vigthoria-cli 1.9.2 → 1.9.5

package/dist/commands/legion.js

@@ -51,6 +51,8 @@ const chalk_1 = __importDefault(require("chalk"));
 const fs = __importStar(require("fs"));
 const path = __importStar(require("path"));
 const readline = __importStar(require("readline/promises"));
+const node_child_process_1 = require("node:child_process");
+const node_util_1 = require("node:util");
 const logger_js_1 = require("../utils/logger.js");
 const api_js_1 = require("../utils/api.js");
 // Hyper Loop / Legion runs on the Vigthoria backend only. Local user installs
@@ -71,6 +73,10 @@ function buildServerHyperloopUrls() {
 const HYPERLOOP_URLS = (0, api_js_1.isServerRuntime)()
     ? buildServerHyperloopUrls()
     : (process.env.VIGTHORIA_HYPERLOOP_URL ? [process.env.VIGTHORIA_HYPERLOOP_URL] : []);
+const CORTEX_WARN_BUDGET_USD = 3.5;
+const CORTEX_HARD_BUDGET_USD = 5.0;
+const CORTEX_MAX_ROUNDS = 2;
+const execAsync = (0, node_util_1.promisify)(node_child_process_1.exec);
 class LegionCommand {
     config;
     logger;
@@ -145,8 +151,8 @@ class LegionCommand {
         });
     }
     async run(request, options) {
-        if (options.godmode) {
-            await this.runGodmode(request, options);
+        if (options.cortex) {
+            await this.runCortex(request, options);
             return;
         }
         if (options.workers) {
@@ -161,71 +167,213 @@ class LegionCommand {
             console.log(chalk_1.default.yellow('Usage: vigthoria legion "<task description>"'));
             console.log(chalk_1.default.gray('  --workers     List available Legion workers'));
             console.log(chalk_1.default.gray('  --status      Show Legion infrastructure status'));
-            console.log(chalk_1.default.gray('  --godmode     Run calculator + high-intelligence orchestration'));
+            console.log(chalk_1.default.gray('  --cortex      Vigthoria Cortex: maximum intelligence execution'));
             return;
         }
         await this.planAndExecute(request, options);
     }
-    async runGodmode(request, options) {
+    async runCortex(request, options) {
         if (!request) {
-            console.log(chalk_1.default.yellow('Usage: vigthoria legion --godmode "<task description>"'));
-            console.log(chalk_1.default.gray('  --plan-only      Run calculator only (no execution)'));
-            console.log(chalk_1.default.gray('  --approve        Skip confirmation prompt and execute'));
-            console.log(chalk_1.default.gray('  --auto-charge    Attempt direct VigCoin top-up when balance is low'));
-            console.log(chalk_1.default.gray('  --timeout <sec>  Abort remote execution if no result within timeout (default: 120)'));
+            console.log(chalk_1.default.yellow('Usage: vigthoria legion --cortex "<task description>"'));
+            console.log(chalk_1.default.gray('  --plan-only          Run calculator only (no execution)'));
+            console.log(chalk_1.default.gray('  --force-budget       Allow execution above hard safe-stop budget'));
+            console.log(chalk_1.default.gray('  --ignore-preflight   Bypass mandatory preflight checks (no warranty)'));
+            console.log(chalk_1.default.gray('  --speed              Speed mode: optional role skip on convergence'));
+            console.log(chalk_1.default.gray('  --tier heavy|lite    Model tier: heavy=strongest LLMs (default), lite=efficient+affordable'));
+            console.log(chalk_1.default.gray('  --repro-cmd <cmd>    Run a local reproducibility command before spend'));
+            console.log(chalk_1.default.gray('  --expect-repro-fail  Require repro command to fail before execution'));
+            console.log(chalk_1.default.gray('  --approve            Skip initial confirmation prompt and execute'));
+            console.log(chalk_1.default.gray('  --auto-charge        Attempt direct VigCoin top-up when balance is low'));
+            console.log(chalk_1.default.gray('  --timeout <sec>      Abort remote execution if no result within timeout (default: 120)'));
             return;
         }
         const workspace = options.project || process.cwd();
         const scan = this.scanProject(workspace);
-        const selectedModels = this.resolveModelProfiles(options.models);
-        const quote = this.buildRoleQuote(scan, selectedModels);
-        const billingQuote = this.buildBillingQuote(quote);
+        if (!options.ignorePreflight) {
+            const preflight = await this.runMandatoryPreflight(workspace, options.reproCmd, options.expectReproFail === true);
+            if (!preflight.ok) {
+                console.log(chalk_1.default.red('Cortex preflight failed.'));
+                console.log(chalk_1.default.red(`  ${preflight.reason}`));
+                console.log(chalk_1.default.yellow('Execution halted before any cloud spend. Use --ignore-preflight to override (no warranty).'));
+                return;
+            }
+            console.log(chalk_1.default.green('Preflight passed.'));
+        }
+        else {
+            console.log(chalk_1.default.yellow('Preflight bypassed by --ignore-preflight (no warranty).'));
+        }
+        const tier = (options.tier === 'lite') ? 'lite' : 'heavy';
+        const selectedModels = this.resolveModelProfiles(options.models, tier);
+        const quote = this.buildRoleQuote(scan, selectedModels, tier);
+        const billingQuote = this.buildBillingQuote(quote, tier);
+        if (billingQuote.retryAdjustedUsd > CORTEX_WARN_BUDGET_USD) {
+            console.log(chalk_1.default.yellow(`Estimated spend exceeds warning threshold ($${CORTEX_WARN_BUDGET_USD.toFixed(2)}).`));
+        }
+        if (billingQuote.retryAdjustedUsd > CORTEX_HARD_BUDGET_USD && !options.forceBudget) {
+            console.log(chalk_1.default.red(`Estimated spend exceeds hard budget ceiling ($${CORTEX_HARD_BUDGET_USD.toFixed(2)}).`));
+            console.log(chalk_1.default.yellow('Re-run with --force-budget to continue.'));
+            return;
+        }
         let billingGate = await this.evaluateBillingGate(billingQuote);
-        this.printGodmodeQuote(workspace, scan, quote, billingQuote, billingGate);
