@hecer/yoke 1.20.0 → 1.22.0

package/dist/routing/router.js

@@ -1,9 +1,17 @@
+import { randomUUID } from 'node:crypto';
 import { buildWatchdogInvocation, makeRunner, runCapturedAgent, runnerInvocation, contextBlockFor, } from '../loop/runner.js';
 import { isAcceptanceCriterion, criterionCommandProblem } from '../loop/prd.js';
 import { historyForWorkers, projectHash, readRoutingObservations, recordRoutingObservation, storyHash } from './registry.js';
 import { assessmentInstructions, parseAssessment, tiers } from './assessment.js';
 import { chooseCapability, readAssessment, saveAssessment, routingAssessmentKey, knownInfrastructureFailure } from './capability.js';
 import { readPlanningFile } from './contracts.js';
+import { finishRoutingAttempt, markRoutingAttemptUsageIncomplete, readRoutingAttempts, recordRoutingAttemptUsage, reserveRoutingAttempt, routingEpisodeSummary } from './attempts.js';
+import { assessmentSignature } from './optimization.js';
+export function buildAssessmentPrompt(root, ctx) {
+    return [assessmentInstructions, 'Use the supplied task contract and project context to produce a bounded plan. Do not implement or change files.',
+        'Approved planning brief:', readPlanningFile(root, '.yoke/plan.md', 80_000) ?? '',
+        contextBlockFor(ctx.targetDir, ctx.story), JSON.stringify(ctx.story), 'Return exactly one line: YOKE_ASSESS {"taskClass":"implementation","difficulty":"medium","uncertainty":"low","risk":"low","scope":"low","testability":"high","reason":"evidence","approach":"steps and tests"}'].join('\n');
+}
 const costRank = { low: 0, medium: 1, high: 2 };
 export function rankWorkers(workers, strategy, maxCandidates) {
     const history = historyForWorkers(workers);
@@ -113,6 +121,7 @@ function callUsage(role, provider, selection, tokens, durationMs, profile) {
         usageAvailable: tokens !== undefined && tokens.measurementComplete !== false,
         ...(tokens?.reasoningOutputTokens !== undefined ? { reasoningOutputTokens: tokens.reasoningOutputTokens } : {}),
         ...(tokens?.totalCostUsd !== undefined ? { totalCostUsd: tokens.totalCostUsd } : {}),
+        costMeasurementComplete: tokens?.totalCostUsd !== undefined && tokens.costMeasurementComplete !== false,
         durationMs,
     };
 }
@@ -121,8 +130,17 @@ export function makeAdaptiveRunner(options) {
     return ctx => {
         const steps = run(ctx);
         let next = steps.next();
-        while (!next.done)
-            next = steps.next(next.value());
+        while (!next.done) {
+            let result;
+            try {
+                result = next.value();
+            }
+            catch (error) {
+                next = steps.throw(error);
+                continue;
+            }
+            next = steps.next(result);
+        }
         return next.value;
     };
 }
@@ -131,8 +149,17 @@ export function makeAsyncAdaptiveRunner(options) {
     return async (ctx) => {
         const steps = run(ctx);
         let next = steps.next();
-        while (!next.done)
-            next = steps.next(await next.value());
+        while (!next.done) {
+            let result;
+            try {
+                result = await next.value();
+            }
+            catch (error) {
+                next = steps.throw(error);
+                continue;
+            }
+            next = steps.next(result);
+        }
         return next.value;
     };
 }
@@ -142,26 +169,100 @@ function routingSteps(options) {
     const eligibleWorkers = options.workers.filter(worker => available(worker.agent));
     const failedStories = new Set();
     const makeWorker = options.makeWorker ?? ((agent, selection) => makeRunner(agent, options.idleTimeoutMs ?? 0, {
-        ...options.runnerOpts,
-        permissions: options.permissions ?? 'safe',
-        selection,
+        ...options.runnerOpts, permissions: options.permissions ?? 'safe', selection,
     }));
     return function* (ctx) {
-        const blocked = (summary) => ({ success: false, summary, routing: { blocked: true, recordOutcome: () => undefined } });
+        const root = options.projectRoot ?? ctx.targetDir;
+        const calls = [];
+        let reservation;
+        const routeStartedAt = new Date().toISOString();
+        const blocked = (summary) => ({ success: false, summary,
+            ...(calls.length ? { tokens: { ...aggregateCalls(calls), storyId: ctx.story.id, ...(reservation ? { routingAttemptId: reservation.id } : {}) } } : {}),
+            routing: { blocked: true, recordOutcome: () => undefined },
+        });
+        function* perform(role, provider, selection, work, profile, beforeInvoke) {
+            const request = { callId: randomUUID(), storyId: ctx.story.id, role, provider, selection: { ...selection } };
+            try {
+                if (!available(provider))
+                    return { blocked: `Configured ${role} provider is unavailable` };
+                const refusal = options.admitCall?.(request);
+                if (refusal)
+                    return { blocked: refusal };
+                beforeInvoke?.();
+            }
+            catch (error) {
+                return { blocked: `Call admission failed: ${error.message}` };
+            }
+            const started = now();
+            let result;
+            try {
+                result = yield work;
+            }
+            catch (error) {
+                const reported = error && typeof error === 'object' ? error.tokens : undefined;
+                result = { success: false, infrastructureFailure: true, summary: error?.message ?? String(error), output: '', ...(reported ? { tokens: reported } : {}) };
+            }
+            const durationMs = Math.max(0, now() - started);
+            const callId = result.tokens?.callId ?? (result.tokens?.calls?.length === 1 ? result.tokens.calls[0].callId : undefined) ?? request.callId;
+            const call = { ...callUsage(role, provider, selection, result.tokens, durationMs, profile), callId,
+                ...(reservation ? { routingAttemptId: reservation.id } : {}) };
+            calls.push(call);
+            const usage = { ...aggregateCalls([call]), callId, storyId: ctx.story.id,
+                ...(reservation ? { routingAttemptId: reservation.id } : {}) };
+            let accountingError;
+            try {
+                // Earlier planning calls acquire the same execution identity once a worker
+                // has actually been admitted. Stable call IDs make reporter joins idempotent.
