairCount: 0,
+        bounceCount: 0,
+        failureCount: 0,
+        lastUsedAt: at,
+    };
+}
+function finiteNonNegative(value) {
+    return typeof value === "number" && Number.isFinite(value) && value >= 0 ? value : undefined;
+}
+function updateEwma(previous, next) {
+    if (next === undefined)
+        return previous;
+    return previous === undefined ? next : previous + EWMA_ALPHA * (next - previous);
+}
+function updateDeviation(previousValue, previousDeviation, next) {
+    if (next === undefined)
+        return previousDeviation;
+    const deviation = previousValue === undefined ? 0 : Math.abs(next - previousValue);
+    return updateEwma(previousDeviation, deviation);
+}
+function parseHost(value) {
+    if (!isRecord(value) || !isRecord(value.host) || !isRecord(value.stats) || !Array.isArray(value.observations))
+        return undefined;
+    const host = value.host;
+    if (typeof host.id !== "string" ||
+        typeof host.cpu !== "string" ||
+        typeof host.cores !== "number" ||
+        typeof host.totalMemGb !== "number")
+        return undefined;
+    return {
+        host: { id: host.id, cpu: host.cpu, cores: host.cores, totalMemGb: host.totalMemGb },
+        stats: Object.fromEntries(Object.entries(value.stats).filter((entry) => isStats(entry[1]))),
+        observations: value.observations.filter(isObservation),
+    };
+}
+function parseFile(value) {
+    if (!isRecord(value) || value.version !== STORE_VERSION || !isRecord(value.hosts)) {
+        return { version: STORE_VERSION, hosts: {} };
+    }
+    return {
+        version: STORE_VERSION,
+        hosts: Object.fromEntries(Object.entries(value.hosts)
+            .map(([hostId, host]) => [hostId, parseHost(host)])
+            .filter((entry) => entry[1] !== undefined)),
+    };
+}
+function trimStats(stats) {
+    const entries = Object.entries(stats);
+    if (entries.length <= MAX_STATS_PER_HOST)
+        return stats;
+    return Object.fromEntries(entries
+        .sort(([, left], [, right]) => Date.parse(right.lastUsedAt) - Date.parse(left.lastUsedAt))
+        .slice(0, MAX_STATS_PER_HOST));
+}
+export class ToolPerformanceStore {
+    filePath;
+    fingerprint;
+    constructor(filePath, options = {}) {
+        this.filePath = filePath;
+        this.fingerprint = options.fingerprint ?? currentHostFingerprint;
+    }
+    static forAgentDir(agentDir, options = {}) {
+        return new ToolPerformanceStore(join(agentDir, "state", "tool-performance.json"), options);
+    }
+    load() {
+        try {
+            if (!existsSync(this.filePath))
+                return { version: STORE_VERSION, hosts: {} };
+            return parseFile(JSON.parse(readFileSync(this.filePath, "utf8")));
+        }
+        catch {
+            return { version: STORE_VERSION, hosts: {} };
+        }
+    }
+    save(file) {
+        mkdirSync(dirname(this.filePath), { recursive: true });
+        writeFileSync(this.filePath, `${JSON.stringify(file, null, "\t")}\n`, "utf8");
+    }
+    hostData(file) {
+        const fingerprint = this.fingerprint();
+        const existing = file.hosts[fingerprint.id];
+        if (existing)
+            return existing;
+        const created = { host: fingerprint, stats: {}, observations: [] };
+        file.hosts[fingerprint.id] = created;
+        return created;
+    }
+    get(key) {
+        const file = this.load();
+        const host = file.hosts[this.fingerprint().id];
+        const stats = host?.stats[statKey(key)];
+        return stats ? { ...stats } : emptyStats(key, new Date(0).toISOString());
+    }
+    recordValidation(key, outcome, at = new Date().toISOString()) {
+        const file = this.load();
+        const host = this.hostData(file);
+        const storageKey = statKey(key);
+        const current = host.stats[storageKey] ?? emptyStats(key, at);
+        const next = {
+            ...current,
+            repairCount: current.repairCount + (outcome === "repaired" ? 1 : 0),
+            bounceCount: current.bounceCount + (outcome === "bounced" ? 1 : 0),
+            lastUsedAt: at,
+        };
+        host.stats[storageKey] = next;
+        host.stats = trimStats(host.stats);
+        this.save(file);
+        return { ...next };
+    }
+    recordExecution(observation) {
+        const at = observation.at ?? new Date().toISOString();
+        const file = this.load();