+        this.printCortexQuote(workspace, scan, quote, billingQuote, billingGate);
         if (options.planOnly) {
-            console.log(chalk_1.default.green('Godmode calculator complete (plan-only).'));
+            console.log(chalk_1.default.green('Cortex estimator complete (plan-only).'));
             return;
         }
-        if (!billingGate.canProceed) {
-            const resolved = await this.resolveBillingInsufficientFunds(billingQuote, billingGate, options);
-            if (!resolved) {
-                console.log(chalk_1.default.yellow('Godmode cancelled due to insufficient VigCoin balance.'));
+        const autoApprove = options.approve === true && options.noApprove !== true;
+        const approved = autoApprove ? true : await this.confirmExecution();
+        if (!approved) {
+            console.log(chalk_1.default.yellow('Cortex cancelled by user.'));
+            return;
+        }
+        let round = 1;
+        let cumulativeUsd = 0;
+        let currentQuote = billingQuote;
+        let lastFailure = null;
+        while (round <= CORTEX_MAX_ROUNDS) {
+            if (!billingGate.canProceed) {
+                const resolved = await this.resolveBillingInsufficientFunds(currentQuote, billingGate, options);
+                if (!resolved) {
+                    console.log(chalk_1.default.yellow('Cortex cancelled due to insufficient VigCoin balance.'));
+                    return;
+                }
+                billingGate = await this.evaluateBillingGate(currentQuote);
+                if (!billingGate.canProceed) {
+                    this.printBillingGateSummary(currentQuote, billingGate);
+                    console.log(chalk_1.default.red('Billing gate still blocked after charge attempt.'));
+                    return;
+                }
+            }
+            const charged = await this.collectExecutionCharge(currentQuote, billingGate);
+            if (!charged) {
+                console.log(chalk_1.default.yellow('Cortex cancelled because wallet charge was not completed.'));
                 return;
             }
-            billingGate = await this.evaluateBillingGate(billingQuote);
-            if (!billingGate.canProceed) {
-                this.printBillingGateSummary(billingQuote, billingGate);
-                console.log(chalk_1.default.red('Billing gate still blocked after charge attempt.'));
+            cumulativeUsd += currentQuote.finalUsd;
+            const enrichedRequest = [
+                '[CORTEX EXECUTION]',
+                request,
+                `Workspace: ${workspace}`,
+                `PhaseA: files=${scan.files}, lines=${scan.lines}, import_edges=${scan.importEdges}`,
+                'Role model assignment:',
+                ...quote.map((q) => `- ${q.role}: ${q.model} (requested_model=${q.requestedModel})`),
+                `Billing: round=${round}, cumulative_usd=${cumulativeUsd.toFixed(4)}, total_usd=${currentQuote.finalUsd.toFixed(4)}, plan=${billingGate.plan}, master_admin_free=${billingGate.masterAdminFree ? 'true' : 'false'}`,
+                `Execution mode: cloud-only core roles, speed_mode=${options.speed === true ? 'true' : 'false'}, tier=${tier}`,
+                lastFailure ? `Repair focus: previous failure at step=${lastFailure.failedStepId || 'unknown'} worker=${lastFailure.failedWorker || 'unknown'}` : '',
+            ].filter(Boolean).join('\\n');
+            const execution = await this.planAndExecute(enrichedRequest, options, {
+                workspace,
+                originalRequest: request,
+                scan,
+                quote,
+                tier,
+            });
+            if (execution.status === 'completed') {
                 return;
             }
+            const failedRole = String(execution.failedStepId || '').toLowerCase();
+            if (this.isCriticalRoleFailure(failedRole)) {
+                console.log(chalk_1.default.red(`Fail-fast: critical role '${failedRole || 'unknown'}' failed. Manual correction required.`));
+                return;
+            }
+            if (!this.isOptionalRepairRoleFailure(failedRole)) {
+                console.log(chalk_1.default.red('Execution failed and is not eligible for automatic optional-role repair.'));
+                return;
+            }
+            if (round >= CORTEX_MAX_ROUNDS) {
+                console.log(chalk_1.default.red('Optional-role auto-repair budget exhausted.'));
+                return;
+            }
+            const nextQuote = this.estimateAdditionalLoopQuote(currentQuote, execution);
+            const projectedTotal = cumulativeUsd + nextQuote.retryAdjustedUsd;
+            if (projectedTotal > CORTEX_HARD_BUDGET_USD && !options.forceBudget) {
+                console.log(chalk_1.default.red(`Additional loop would exceed hard budget ceiling ($${CORTEX_HARD_BUDGET_USD.toFixed(2)}).`));
+                console.log(chalk_1.default.yellow('Re-run with --force-budget to allow paid continuation.'));
+                return;
+            }
+            const continueApproved = await this.confirmAdditionalLoopCharge(nextQuote, round + 1, execution);
+            if (!continueApproved) {
+                console.log(chalk_1.default.yellow('Termination: user declined additional budget for next round.'));
+                return;
+            }
+            currentQuote = nextQuote;
+            billingGate = await this.evaluateBillingGate(currentQuote);
+            lastFailure = execution;
+            round += 1;
         }
-        const autoApprove = options.approve === true && options.noApprove !== true;
-        const approved = autoApprove ? true : await this.confirmExecution();
-        if (!approved) {
-            console.log(chalk_1.default.yellow('Godmode cancelled by user.'));
-            return;
+    }
+    isCriticalRoleFailure(role) {
+        return role === 'logic' || role === 'security';
+    }
+    isOptionalRepairRoleFailure(role) {
+        return role === 'performance' || role === 'edge_case';
+    }
+    estimateAdditionalLoopQuote(previousQuote, execution) {
+        const failedRole = String(execution.failedStepId || '').toLowerCase();
+        const factor = this.isOptionalRepairRoleFailure(failedRole) ? 0.35 : 0.55;