+                if (reservation)
+                    recordRoutingAttemptUsage(root, reservation.id, { ...aggregateCalls(calls), routingAttemptId: reservation.id });
+            }
+            catch (error) {
+                accountingError = error;
+            }
+            // A failed optional join must never suppress delivery of already paid
+            // usage to the caller's own durable budget account.
+            try {
+                options.onCallUsage?.({ ...request, callId, durationMs, usage });
+            }
+            catch (error) {
+                accountingError ??= error;
+            }
+            if (accountingError) {
+                if (reservation) {
+                    try {
+                        markRoutingAttemptUsageIncomplete(root, reservation.id);
+                    }
+                    catch { /* unreadable state cannot provide a complete sample */ }
+                    try {
+                        finishRoutingAttempt(root, reservation, { verificationSuccess: false, failureKind: 'infrastructure', actualModel: result.tokens?.model });
+                    }
+                    catch { /* reservation remains charged and unresolved */ }
+                }
+                return { blocked: `Call accounting failed: ${accountingError.message}` };
+            }
+            return { result };
+        }
+        const capture = (provider, prompt, selection) => () => options.captureRoute
+            ? options.captureRoute(provider, ctx, prompt, selection)
+            : runCapturedAgent(provider, buildWatchdogInvocation(runnerInvocation(provider, prompt, ctx.targetDir, true, 'read-only', selection), options.idleTimeoutMs ?? 0));
+        const reserveWorker = (provider, selection, profile, limit) => {
+            reservation = reserveRoutingAttempt({ root, storyId: ctx.story.id, contractKey: routingAssessmentKey(root, ctx.story), limit,
+                provider, profile, selection, startedAt: routeStartedAt, accountingScope: options.accountingScope,
+                executionPolicyKey: options.executionPolicyKey });
+        };
         if (options.strategy === 'capability' && options.assessmentPolicy === 'prepared') {
             try {
-                const criteria = ctx.story.acceptance;
-                if (criteria.length < 2 || criteria.length > 5 || criteria.some(c => !isAcceptanceCriterion(c) || criterionCommandProblem(c)))
-                    return blocked('Prepared routing requires 2-5 executable acceptance criteria');
-                if (!readAssessment(options.projectRoot ?? ctx.targetDir, ctx.story, true))
-                    return blocked('Task assessment is missing or stale. Run yoke prd assess before execution.');
+                const criteria = ctx.story.acceptance, goal = options.contractKind === 'goal';
+                if (criteria.length < (goal ? 1 : 2) || (!goal && criteria.length > 5) || criteria.some(c => !isAcceptanceCriterion(c) || (!goal && criterionCommandProblem(c)) || (goal && (!c.verify.length || c.verify.some(command => !command.trim())))))
+                    return blocked(goal ? 'Prepared goal routing requires bound executable acceptance criteria' : 'Prepared routing requires 2-5 executable acceptance criteria');
+                if (!readAssessment(root, ctx.story, true))
+                    return blocked(`Task assessment is missing or stale. Run yoke ${goal ? 'goal' : 'prd'} assess before execution.`);
             }
             catch (error) {
                 return blocked(`Cannot read prepared assessment: ${error.message}`);
             }
         }
         if (options.strategy === 'capability' && !options.rules?.some(rule => (!rule.area || rule.area === ctx.story.area) && (!rule.storyId || rule.storyId === ctx.story.id))) {
-            const root = options.projectRoot ?? ctx.targetDir;
             let assessment;
             try {
                 assessment = readAssessment(root, ctx.story, options.assessmentPolicy === 'prepared');
@@ -169,79 +270,151 @@ function routingSteps(options) {
             catch (error) {
                 return blocked(`Cannot read assessment: ${error.message}`);
             }
-            let planning;
-            const calls = [];
             if (!assessment) {
-                const inputKey = routingAssessmentKey(root, ctx.story);
-                const prompt = [assessmentInstructions, 'Use the supplied task contract and project context to produce a bounded plan. Do not implement or change files.',
-                    'Approved planning brief:', readPlanningFile(root, '.yoke/plan.md', 80_000) ?? '',
-                    contextBlockFor(ctx.targetDir, ctx.story), JSON.stringify(ctx.story), 'Return exactly one line: YOKE_ASSESS {"taskClass":"implementation","difficulty":"medium","uncertainty":"low","risk":"low","scope":"low","testability":"high","reason":"evidence","approach":"steps and tests"}'].join('\n');
                 const planner = options.planner?.agent ?? options.parent;
-                if (!available(planner))
-                    return blocked('Configured planning provider is unavailable');
                 const selection = { ...(options.planner?.selection ?? options.parentSelection), nativeMultiAgent: false };
-                const started = now();
-                planning = yield () => options.captureRoute ? options.captureRoute(planner, ctx, prompt, selection)
-                    : runCapturedAgent(planner, buildWatchdogInvocation(runnerInvocation(planner, prompt, ctx.targetDir, true, 'read-only', selection), options.idleTimeoutMs ?? 0));