+        const host = this.hostData(file);
+        const storageKey = statKey(observation.key);
+        const current = host.stats[storageKey] ?? emptyStats(observation.key, at);
+        const latencyMs = finiteNonNegative(observation.latencyMs);
+        const inputTokenEstimate = finiteNonNegative(observation.inputTokenEstimate);
+        const outputTokenEstimate = finiteNonNegative(observation.outputTokenEstimate);
+        const next = {
+            ...current,
+            alpha: current.alpha + (observation.success ? 1 : 0),
+            beta: current.beta + (observation.success ? 0 : 1),
+            sampleCount: current.sampleCount + 1,
+            latencyEwmaMs: updateEwma(current.latencyEwmaMs, latencyMs),
+            latencyDeviationEwmaMs: updateDeviation(current.latencyEwmaMs, current.latencyDeviationEwmaMs, latencyMs),
+            inputTokenEstimateEwma: updateEwma(current.inputTokenEstimateEwma, inputTokenEstimate),
+            outputTokenEstimateEwma: updateEwma(current.outputTokenEstimateEwma, outputTokenEstimate),
+            failureCount: current.failureCount + (observation.success ? 0 : 1),
+            lastUsedAt: at,
+        };
+        host.stats[storageKey] = next;
+        host.stats = trimStats(host.stats);
+        host.observations.push({
+            ...observation.selection,
+            at,
+            modelRef: observation.key.modelRef,
+            intentClass: observation.key.intentClass,
+            actualTool: observation.key.tool,
+            succeeded: observation.success,
+            ranked: observation.selection.ranked.slice(0, 6),
+            shortlist: observation.selection.shortlist.slice(0, 3),
+            latencyMs,
+            inputTokenEstimate,
+            outputTokenEstimate,
+        });
+        if (host.observations.length > MAX_OBSERVATIONS_PER_HOST) {
+            host.observations = host.observations.slice(-MAX_OBSERVATIONS_PER_HOST);
+        }
+        this.save(file);
+        return { ...next };
+    }
+    getMetrics(modelRef) {
+        const host = this.load().hosts[this.fingerprint().id];
+        const observations = (host?.observations ?? []).filter((observation) => modelRef === undefined || observation.modelRef === modelRef);
+        const firstTools = observations.filter((observation) => observation.firstTool);
+        const recommended = observations.filter((observation) => observation.disposition === "recommend");
+        const average = (values) => {
+            const present = values.filter((value) => value !== undefined);
+            return present.length > 0 ? present.reduce((sum, value) => sum + value, 0) / present.length : undefined;
+        };
+        return {
+            firstToolAttempts: firstTools.length,
+            firstToolSuccesses: firstTools.filter((observation) => observation.succeeded).length,
+            wrongToolOrFailureCount: observations.filter((observation) => !observation.succeeded ||
+                (observation.disposition === "recommend" && observation.recommendation !== observation.actualTool)).length,
+            recommendationCount: recommended.length,
+            recommendationMatchedCount: recommended.filter((observation) => observation.recommendation === observation.actualTool).length,
+            shortlistCount: observations.filter((observation) => observation.disposition === "shortlist").length,
+            abstentionCount: observations.filter((observation) => observation.disposition === "abstain").length,
+            averageLatencyMs: average(observations.map((observation) => observation.latencyMs)),
+            averageInputTokenEstimate: average(observations.map((observation) => observation.inputTokenEstimate)),
+            averageOutputTokenEstimate: average(observations.map((observation) => observation.outputTokenEstimate)),
+        };
+    }
+    getObservations(modelRef) {
+        const host = this.load().hosts[this.fingerprint().id];
+        return (host?.observations ?? [])
+            .filter((observation) => modelRef === undefined || observation.modelRef === modelRef)
+            .map((observation) => ({
+            ...observation,
+            shortlist: [...observation.shortlist],
+            ranked: observation.ranked.map((candidate) => ({ ...candidate })),
+        }));
+    }
+}
+//# sourceMappingURL=tool-performance-store.js.map