+        const baseUsd = Math.max(0.1, previousQuote.baseUsd * factor);
+        const finalUsd = Math.max(0.15, previousQuote.finalUsd * factor);
+        const retryAdjustedUsd = Math.max(0.15, previousQuote.retryAdjustedUsd * factor);
+        const rangeMinUsd = finalUsd;
+        const rangeMaxUsd = retryAdjustedUsd;
+        return {
+            baseUsd,
+            marginPct: previousQuote.marginPct,
+            finalUsd,
+            retryAdjustedUsd,
+            rangeMinUsd,
+            rangeMaxUsd,
+            tier: previousQuote.tier,
+            vigcoinRateUsd: previousQuote.vigcoinRateUsd,
+            vigcoinRequired: retryAdjustedUsd / previousQuote.vigcoinRateUsd,
+        };
+    }
+    async confirmAdditionalLoopCharge(nextQuote, nextRound, execution) {
+        if (!process.stdin.isTTY || !process.stdout.isTTY) {
+            return false;
         }
-        const charged = await this.collectExecutionCharge(billingQuote, billingGate);
-        if (!charged) {
-            console.log(chalk_1.default.yellow('Godmode cancelled because wallet charge was not completed.'));
-            return;
+        const rl = readline.createInterface({ input: process.stdin, output: process.stdout });
+        try {
+            console.log();
+            console.log(chalk_1.default.yellow('Budget depleted for current round. Additional paid loop required.'));
+            console.log(chalk_1.default.gray(`  Next round: ${nextRound}`));
+            console.log(chalk_1.default.gray(`  Failed step: ${execution.failedStepId || 'unknown'} (${execution.failedWorker || 'unknown'})`));
+            console.log(chalk_1.default.gray(`  Additional estimate: $${nextQuote.finalUsd.toFixed(4)} / ${nextQuote.vigcoinRequired.toFixed(3)} VIG`));
+            const answer = (await rl.question('Confirm additional VigCoin deduction and continue? (y/N): ')).trim().toLowerCase();
+            return answer === 'y' || answer === 'yes';
+        }
+        finally {
+            rl.close();
+        }
+    }
+    async runMandatoryPreflight(workspace, reproCmd, expectReproFail = false) {
+        try {
+            const stat = fs.statSync(workspace);
+            if (!stat.isDirectory()) {
+                return { ok: false, reason: `Workspace is not a directory: ${workspace}` };
+            }
+            fs.accessSync(workspace, fs.constants.R_OK | fs.constants.W_OK);
+        }
+        catch (err) {
+            return { ok: false, reason: `Workspace access check failed: ${err?.message || err}` };
+        }
+        if (!reproCmd || !reproCmd.trim()) {
+            return { ok: true, reason: 'Workspace checks passed (no repro command supplied).' };
+        }
+        try {
+            await execAsync(reproCmd, { cwd: workspace, timeout: 180000, maxBuffer: 1024 * 1024 * 8 });
+            if (expectReproFail) {
+                return { ok: false, reason: 'Repro command succeeded but --expect-repro-fail requires a failure signal.' };
+            }
+            return { ok: true, reason: 'Repro command succeeded.' };
+        }
+        catch (err) {
+            if (expectReproFail) {
+                return { ok: true, reason: 'Repro command failed as expected.' };
+            }
+            const stderr = String(err?.stderr || err?.message || err).slice(0, 300);
+            return { ok: false, reason: `Repro command failed: ${stderr}` };
         }
-        const enrichedRequest = [
-            '[GODMODE EXECUTION]',
-            request,
-            `Workspace: ${workspace}`,
-            `PhaseA: files=${scan.files}, lines=${scan.lines}, import_edges=${scan.importEdges}`,
-            'Role model assignment:',
-            ...quote.map((q) => `- ${q.role}: ${q.model} (requested_model=${q.requestedModel})`),
-            `Billing: plan=${billingGate.plan}, master_admin_free=${billingGate.masterAdminFree ? 'true' : 'false'}, vigcoin_required=${billingQuote.vigcoinRequired.toFixed(3)}, vigcoin_rate_usd=${billingQuote.vigcoinRateUsd.toFixed(4)}, total_usd=${billingQuote.finalUsd.toFixed(4)}`,
-            'Required flow: Detective repro -> 6-role parallel attack -> Architect synthesis -> test loop until pass.',
-        ].join('\n');
-        await this.planAndExecute(enrichedRequest, options, {
-            workspace,
-            originalRequest: request,
-            scan,
-            quote,
-        });
     }
     scanProject(workspace) {
         const files = [];
@@ -274,8 +422,9 @@ class LegionCommand {
         topFiles.sort((a, b) => (b.lines + b.imports * 25) - (a.lines + a.imports * 25));
         return { files: files.length, lines, importEdges, topFiles: topFiles.slice(0, 8) };
     }
-    resolveModelProfiles(modelsArg) {
-        const catalog = [
+    resolveModelProfiles(modelsArg, tier = 'heavy') {
+        // HEAVY tier: strongest available models — highest quality, highest cost.
+        const heavyCatalog = [
             { id: 'openrouter:openai/gpt-5.5', requestedModel: 'gpt-5.5', provider: 'openrouter', estInputPer1M: 5.0, estOutputPer1M: 30.0, capability: { reasoning: 10, coding: 10, security: 9, speed: 6, synthesis: 10 } },
             { id: 'openrouter:anthropic/claude-opus-4.7', requestedModel: 'opus-4.7', provider: 'openrouter', estInputPer1M: 5.0, estOutputPer1M: 25.0, capability: { reasoning: 10, coding: 9, security: 10, speed: 6, synthesis: 10 } },
             { id: 'openrouter:openai/o3', requestedModel: 'o3', provider: 'openrouter', estInputPer1M: 2.0, estOutputPer1M: 8.0, capability: { reasoning: 10, coding: 9, security: 9, speed: 6, synthesis: 9 } },
@@ -285,6 +434,24 @@ class LegionCommand {
             { id: 'openrouter:deepseek/deepseek-r1', requestedModel: 'cloud-reason', provider: 'openrouter', estInputPer1M: 0.7, estOutputPer1M: 2.5, capability: { reasoning: 9, coding: 8, security: 8, speed: 7, synthesis: 8 } },
             { id: 'openrouter:deepseek/deepseek-chat', requestedModel: 'cloud-pro', provider: 'openrouter', estInputPer1M: 0.32, estOutputPer1M: 0.89, capability: { reasoning: 7, coding: 8, security: 7, speed: 10, synthesis: 7 } },
         ];
+        // LITE tier: efficient models — high quality at ~5-10x lower cost than heavy.