-                calls.push(callUsage('orchestrator', planner, selection, planning.tokens, now() - started));
-                if (routingAssessmentKey(root, ctx.story) !== inputKey)
-                    return { ...blocked('Planning inputs changed during assessment; retry planning with the current contract'), tokens: aggregateCalls(calls) };
-                assessment = planning.success ? parseAssessment(planning.output) : undefined;
-                if (assessment)
-                    saveAssessment(root, ctx.story, assessment, { provider: planner, model: planning.tokens?.model ?? selection.model });
+                let inputKey, prompt;
+                try {
+                    inputKey = routingAssessmentKey(root, ctx.story);
+                    prompt = buildAssessmentPrompt(root, ctx);
+                }
+                catch (error) {
+                    return blocked(`Cannot prepare assessment: ${error.message}`);
+                }
+                const planningCall = yield* perform('orchestrator', planner, selection, capture(planner, prompt, selection));
+                if (planningCall.blocked)
+                    return blocked(planningCall.blocked);
+                const planning = planningCall.result;
+                try {
+                    if (routingAssessmentKey(root, ctx.story) !== inputKey)
+                        return blocked('Planning inputs changed during assessment; retry planning with the current contract');
+                    assessment = planning.success ? parseAssessment(planning.output) : undefined;
+                    if (assessment)
+                        saveAssessment(root, ctx.story, assessment, { provider: planner, model: planning.tokens?.model ?? selection.model });
+                }
+                catch (error) {
+                    return blocked(`Cannot persist current assessment: ${error.message}`);
+                }
             }
             if (!assessment)
-                return { success: false, summary: 'Routing assessment unavailable or invalid; implementation was not started', tokens: aggregateCalls(calls), routing: { recordOutcome: () => undefined, blocked: true } };
-            const choice = chooseCapability({ root, story: ctx.story, assessment, workers: eligibleWorkers, parent: options.parent, parentSelection: options.parentSelection, maxAttempts: options.maxAttempts, fallback: options.fallback, maxTier: options.maxTier });
+                return blocked('Routing assessment unavailable or invalid; implementation was not started');
+            let choice;
+            try {
+                choice = chooseCapability({ root, story: ctx.story, assessment, workers: eligibleWorkers, parent: options.parent, parentSelection: options.parentSelection,
+                    maxAttempts: options.maxAttempts, fallback: options.fallback, maxTier: options.maxTier,
+                    optimization: options.optimization, executionPolicyKey: options.executionPolicyKey });
+            }
+            catch (error) {
+                return blocked(`Cannot admit routing attempt: ${error.message}`);
+            }
             options.onDecision?.(ctx.story.id, { profile: choice.worker?.id ?? 'SELF', provider: choice.provider, model: choice.selection.model, reasoningEffort: choice.selection.reasoningEffort, variant: choice.selection.variant, providerModel: choice.selection.provider, reason: choice.reason, next: choice.next, assessment });
             if (choice.blocked)
-                return { ...blocked(choice.reason), tokens: aggregateCalls(calls) };
+                return blocked(choice.reason);
             if (choice.exhausted)
-                return { success: false, summary: 'Routing attempt budget exhausted; replan this task before retrying', tokens: aggregateCalls(calls), routing: { recordOutcome: () => undefined, blocked: true } };
-            const started = now();
-            const result = yield () => makeWorker(choice.provider, choice.selection)({ ...ctx, attempt: choice.failures + 1, story: { ...ctx.story, assessment } });
-            calls.push(callUsage(choice.worker ? 'worker' : 'parent', choice.provider, choice.selection, result.tokens, now() - started, choice.worker?.id ?? 'SELF'));
+                return blocked('Routing attempt budget exhausted; replan this task before retrying');
+            let runner;
+            const implementation = yield* perform(choice.worker ? 'worker' : 'parent', choice.provider, choice.selection, () => runner({ ...ctx, attempt: reservation.ordinal, story: { ...ctx.story, assessment } }), choice.worker?.id ?? 'SELF', () => { runner = makeWorker(choice.provider, choice.selection); reserveWorker(choice.provider, choice.selection, choice.worker?.id ?? 'SELF', choice.attemptLimit); });
+            if (implementation.blocked)
+                return blocked(implementation.blocked);
+            const result = implementation.result;
             let recorded = false;
             const infrastructureFailure = result.infrastructureFailure || (!result.success && knownInfrastructureFailure(result.summary));
+            const recordOutcome = (verified, failureKind) => {
+                if (recorded)
+                    return;
+                const kind = infrastructureFailure ? 'infrastructure' : failureKind ?? 'implementation';
+                const measured = finishRoutingAttempt(root, reservation, { verificationSuccess: infrastructureFailure ? false : verified, failureKind: kind, actualModel: result.tokens?.model });
+                const episode = routingEpisodeSummary(root, reservation);
+                recorded = true;
+                recordRoutingObservation({ projectHash: projectHash(root), storyHash: storyHash(projectHash(root), ctx.story.id), assessmentKey: reservation.contractKey, taskClass: assessment.taskClass, requiredTier: choice.requiredTier,
+                    role: 'implementation', strategy: 'capability', selected: choice.worker?.id ?? 'SELF', provider: choice.provider,
+                    requestedProvider: choice.selection.provider, requestedModel: choice.selection.model, requestedReasoningEffort: choice.selection.reasoningEffort, requestedVariant: choice.selection.variant,
+                    actualModel: result.tokens?.model, orchestratorProvider: options.planner?.agent ?? options.parent, orchestratorModel: (options.planner?.selection ?? options.parentSelection)?.model,
+                    orchestratorDurationMs: calls.filter(c => c.role === 'orchestrator').reduce((sum, call) => sum + call.durationMs, 0), workerDurationMs: calls.at(-1).durationMs,
+                    processSuccess: result.success, verificationSuccess: infrastructureFailure ? false : verified, failureKind: kind,
+                    usageAvailable: measured.usageComplete, inputTokens: measured.inputTokens, outputTokens: measured.outputTokens,
+                    totalCostUsd: measured.totalCostUsd, costMeasurementComplete: measured.costComplete, accountingScope: measured.accountingScope,
+                    totalDurationMs: measured.durationMs, executionPolicyKey: options.executionPolicyKey,
+                    economicEpisode: { ...episode, assessmentSignature: assessmentSignature(assessment) },
+                });
+            };
+            if (infrastructureFailure)
+                recordOutcome(false, 'infrastructure');
             return { ...result, summary: `route=${choice.worker?.id ?? 'SELF'} (${choice.reason}); ${result.summary}`,
                 ...(infrastructureFailure ? { success: false, infrastructureFailure: true } : {}),
-                tokens: { ...aggregateCalls(calls), storyId: ctx.story.id, escalated: choice.failures > 1 },
-                routing: { blocked: infrastructureFailure || undefined, canRetry: !infrastructureFailure && choice.failures + 1 < (options.maxAttempts ?? 5), recordOutcome: (verified, failureKind) => {
-                        if (recorded)
-                            return;
-                        recorded = true;
-                        recordRoutingObservation({ projectHash: projectHash(root), storyHash: storyHash(projectHash(root), ctx.story.id), assessmentKey: routingAssessmentKey(root, ctx.story), taskClass: assessment.taskClass, requiredTier: choice.requiredTier,
-                            role: 'implementation', strategy: 'capability', selected: choice.worker?.id ?? 'SELF', provider: choice.provider, requestedProvider: choice.selection.provider, requestedModel: choice.selection.model, requestedReasoningEffort: choice.selection.reasoningEffort, requestedVariant: choice.selection.variant,
-                            actualModel: result.tokens?.model, orchestratorProvider: options.planner?.agent ?? options.parent, orchestratorModel: (options.planner?.selection ?? options.parentSelection)?.model, orchestratorDurationMs: calls.filter(c => c.role === 'orchestrator').reduce((s, c) => s + c.durationMs, 0), workerDurationMs: calls[calls.length - 1].durationMs,
-                            processSuccess: result.success, verificationSuccess: infrastructureFailure ? false : verified, failureKind: infrastructureFailure ? 'infrastructure' : failureKind ?? 'implementation', usageAvailable: result.tokens !== undefined && result.tokens.measurementComplete !== false,
-                            inputTokens: result.tokens?.inputTokens ?? 0, outputTokens: result.tokens?.outputTokens ?? 0, totalCostUsd: result.tokens?.totalCostUsd });
-                    } } };
+                tokens: { ...aggregateCalls(calls), storyId: ctx.story.id, routingAttemptId: reservation.id, escalated: choice.failures > 1 },
+                routing: { blocked: infrastructureFailure || undefined, canRetry: !infrastructureFailure && reservation.ordinal < choice.attemptLimit, recordOutcome } };
         }
-        // Re-rank per story so a long-running loop can use gate outcomes learned by
-        // earlier stories without rebuilding the runner.
         const rule = options.rules?.find(rule => (!rule.area || rule.area === ctx.story.area) && (!rule.storyId || rule.storyId === ctx.story.id) && (rule.area || rule.storyId));
+        const bounded = options.strategy === 'capability' || options.maxAttempts !== undefined;
+        let priorAttempts = [];
+        let failureKey = ctx.story.id;
+        if (bounded) {
+            try {
+                failureKey = routingAssessmentKey(root, ctx.story);
+                priorAttempts = readRoutingAttempts(root, ctx.story.id, failureKey);
+                if (priorAttempts.length >= (options.maxAttempts ?? 5))
+                    return blocked('Routing attempt budget exhausted; replan this task before retrying');
+            }
+            catch (error) {
+                return blocked(`Cannot read routing attempt state: ${error.message}`);
+            }
+        }
         if (rule) {
-            const project = projectHash(options.projectRoot ?? ctx.targetDir);
-            const prior = readRoutingObservations().reverse().find(event => event.projectHash === project && event.storyHash === storyHash(project, ctx.story.id) && typeof event.verificationSuccess === 'boolean');