+        // detective/architect → claude-sonnet-4-5 (best reasoning per dollar)
+        // security/reviewer   → claude-haiku-3-5  (strong safety, very cheap)
+        // logic               → o4-mini           (reasoning specialist, fraction of o3 cost)
+        // edge_case           → gemini-2.0-flash  (long context, fast, cheap)
+        // integration         → kimi-k2.5         (best codebase mapping at its price point)
+        // performance         → deepseek-v3       (coding analysis, ~20% of DeepSeek-V4-Pro cost)
+        const liteCatalog = [
+            { id: 'openrouter:anthropic/claude-sonnet-4-5', requestedModel: 'claude-sonnet-4-5', provider: 'openrouter', estInputPer1M: 3.0, estOutputPer1M: 15.0, capability: { reasoning: 9, coding: 9, security: 8, speed: 8, synthesis: 9 } },
+            { id: 'openrouter:anthropic/claude-haiku-3-5', requestedModel: 'claude-haiku-3-5', provider: 'openrouter', estInputPer1M: 0.8, estOutputPer1M: 4.0, capability: { reasoning: 8, coding: 8, security: 9, speed: 10, synthesis: 8 } },
+            { id: 'openrouter:openai/o4-mini', requestedModel: 'o4-mini', provider: 'openrouter', estInputPer1M: 1.1, estOutputPer1M: 4.4, capability: { reasoning: 9, coding: 8, security: 8, speed: 8, synthesis: 8 } },
+            { id: 'openrouter:google/gemini-2.0-flash', requestedModel: 'gemini-2.0-flash', provider: 'openrouter', estInputPer1M: 0.1, estOutputPer1M: 0.4, capability: { reasoning: 7, coding: 8, security: 7, speed: 10, synthesis: 7 } },
+            { id: 'openrouter:moonshotai/kimi-k2.5', requestedModel: 'kimi-k2.5', provider: 'openrouter', estInputPer1M: 0.44, estOutputPer1M: 2.0, capability: { reasoning: 9, coding: 9, security: 8, speed: 8, synthesis: 9 } },
+            { id: 'openrouter:deepseek/deepseek-chat-v3', requestedModel: 'deepseek-v3', provider: 'openrouter', estInputPer1M: 0.27, estOutputPer1M: 1.1, capability: { reasoning: 8, coding: 9, security: 7, speed: 10, synthesis: 8 } },
+            { id: 'openrouter:deepseek/deepseek-r1-distill-llama-70b', requestedModel: 'cloud-reason-lite', provider: 'openrouter', estInputPer1M: 0.23, estOutputPer1M: 0.69, capability: { reasoning: 8, coding: 7, security: 7, speed: 8, synthesis: 7 } },
+            { id: 'openrouter:deepseek/deepseek-chat', requestedModel: 'cloud-pro', provider: 'openrouter', estInputPer1M: 0.32, estOutputPer1M: 0.89, capability: { reasoning: 7, coding: 8, security: 7, speed: 10, synthesis: 7 } },
+        ];
+        const catalog = tier === 'lite' ? liteCatalog : heavyCatalog;
         if (!modelsArg || !modelsArg.trim())
             return catalog;
         const allow = new Set(modelsArg.split(',').map((m) => m.trim()).filter(Boolean));
@@ -293,7 +460,7 @@ class LegionCommand {
             || allow.has(m.requestedModel)));
         return filtered.length > 0 ? filtered : catalog;
     }
-    buildRoleQuote(scan, models) {
+    buildRoleQuote(scan, models, tier = 'heavy') {
         const roleWeights = {
             detective: { reasoning: 10, coding: 8, security: 6, speed: 3, synthesis: 7 },
             logic: { reasoning: 8, coding: 10, security: 3, speed: 6, synthesis: 6 },
@@ -304,19 +471,35 @@ class LegionCommand {
             reviewer: { reasoning: 9, coding: 7, security: 7, speed: 6, synthesis: 8 },
             architect: { reasoning: 10, coding: 9, security: 8, speed: 4, synthesis: 10 },
         };
+        // Preferred model per role for each tier.
+        const preferredByRoleHeavy = {
+            detective: 'gpt-5.5',
+            logic: 'o3',
+            security: 'opus-4.7',
+            performance: 'deepseek-v4-pro',
+            edge_case: 'gemini-2.5-pro',
+            integration: 'kimi-k2.5',
+            reviewer: 'opus-4.7',
+            architect: 'gpt-5.5',
+        };
+        const preferredByRoleLite = {
+            detective: 'claude-sonnet-4-5',
+            logic: 'o4-mini',
+            security: 'claude-haiku-3-5',
+            performance: 'deepseek-v3',
+            edge_case: 'gemini-2.0-flash',
+            integration: 'kimi-k2.5',
+            reviewer: 'claude-haiku-3-5',
+            architect: 'claude-sonnet-4-5',
+        };
+        const preferredByRole = tier === 'lite' ? preferredByRoleLite : preferredByRoleHeavy;
+        // Real-world token volumes from observed Cortex runs (not theoretical minimums).