-            if (prior?.verificationSuccess === false)
-                failedStories.add(ctx.story.id);
+            const project = projectHash(root);
+            try {
+                const prior = bounded ? priorAttempts.filter(entry => entry.outcome?.failureKind !== 'infrastructure').at(-1)?.outcome
+                    : readRoutingObservations().reverse().find(event => event.projectHash === project && event.storyHash === storyHash(project, ctx.story.id) && typeof event.verificationSuccess === 'boolean');
+                if (prior?.verificationSuccess === false)
+                    failedStories.add(failureKey);
+            }
+            catch (error) {
+                return blocked(`Cannot read routing attempt state: ${error.message}`);
+            }
         }
-        const ruleWorker = rule && failedStories.has(ctx.story.id) ? rule.escalateTo ?? 'SELF' : rule?.worker;
+        const ruleWorker = rule && failedStories.has(failureKey) ? rule.escalateTo ?? 'SELF' : rule?.worker;
         const candidates = rule ? eligibleWorkers : rankWorkers(eligibleWorkers, options.strategy, options.maxCandidates);
         if (candidates.length === 0) {
             if (options.fallback === 'block' || options.maxTier)
                 return blocked('No eligible routing profiles; parent fallback is disabled');
-            return yield () => makeWorker(options.parent, options.parentSelection ?? {})(ctx);
+            let runner;
+            const selection = options.parentSelection ?? {};
+            options.onDecision?.(ctx.story.id, { profile: 'SELF', provider: options.parent, model: selection.model, providerModel: selection.provider,
+                reasoningEffort: selection.reasoningEffort, variant: selection.variant, reason: 'No eligible profile; configured parent fallback', next: 'independent gate verification' });
+            const execution = yield* perform('parent', options.parent, selection, () => runner(ctx), 'SELF', () => {
+                runner = makeWorker(options.parent, selection);
+                if (bounded)
+                    reserveWorker(options.parent, selection, 'SELF', options.maxAttempts ?? 5);
+            });
+            if (execution.blocked)
+                return blocked(execution.blocked);
+            const result = execution.result;
+            const infrastructureFailure = result.infrastructureFailure || (!result.success && knownInfrastructureFailure(result.summary));
+            let recorded = false;
+            const recordOutcome = (verified, failureKind) => {
+                if (recorded)
+                    return;
+                const kind = infrastructureFailure ? 'infrastructure' : failureKind ?? 'implementation';
+                if (reservation)
+                    finishRoutingAttempt(root, reservation, { verificationSuccess: infrastructureFailure ? false : verified, failureKind: kind, actualModel: result.tokens?.model });
+                recorded = true;
+            };
+            if (infrastructureFailure)
+                recordOutcome(false, 'infrastructure');
+            return { ...result, ...(infrastructureFailure ? { success: false, infrastructureFailure: true } : {}),
+                tokens: { ...aggregateCalls(calls), storyId: ctx.story.id, ...(reservation ? { routingAttemptId: reservation.id } : {}) },
+                routing: { recordOutcome, ...(infrastructureFailure ? { blocked: true } : {}), ...(bounded ? { canRetry: !infrastructureFailure && reservation.ordinal < (options.maxAttempts ?? 5) } : {}) } };
         }
-        const prompt = buildRoutingPrompt(ctx, candidates, options.strategy);
         const orchestratorSelection = { ...(options.parentSelection ?? {}), ...(options.orchestratorSelection ?? {}), nativeMultiAgent: false };
-        const orchestratorStarted = now();
-        const routeRun = rule ? { success: true, summary: 'Explicit rule', output: '', tokens: { inputTokens: 0, outputTokens: 0 } } : yield () => options.captureRoute
-            ? options.captureRoute(options.parent, ctx, prompt, orchestratorSelection)
-            : runCapturedAgent(options.parent, buildWatchdogInvocation(runnerInvocation(options.parent, prompt, ctx.targetDir, true, 'read-only', orchestratorSelection), options.idleTimeoutMs ?? 0));
-        const orchestratorDurationMs = Math.max(0, now() - orchestratorStarted);
+        let routeRun = { success: true, summary: 'Explicit rule', output: '' };
+        if (!rule) {
+            const routed = yield* perform('orchestrator', options.parent, orchestratorSelection, capture(options.parent, buildRoutingPrompt(ctx, candidates, options.strategy), orchestratorSelection));
+            if (routed.blocked)
+                return blocked(routed.blocked);
+            routeRun = routed.result;
+        }
         const decision = rule
-            ? { worker: ruleWorker === 'SELF' || candidates.some(w => w.id === ruleWorker) ? ruleWorker : 'SELF', reason: failedStories.has(ctx.story.id) ? 'gate failure escalated by project rule' : 'explicit project routing rule' }
+            ? { worker: ruleWorker === 'SELF' || candidates.some(w => w.id === ruleWorker) ? ruleWorker : 'SELF', reason: failedStories.has(failureKey) ? 'gate failure escalated by project rule' : 'explicit project routing rule' }
             : routeRun.success ? parseRouteDecision(routeRun.output, candidates.map(worker => worker.id)) : null;
         const selected = decision?.worker ?? 'SELF';
         const worker = selected === 'SELF' ? undefined : candidates.find(candidate => candidate.id === selected);