+        // Base: 8k input tokens per role (context + project scan + prompt).
         const complexity = Math.max(1, Math.ceil((scan.lines / 4000) + (scan.importEdges / 200)));
+        const baseInputTokens = 8000 * complexity; // observed avg: 8k-25k per role
+        const baseOutputTokens = 2200 * complexity; // observed avg: 1.8k-3.6k per role
         return Object.keys(roleWeights).map((role) => {
             const w = roleWeights[role];
-            const preferredByRole = {
-                detective: 'gpt-5.5',
-                logic: 'o3',
-                security: 'opus-4.7',
-                performance: 'deepseek-v4-pro',
-                edge_case: 'gemini-2.5-pro',
-                integration: 'kimi-k2.5',
-                reviewer: 'opus-4.7',
-                architect: 'gpt-5.5',
-            };
             let best = models.find((m) => m.requestedModel === preferredByRole[role]) || models[0];
             if (!best)
                 best = models[0];
@@ -333,13 +516,13 @@ class LegionCommand {
                     }
                 }
             }
-            const estInputTokens = 1200 * complexity;
-            const estOutputTokens = 1800 * complexity;
+            const estInputTokens = baseInputTokens;
+            const estOutputTokens = baseOutputTokens;
             const estCostUsd = (estInputTokens / 1_000_000) * best.estInputPer1M + (estOutputTokens / 1_000_000) * best.estOutputPer1M;
             return { role, model: best.id, requestedModel: best.requestedModel, estInputTokens, estOutputTokens, estCostUsd };
         });
     }
-    buildGodmodeExplicitSteps(execution) {
+    buildCortexExplicitSteps(execution, speedMode = false) {
         const { originalRequest, workspace, scan, quote } = execution;
         const quoteByRole = new Map(quote.map((row) => [row.role, row]));
         const topFiles = scan.topFiles.slice(0, 5).map((file) => `${file.file} (${file.lines} lines, ${file.imports} imports)`);
@@ -356,26 +539,47 @@ class LegionCommand {
         const roleIterationBudget = {
             detective: 4,
             logic: 5,
-            security: 5,
-            performance: 4,
-            edge_case: 4,
-            integration: 5,
-            reviewer: 5,
+            security: speedMode ? 3 : 5,
+            performance: speedMode ? 3 : 5,
+            edge_case: speedMode ? 3 : 4,
+            integration: speedMode ? 3 : 6,
+            reviewer: speedMode ? 3 : 5,
             architect: 5,
         };
+        const optionalRoles = new Set(['security', 'performance', 'edge_case', 'integration', 'reviewer']);
         const steps = roleSequence.map(({ role, dependsOn }) => {
             const row = quoteByRole.get(role);
-            const requestedModel = row?.requestedModel || 'cloud-pro';
-            const model = row?.model || 'openrouter:deepseek/deepseek-chat';
+            const isOptionalRole = optionalRoles.has(role);
+            const requestedModel = role === 'detective'
+                ? 'gpt-5.5'
+                : role === 'architect'
+                    ? (scan.lines <= 3000 ? 'opus-4.7' : 'gpt-5.5')
+                    : (row?.requestedModel || 'cloud-pro');
+            const model = role === 'detective'
+                ? 'openrouter:openai/gpt-5.5'
+                : role === 'architect'
+                    ? (requestedModel === 'opus-4.7' ? 'openrouter:anthropic/claude-opus-4.7' : 'openrouter:openai/gpt-5.5')
+                    : (row?.model || 'openrouter:deepseek/deepseek-chat');
+            // Compact instruction packet: role tag + request digest + role micro-prompt.
+            // Full workspace, scan stats, and key files are already in the payload.
+            const ROLE_MICRO_PROMPTS = {
+                detective: 'Root-cause analysis. Identify failing paths, unknown risks, and systemic gaps. Output: numbered findings with confidence level.',
+                logic: 'Verify iteration bounds, fallback chains, state transitions, race conditions. Output: PASS/FAIL per subsystem with evidence.',
+                security: 'Audit for auth bypass, injection flaws, unsafe exec, secret exposure, SSRF. Verdict: SAFE/RISK/CRITICAL per finding.',
+                performance: 'Find N+1 queries, unbounded loops, sync I/O in async paths, memory leaks. Output: hotspot list with severity.',
+                edge_case: 'Probe boundary conditions, null/undefined paths, concurrent mutations, retry storms. Output: concrete failure scenarios.',
+                integration: 'Verify endpoint registration, auth middleware wiring, env var presence, cross-service contracts. PASS/FAIL per check.',
+                reviewer: 'Final gate review. Confirm contract coverage, flag regressions. Verdict: CONDITIONAL-GO or NO-GO with rationale.',
+                architect: 'Synthesize all role findings into a production-readiness verdict with a prioritized action list.',
+            };
+            const reqDigest = originalRequest.length > 220
+                ? originalRequest.slice(0, 220) + '\u2026'
+                : originalRequest;
             const roleObjective = [
-                `[GODMODE:${role.toUpperCase()}]`,
-                originalRequest,
-                `Workspace: ${workspace}`,
-                `Execution contract: force requested_model=${requestedModel} via ${model}; do not substitute a local-only model.`,
-                `Project scan: files=${scan.files}, lines=${scan.lines}, import_edges=${scan.importEdges}`,
-                topFiles.length > 0 ? `Key files: ${topFiles.join('; ')}` : 'Key files: unavailable',
-                `Execution budget: at most ${roleIterationBudget[role] || 3} reasoning iterations; prioritize a one-pass final answer and avoid redundant full-repo rescans.`,
-            ].join('\n');
+                `[CORTEX:${role.toUpperCase()}] model=${requestedModel} budget=${roleIterationBudget[role] || 3}iter`,
+                reqDigest,
+                ROLE_MICRO_PROMPTS[role] || 'Analyse the codebase and deliver your findings concisely.',
+            ].join('\\n');
             return {
                 step_id: role,
                 worker_name: 'v3_agent_worker',
@@ -383,7 +587,7 @@ class LegionCommand {
                 depends_on: dependsOn,
                 priority: role === 'architect' ? 2 : 3,
                 retry_policy: {
-                    max_attempts: 1,
+                    max_attempts: role === 'performance' || role === 'edge_case' ? 2 : 1,
                     strategy: 'repair',
                     requires_validation_failure: false,
                 },
@@ -391,14 +595,20 @@ class LegionCommand {
                     role,
                     requested_model: requestedModel,
                     quoted_model: model,
+                    enforce_cloud_only: role === 'detective' || role === 'architect',
+                    speed_mode: speedMode,
+                    optional_role: isOptionalRole,
+                    allow_convergence_skip: speedMode && isOptionalRole,
+                    convergence_sources: ['detective', 'logic'],
                     workspace,
                     top_files: topFiles,
                     max_iterations: roleIterationBudget[role] || 4,
                     compact_context: true,
-                    max_dependency_chars: role === 'architect' ? 18000 : 12000,
-                    max_artifacts: role === 'architect' ? 24 : 16,
-                    max_context_chars: role === 'architect' ? 240000 : 180000,
-                    max_output_tokens: role === 'detective' ? 3600 : role === 'architect' ? 3400 : role === 'reviewer' ? 3000 : 2400,
+                    context_sniper_max_depth: 2,
+                    max_dependency_chars: role === 'architect' ? 8000 : 4000,
+                    max_artifacts: role === 'architect' ? 12 : 8,
+                    max_context_chars: role === 'architect' ? 90000 : 60000,
+                    max_output_tokens: role === 'detective' ? 3600 : role === 'architect' ? 3400 : role === 'reviewer' ? 2800 : 2200,
                     request_timeout_seconds: role === 'detective' || role === 'architect' ? 1200 : 720,
                 },
             };
@@ -407,11 +617,11 @@ class LegionCommand {
             step_id: 'testing',
             worker_name: 'testing_worker',
             objective: [
-                '[GODMODE:TESTING]',
+                '[CORTEX:TESTING]',
                 originalRequest,
                 `Workspace: ${workspace}`,
-                'Validate the architect output, run the narrowest relevant checks, and report concrete failures if any remain.',
-            ].join('\n'),
+                'Validate architect output with narrow relevant checks and report concrete failures only.',
+            ].join('\\n'),
             depends_on: ['architect'],
             priority: 2,
             retry_policy: {
@@ -426,24 +636,45 @@ class LegionCommand {
         });
         return steps;
     }
-    buildBillingQuote(quote) {
+    buildBillingQuote(quote, tier = 'heavy') {
         const baseUsd = quote.reduce((sum, r) => sum + r.estCostUsd, 0);
-        const marginPctRaw = Number.parseFloat(String(process.env.VIGTHORIA_GODMODE_MARGIN_PCT || '10'));
+        const marginPctRaw = Number.parseFloat(String(process.env.VIGTHORIA_CORTEX_MARGIN_PCT || process.env.VIGTHORIA_GODMODE_MARGIN_PCT || '10'));
         const marginPct = Number.isFinite(marginPctRaw) ? Math.max(0, marginPctRaw) : 10;
         const finalUsd = baseUsd * (1 + (marginPct / 100));
+        // Retry-adjusted estimate: all 7 critical roles can trigger the quality-gate
+        // retry in state_manager.py (adds +2 iterations per degraded role).
+        // Probability of retry per critical role ≈ 45% (from production telemetry).
+        // RETRY_ITER_RATIO = additional spend when retry fires: +2 iters / avg 5 iters base ≈ 40%.
+        const CRITICAL_ROLES = new Set(['detective', 'logic', 'security', 'performance', 'integration', 'reviewer', 'architect']);
+        const RETRY_PROB = 0.45;
+        const RETRY_ITER_RATIO = 0.40;
+        const retryExtra = quote
+            .filter((r) => CRITICAL_ROLES.has(r.role))
+            .reduce((sum, r) => sum + r.estCostUsd * RETRY_PROB * RETRY_ITER_RATIO, 0);
+        const retryAdjustedUsd = (baseUsd + retryExtra) * (1 + (marginPct / 100));
+        // Range bounds.
+        const rangeMinUsd = finalUsd; // single pass, no retries
+        const maxRetryExtra = quote
+            .filter((r) => CRITICAL_ROLES.has(r.role))
+            .reduce((sum, r) => sum + r.estCostUsd * RETRY_ITER_RATIO, 0);
+        const rangeMaxUsd = (baseUsd + maxRetryExtra) * (1 + (marginPct / 100));
         const vigcoinRateRaw = Number.parseFloat(String(process.env.VIGTHORIA_VIGCOIN_USD_RATE || '1'));
         const vigcoinRateUsd = Number.isFinite(vigcoinRateRaw) && vigcoinRateRaw > 0 ? vigcoinRateRaw : 1;
-        const vigcoinRequired = finalUsd / vigcoinRateUsd;
+        const vigcoinRequired = retryAdjustedUsd / vigcoinRateUsd;
         return {
             baseUsd,
             marginPct,
             finalUsd,
+            retryAdjustedUsd,
+            rangeMinUsd,
+            rangeMaxUsd,
+            tier,
             vigcoinRateUsd,
             vigcoinRequired,
         };
     }
     async evaluateBillingGate(billingQuote) {
-        const forcedPlan = String(process.env.VIGTHORIA_GODMODE_FORCE_PLAN || '').trim().toLowerCase();
+        const forcedPlan = String(process.env.VIGTHORIA_CORTEX_FORCE_PLAN || process.env.VIGTHORIA_GODMODE_FORCE_PLAN || '').trim().toLowerCase();
         // On-server invocations using a service key run as trusted infrastructure.