@@ -251,75 +424,55 @@ function routingSteps(options) {
         const selection = worker
             ? { provider: worker.provider, model: worker.model, reasoningEffort: worker.reasoningEffort, variant: worker.variant, nativeMultiAgent: false, ...(provider !== 'gemini' && provider !== 'qwen' && provider !== 'pi' && provider !== 'hermes' ? { bare: options.parentSelection?.bare } : {}) }
             : { ...(options.parentSelection ?? {}), nativeMultiAgent: false };
-        const workerStarted = now();
-        const result = yield () => makeWorker(provider, selection)(ctx);
-        const workerDurationMs = Math.max(0, now() - workerStarted);
-        const calls = [
-            ...(!rule ? [callUsage('orchestrator', options.parent, orchestratorSelection, routeRun.tokens, orchestratorDurationMs)] : []),
-            callUsage(worker ? 'worker' : 'parent', provider, selection, result.tokens, workerDurationMs, selected),
-        ];
-        const tokens = {
-            storyId: ctx.story.id,
-            escalated: Boolean(rule && failedStories.has(ctx.story.id)),
-            inputTokens: (routeRun.tokens?.inputTokens ?? 0) + (result.tokens?.inputTokens ?? 0),
-            ...((routeRun.tokens?.cachedInputTokens !== undefined || result.tokens?.cachedInputTokens !== undefined)
-                ? { cachedInputTokens: (routeRun.tokens?.cachedInputTokens ?? 0) + (result.tokens?.cachedInputTokens ?? 0) }
-                : {}),
-            ...((routeRun.tokens?.cacheWriteInputTokens !== undefined || result.tokens?.cacheWriteInputTokens !== undefined)
-                ? { cacheWriteInputTokens: (routeRun.tokens?.cacheWriteInputTokens ?? 0) + (result.tokens?.cacheWriteInputTokens ?? 0) }
-                : {}),
-            outputTokens: (routeRun.tokens?.outputTokens ?? 0) + (result.tokens?.outputTokens ?? 0),
-            ...((routeRun.tokens?.reasoningOutputTokens !== undefined || result.tokens?.reasoningOutputTokens !== undefined)
-                ? { reasoningOutputTokens: (routeRun.tokens?.reasoningOutputTokens ?? 0) + (result.tokens?.reasoningOutputTokens ?? 0) }
-                : {}),
-            ...((routeRun.tokens?.totalCostUsd !== undefined || result.tokens?.totalCostUsd !== undefined)
-                ? { totalCostUsd: (routeRun.tokens?.totalCostUsd ?? 0) + (result.tokens?.totalCostUsd ?? 0) }
-                : {}),
-            ...(result.tokens?.model ? { model: result.tokens.model } : routeRun.tokens?.model ? { model: routeRun.tokens.model } : {}),
-            calls,
-            measurementComplete: calls.every(call => call.usageAvailable),
-            costMeasurementComplete: calls.every(call => call.totalCostUsd !== undefined),
-        };
+        options.onDecision?.(ctx.story.id, { profile: selected, provider, model: selection.model, providerModel: selection.provider,
+            reasoningEffort: selection.reasoningEffort, variant: selection.variant, reason: decision?.reason ?? 'Invalid controller response; configured parent fallback',
+            next: rule?.escalateTo ?? 'independent gate verification', ...(ctx.story.assessment ? { assessment: ctx.story.assessment } : {}) });
+        let runner;
+        const execution = yield* perform(worker ? 'worker' : 'parent', provider, selection, () => runner(ctx), selected, () => {
+            runner = makeWorker(provider, selection);
+            if (bounded)
+                reserveWorker(provider, selection, selected, options.maxAttempts ?? 5);
+        });
+        if (execution.blocked)
+            return blocked(execution.blocked);
+        const result = execution.result;
+        const tokens = { ...aggregateCalls(calls), storyId: ctx.story.id, escalated: Boolean(rule && failedStories.has(failureKey)), ...(reservation ? { routingAttemptId: reservation.id } : {}) };
         let recorded = false;
-        const recordOutcome = (verificationSuccess) => {
+        const recordOutcome = (verificationSuccess, failureKind) => {
             if (recorded)
                 return;
+            if (reservation)
+                finishRoutingAttempt(root, reservation, { verificationSuccess, failureKind, actualModel: result.tokens?.model });
             recorded = true;
-            if (!verificationSuccess)
-                failedStories.add(ctx.story.id);
-            else
-                failedStories.delete(ctx.story.id);
-            const project = projectHash(options.projectRoot ?? ctx.targetDir);
-            recordRoutingObservation({
-                projectHash: project,
-                storyHash: storyHash(project, ctx.story.id),
-                strategy: options.strategy,
-                selected,
-                provider,
-                ...(selection.model ? { requestedModel: selection.model } : {}),
-                ...(selection.reasoningEffort ? { requestedReasoningEffort: selection.reasoningEffort } : {}),
-                ...(selection.provider ? { requestedProvider: selection.provider } : {}),
-                ...(selection.variant ? { requestedVariant: selection.variant } : {}),
-                ...(result.tokens?.model ? { actualModel: result.tokens.model } : {}),
-                orchestratorProvider: options.parent,
-                ...(orchestratorSelection.model ? { orchestratorModel: orchestratorSelection.model } : {}),
-                orchestratorDurationMs,
-                workerDurationMs,
-                processSuccess: result.success,
-                verificationSuccess,
-                inputTokens: tokens.inputTokens,
-                outputTokens: tokens.outputTokens,
-            });