         // No user wallet check is needed — cost is tracked at the service level.
         const hasServiceKey = !!(process.env.HYPERLOOP_SERVICE_KEY || process.env.V3_SERVICE_KEY);
@@ -460,7 +691,7 @@ class LegionCommand {
                 },
             };
         }
-        const entitlement = await this.fetchGodmodeEntitlement();
+        const entitlement = await this.fetchCortexEntitlement();
         const normalizedPlan = forcedPlan || entitlement.plan || this.config.getNormalizedPlan() || 'free';
         const masterAdminFree = this.isMasterAdminFree(normalizedPlan, entitlement.masterAccess, entitlement.isMasterAdmin);
         if (masterAdminFree) {
@@ -512,8 +743,8 @@ class LegionCommand {
         }
         return null;
     }
-    async fetchGodmodeEntitlement() {
-        if (process.env.VIGTHORIA_GODMODE_FORCE_MASTER_ACCESS === '1') {
+    async fetchCortexEntitlement() {
+        if (process.env.VIGTHORIA_CORTEX_FORCE_MASTER_ACCESS === '1' || process.env.VIGTHORIA_GODMODE_FORCE_MASTER_ACCESS === '1') {
             return { plan: this.config.getNormalizedPlan() || 'free', masterAccess: true, isMasterAdmin: true };
         }
         const baseUrl = this.getBillingBaseUrl();
@@ -568,7 +799,7 @@ class LegionCommand {
                 }
             }
             catch (err) {
-                this.logger.warn(this.formatLegionError(`Godmode entitlement request ${endpoint}`, err));
+                this.logger.warn(this.formatLegionError(`Cortex entitlement request ${endpoint}`, err));
                 continue;
             }
         }
@@ -649,8 +880,8 @@ class LegionCommand {
     async fetchWalletState() {
         const baseUrl = this.getBillingBaseUrl();
         const headers = this.getHeaders();
-        const forcedLow = process.env.VIGTHORIA_GODMODE_FORCE_LOW_CREDIT === '1';
-        const forcedBalanceRaw = process.env.VIGTHORIA_GODMODE_FORCE_BALANCE;
+        const forcedLow = process.env.VIGTHORIA_CORTEX_FORCE_LOW_CREDIT === '1' || process.env.VIGTHORIA_GODMODE_FORCE_LOW_CREDIT === '1';
+        const forcedBalanceRaw = process.env.VIGTHORIA_CORTEX_FORCE_BALANCE || process.env.VIGTHORIA_GODMODE_FORCE_BALANCE;
         if (forcedLow) {
             return {
                 available: true,
@@ -720,9 +951,9 @@ class LegionCommand {
         const headers = this.getHeaders();
         const amount = Math.max(1, Math.ceil(vigcoinNeeded));
         const chargePayloads = [
-            { endpoint: '/api/viagen6/vigcoin/charge', body: { amount, reason: 'godmode_legion' } },
-            { endpoint: '/api/wallet/charge', body: { amount, currency: 'VIGCOIN', reason: 'godmode_legion' } },
-            { endpoint: '/api/billing/topup', body: { vigcoin: amount, reason: 'godmode_legion' } },
+            { endpoint: '/api/viagen6/vigcoin/charge', body: { amount, reason: 'cortex_legion' } },
+            { endpoint: '/api/wallet/charge', body: { amount, currency: 'VIGCOIN', reason: 'cortex_legion' } },
+            { endpoint: '/api/billing/topup', body: { vigcoin: amount, reason: 'cortex_legion' } },
         ];
         for (const attempt of chargePayloads) {
             try {
@@ -766,7 +997,7 @@ class LegionCommand {
         if (gate.masterAdminFree) {
             return true;
         }
-        const spinner = (0, logger_js_1.createSpinner)('Charging VigCoin wallet for Godmode execution...').start();
+        const spinner = (0, logger_js_1.createSpinner)('Charging VigCoin wallet for Cortex execution...').start();
         const result = await this.attemptDirectCharge(billingQuote.vigcoinRequired);
         spinner.stop();
         if (!result.ok) {
@@ -774,7 +1005,7 @@ class LegionCommand {
             console.log(chalk_1.default.yellow(`Complete purchase first: ${result.checkoutUrl || `${this.getBillingBaseUrl()}/music/store#vigcoins`}`));
             return false;
         }
-        console.log(chalk_1.default.green('Wallet charged for Godmode execution.'));
+        console.log(chalk_1.default.green('Wallet charged for Cortex execution.'));
         return true;
     }
     async resolveBillingInsufficientFunds(billingQuote, gate, options) {
@@ -837,7 +1068,7 @@ class LegionCommand {
             console.log(chalk_1.default.green('    Free tier override applied (Master Admin).'));
             return;
         }
-        console.log(chalk_1.default.gray(`    Estimated total (USD): $${billingQuote.finalUsd.toFixed(4)}`));
+        console.log(chalk_1.default.gray(`    Estimated total (USD): $${billingQuote.retryAdjustedUsd.toFixed(4)}`) + chalk_1.default.gray(' (retry-adjusted expected)'));
         console.log(chalk_1.default.gray(`    VigCoin rate: 1 VIG = $${billingQuote.vigcoinRateUsd.toFixed(4)}`));
         console.log(chalk_1.default.gray(`    VigCoin required: ${billingQuote.vigcoinRequired.toFixed(3)}`));
         if (gate.wallet.vigcoinBalance !== null) {
@@ -851,12 +1082,15 @@ class LegionCommand {
             console.log(chalk_1.default.gray(`    Purchase URL: ${gate.wallet.purchaseUrl}`));
         }
     }
-    printGodmodeQuote(workspace, scan, quote, billingQuote, gate) {
-        const totalCost = billingQuote.finalUsd;
+    printCortexQuote(workspace, scan, quote, billingQuote, gate) {
+        const tierLabel = billingQuote.tier === 'lite'
+            ? chalk_1.default.cyan('LITE') + chalk_1.default.gray(' — efficient (claude-sonnet/haiku, o4-mini, gemini-flash, deepseek-v3)')
+            : chalk_1.default.magenta('HEAVY') + chalk_1.default.gray(' — strongest LLMs (gpt-5.5, opus-4.7, o3, gemini-2.5-pro)');
         console.log();
-        console.log(chalk_1.default.bold.white(`  ${logger_js_1.CH.hLine.repeat(3)} Legion Godmode Calculator ${logger_js_1.CH.hLine.repeat(31)}`));