+            if (!verificationSuccess && failureKind !== 'infrastructure')
+                failedStories.add(failureKey);
+            else if (verificationSuccess)
+                failedStories.delete(failureKey);
+            const project = projectHash(root);
+            recordRoutingObservation({ projectHash: project, storyHash: storyHash(project, ctx.story.id), strategy: options.strategy, selected, provider,
+                requestedModel: selection.model, requestedReasoningEffort: selection.reasoningEffort, requestedProvider: selection.provider, requestedVariant: selection.variant,
+                actualModel: result.tokens?.model, orchestratorProvider: options.parent, orchestratorModel: orchestratorSelection.model,
+                orchestratorDurationMs: calls.filter(call => call.role === 'orchestrator').reduce((sum, call) => sum + call.durationMs, 0), workerDurationMs: calls.at(-1).durationMs,
+                processSuccess: result.success, verificationSuccess, failureKind,
+                inputTokens: tokens.inputTokens, outputTokens: tokens.outputTokens, usageAvailable: tokens.measurementComplete,
+                totalCostUsd: tokens.totalCostUsd, costMeasurementComplete: tokens.costMeasurementComplete, });
         };
-        const routeSummary = decision
-            ? `route=${selected} (${decision.reason})`
-            : `route=SELF (${routeRun.success ? 'invalid routing response' : 'orchestrator failed'})`;
-        return { ...result, summary: `${routeSummary}; ${result.summary}`, tokens, routing: { recordOutcome } };
+        if (result.infrastructureFailure)
+            recordOutcome(false, 'infrastructure');
+        const routeSummary = decision ? `route=${selected} (${decision.reason})` : `route=SELF (${routeRun.success ? 'invalid routing response' : 'orchestrator failed'})`;
+        return { ...result, summary: `${routeSummary}; ${result.summary}`, tokens, routing: { recordOutcome, ...(result.infrastructureFailure ? { blocked: true } : {}),
+                ...(bounded ? { canRetry: !result.infrastructureFailure && reservation.ordinal < (options.maxAttempts ?? 5) } : {}) } };
     };
 }
 function aggregateCalls(calls) {
-    return { inputTokens: calls.reduce((n, c) => n + c.inputTokens, 0), outputTokens: calls.reduce((n, c) => n + c.outputTokens, 0),
-        cachedInputTokens: calls.reduce((n, c) => n + (c.cachedInputTokens ?? 0), 0), cacheWriteInputTokens: calls.reduce((n, c) => n + (c.cacheWriteInputTokens ?? 0), 0),
-        ...(calls.some(c => c.totalCostUsd !== undefined) ? { totalCostUsd: calls.reduce((n, c) => n + (c.totalCostUsd ?? 0), 0) } : {}),
-        calls, measurementComplete: calls.every(c => c.usageAvailable), costMeasurementComplete: calls.every(c => c.totalCostUsd !== undefined) };
+    const optional = {};
+    for (const field of ['cachedInputTokens', 'cacheWriteInputTokens', 'reasoningOutputTokens', 'totalCostUsd']) {
+        if (calls.some(call => call[field] !== undefined))
+            optional[field] = calls.reduce((sum, call) => sum + (call[field] ?? 0), 0);
+    }
+    const model = calls.at(-1)?.actualModel;
+    return { inputTokens: calls.reduce((sum, call) => sum + call.inputTokens, 0), outputTokens: calls.reduce((sum, call) => sum + call.outputTokens, 0),
+        ...optional, ...(model ? { model } : {}), calls,
+        measurementComplete: calls.every(call => call.usageAvailable !== false),
+        costMeasurementComplete: calls.every(call => call.totalCostUsd !== undefined && call.costMeasurementComplete !== false) };
 }

package/dist/setup/command.js

@@ -46,7 +46,10 @@ export function defaultRoutingWorkers(agents) {
             { id: 'hermes-standard', agent: 'hermes', tier: 'standard', costTier: 'medium', capabilities: ['implementation'] },
         ],
     };
-    return agents.flatMap(agent => workers[agent]);
+    return agents.flatMap(agent => workers[agent]).map(worker => ({
+        ...worker,
+        profileMetadata: { version: 1, catalog: 'yoke-1.22.0', source: 'yoke-default', basis: 'configured-prior' },
+    }));
 }
 function parseAgents(value, fallback) {
     if (value.trim().toLowerCase() === 'all')
@@ -55,6 +58,8 @@ function parseAgents(value, fallback) {
     return parsed.length > 0 ? [...new Set(parsed)] : fallback;
 }
 function yes(value, fallback) {
+    if (!value)
+        return fallback;
     const normalized = value.trim().toLowerCase();
     if (['y', 'yes', 'j', 'ja', 'true', '1'].includes(normalized))
         return true;
@@ -99,6 +104,10 @@ export async function runSetup(targetDir, opts = {}) {
         let runner = defaultRunner;
         let decisionPolicy = defaultPolicy;
         let routing = defaultRouting;
+        let runnerModel = opts.runnerModel ?? existing?.runner?.model;
+        let runnerReasoning = opts.runnerReasoning ?? existing?.runner?.reasoningEffort;
+        const agentModels = { ...(opts.agentModels ?? {}) };
+        const agentReasoning = { ...(opts.agentReasoning ?? {}) };
         if (interactive && ask) {
             agents = parseAgents(await ask(`Agents [${defaultAgents.join(',')}] (${SUPPORTED_AGENTS.join(',')}|all): `), defaultAgents);
             const graphAnswer = (await ask(`Code graph [${defaultGraph}] (graphify|serena): `)).trim().toLowerCase();
@@ -112,12 +121,39 @@ export async function runSetup(targetDir, opts = {}) {
             if (policyAnswer === 'auto' || policyAnswer === 'critical')
                 decisionPolicy = policyAnswer;
             routing = yes(await ask(`Enable adaptive multi-model routing? [${defaultRouting ? 'yes' : 'no'}]: `), defaultRouting);
+            if (opts.configureModels) {
+                const configureModels = yes(await ask(`Configure model & reasoning effort per agent? [no]: `), false);
+                if (configureModels) {
+                    const rm = (await ask(`Runner model [${runnerModel ?? 'default'}]: `)).trim();
+                    if (rm)
+                        runnerModel = rm;
+                    const rr = (await ask(`Runner reasoning effort (none|low|medium|high|xhigh) [${runnerReasoning ?? 'default'}]: `)).trim();
+                    if (rr)
+                        runnerReasoning = rr;
+                    for (const a of agents) {
+                        const am = (await ask(`Model for agent "${a}" [${agentModels[a] ?? 'default'}]: `)).trim();
+                        if (am)
+                            agentModels[a] = am;
+                        const ar = (await ask(`Reasoning effort for agent "${a}" (none|low|medium|high|xhigh) [${agentReasoning[a] ?? 'default'}]: `)).trim();
+                        if (ar)
+                            agentReasoning[a] = ar;
+                    }
+                }
+            }
         }
         if (modelProviders.length && !agents.includes('qwen'))
             agents = [...agents, 'qwen'];
         if (!agents.includes(runner))
             agents = [...agents, runner];
-        const code = runRetrofit(targetDir, { loop, agents, codeGraph, codeIntelligence, host });
+        const priorRunner = existing?.runner?.agent ?? existing?.agents[0];
+        const runnerChanged = priorRunner !== undefined && priorRunner !== runner;
+        if (runnerChanged) {
+            if (opts.runnerModel === undefined)
+                runnerModel = undefined;
+            if (opts.runnerReasoning === undefined)
+                runnerReasoning = undefined;
+        }
+        const code = runRetrofit(targetDir, { loop, agents, codeGraph, codeIntelligence, host, cleanWorktrees: opts.cleanWorktrees, runnerModel, runnerReasoning, agentModels, agentReasoning });
         if (code !== 0)
             return code;
         applyActions(presetActions, targetDir, { backupDir: join(targetDir, '.yoke', 'backups', `model-presets-${Date.now()}`) });
@@ -125,14 +161,25 @@ export async function runSetup(targetDir, opts = {}) {
         if (!config)
             return 1;
         config.loop = { parallel: 'auto', isolate: true, ...config.loop, enabled: loop, decisionPolicy };
-        const priorRunner = existing?.runner?.agent ?? existing?.agents[0];
         const selectPresetModel = presetWorkers.length > 0 && runner === 'qwen' && (!existing || (priorRunner !== undefined && priorRunner !== runner));
         if (selectPresetModel && priorRunner !== 'qwen')
             config.runner = { permissions: config.runner?.permissions };
-        config.runner = { ...config.runner, agent: runner, ...(selectPresetModel && !config.runner?.model ? { model: presetWorkers[0].model } : {}) };
+        config.runner = {
+            ...config.runner,
+            agent: runner,
+            ...(runnerModel ? { model: runnerModel } : selectPresetModel && !config.runner?.model ? { model: presetWorkers[0].model } : {}),
+            ...(runnerReasoning ? { reasoningEffort: runnerReasoning } : {}),
+        };
         const existingWorkers = config.routing?.workers ?? [];
         const baseWorkers = existingWorkers.length > 0 && !opts.routingPreset ? existingWorkers : defaultRoutingWorkers(agents).filter(worker => !modelProviders.length || worker.agent !== 'qwen');
-        const workers = [...baseWorkers, ...presetWorkers.filter(worker => !baseWorkers.some(existing => existing.id === worker.id))];
+        const rawWorkers = [...baseWorkers, ...presetWorkers.filter(worker => !baseWorkers.some(existing => existing.id === worker.id))];
+        const workers = rawWorkers.map(w => {
+            const model = agentModels[w.agent] ?? w.model;
+            const reasoningEffort = agentReasoning[w.agent] ?? w.reasoningEffort;
+            return { ...w, ...(model ? { model } : {}), ...(reasoningEffort ? { reasoningEffort } : {}),
+                ...((agentModels[w.agent] || agentReasoning[w.agent]) && w.profileMetadata ? { profileMetadata: { ...w.profileMetadata, source: 'operator' } } : {}),
+            };
+        });
         config.routing = {
             ...config.routing,
             enabled: routing,
@@ -140,12 +187,13 @@ export async function runSetup(targetDir, opts = {}) {
             maxCandidates: config.routing?.maxCandidates ?? 3,
             assessmentPolicy: existing?.routing?.assessmentPolicy ?? (existing ? 'on-demand' : 'prepared'),
             fallback: existing?.routing?.fallback ?? (existing ? 'parent' : 'block'),
+            ...(!existing && !config.routing?.optimization ? { optimization: { version: 1, objective: 'balanced', minSamples: 20 } } : {}),
             ...(config.routing?.orchestrator ? { orchestrator: config.routing.orchestrator } : {}),
             workers,
         };
         YokeConfigSchema.parse(config);
         saveConfig(targetDir, config);
-        console.log(`Yoke setup complete: agents=${agents.join(',')} · runner=${runner} · loop=${loop ? 'on' : 'off'} · routing=${routing ? 'on' : 'off'} · decisions=${decisionPolicy}`);
+        console.log(`Yoke setup complete: agents=${agents.join(',')} · runner=${runner}${runnerModel ? ` (${runnerModel})` : ''}${runnerReasoning ? ` [${runnerReasoning}]` : ''} · loop=${loop ? 'on' : 'off'} · routing=${routing ? 'on' : 'off'} · decisions=${decisionPolicy}`);
         return 0;
     }
     finally {