+        console.log(chalk_1.default.bold.white(`  ${logger_js_1.CH.hLine.repeat(3)} Vigthoria Cortex Estimator ${logger_js_1.CH.hLine.repeat(31)}`));
         console.log();
         console.log(chalk_1.default.gray('  Workspace: ') + chalk_1.default.white(workspace));
+        console.log(chalk_1.default.gray('  Tier:      ') + tierLabel);
         console.log(chalk_1.default.gray('  Files scanned: ') + chalk_1.default.white(String(scan.files)));
         console.log(chalk_1.default.gray('  Lines scanned: ') + chalk_1.default.white(String(scan.lines)));
         console.log(chalk_1.default.gray('  Dependency edges: ') + chalk_1.default.white(String(scan.importEdges)));
@@ -873,7 +1107,11 @@ class LegionCommand {
             console.log(chalk_1.default.gray(`    ${logger_js_1.CH.bullet} ${row.role.padEnd(11)} ${row.model}  $${row.estCostUsd.toFixed(4)}`));
         }
         console.log();
-        console.log(chalk_1.default.yellow(`  Estimated total: $${totalCost.toFixed(4)}`));
+        console.log(chalk_1.default.yellow(`  Cost range (single-pass best case):  $${billingQuote.rangeMinUsd.toFixed(4)}`));
+        console.log(chalk_1.default.yellow(`  Cost range (expected with retries):   $${billingQuote.retryAdjustedUsd.toFixed(4)}`) + chalk_1.default.gray('  ← use this for budget planning'));
+        console.log(chalk_1.default.yellow(`  Cost range (worst case, all retries): $${billingQuote.rangeMaxUsd.toFixed(4)}`));
+        console.log(chalk_1.default.gray('  Retry model: 45% chance per critical role triggers quality-gate (+40% iterations per retry).'));
+        console.log(chalk_1.default.gray('  A mid-run checkpoint will appear when 70% of the expected estimate is consumed.'));
         console.log(chalk_1.default.gray('  Flow: Estimate -> Isolation -> Parallel Attack -> Synthesis'));
         console.log();
         this.printBillingGateSummary(billingQuote, gate);
@@ -886,7 +1124,7 @@ class LegionCommand {
         }
         const rl = readline.createInterface({ input: process.stdin, output: process.stdout });
         try {
-            const answer = (await rl.question('Proceed with Godmode execution? (y/N): ')).trim().toLowerCase();
+            const answer = (await rl.question('Proceed with Cortex execution? (y/N): ')).trim().toLowerCase();
             return answer === 'y' || answer === 'yes';
         }
         finally {
@@ -896,7 +1134,7 @@ class LegionCommand {
     /**
      * SSE streaming URL for the Legion execution endpoint.
      * Always hits Hyper Loop directly (port 8020) with the service key to avoid
-     * gateway JWT expiry killing long-running GodMode jobs.
+     * gateway JWT expiry killing long-running Cortex jobs.
      */
     getLegionStreamUrl() {
         const envOverride = String(process.env.VIGTHORIA_HYPERLOOP_URL || '').trim().replace(/\/$/, '');
@@ -910,12 +1148,16 @@ class LegionCommand {
             process.env.V3_SERVICE_KEY ||
             '');
     }
-    async planAndExecute(request, options, godmodeExecution) {
-        const explicitSteps = godmodeExecution ? this.buildGodmodeExplicitSteps(godmodeExecution) : undefined;
-        const workspace = godmodeExecution?.workspace || options.project || process.cwd();
+    async planAndExecute(request, options, cortexExecution) {
+        const explicitSteps = cortexExecution ? this.buildCortexExplicitSteps(cortexExecution, options.speed === true) : undefined;
+        const workspace = cortexExecution?.workspace || options.project || process.cwd();
         const body = {
             request,
-            context: { workspace },
+            context: {
+                workspace,
+                governor_budget_secs: 600,
+                governor_optional_roles: ['security', 'performance', 'edge_case', 'integration', 'reviewer'],
+            },
             constraints: {
                 active_only: true,
                 execution_timeout_seconds: options.timeoutSec,
@@ -930,7 +1172,6 @@ class LegionCommand {
             headers['X-Service-Key'] = serviceKey;
         }
         else {
-            // Fallback: pass user JWT (works if token is still valid)
             const token = this.config.get('authToken');
             if (token) {
                 headers['Authorization'] = `Bearer ${token}`;
@@ -941,7 +1182,6 @@ class LegionCommand {
         const startTime = Date.now();
         let response;
         try {
-            // No AbortSignal timeout — SSE keeps alive; server controls lifetime
             response = await fetch(streamUrl, {
                 method: 'POST',
                 headers,
@@ -951,20 +1191,19 @@ class LegionCommand {
         catch (connErr) {
             spinner.stop();
             this.logger.error(`Cannot connect to Hyper Loop at ${streamUrl}: ${connErr?.message || connErr}`);
-            return;
+            return { status: 'failed', plannedSteps: 0, completedSteps: 0 };
         }
         if (!response.ok) {
             spinner.stop();
             const errBody = await response.text().catch(() => '');
             this.logger.error(`Legion stream ${response.status}: ${(0, api_js_1.describeUpstreamStatus)(response.status)} — ${errBody.slice(0, 200)}`);
-            return;
+            return { status: 'failed', plannedSteps: 0, completedSteps: 0 };
         }
         if (!response.body) {
             spinner.stop();
             this.logger.error('Legion stream returned no response body');
-            return;
+            return { status: 'failed', plannedSteps: 0, completedSteps: 0 };
         }
-        // ── SSE consumer ─────────────────────────────────────────────
         spinner.stop();
         console.log();
         console.log(chalk_1.default.bold.white(`  ${logger_js_1.CH.hLine.repeat(3)} Legion Execution Report ${logger_js_1.CH.hLine.repeat(34)}`));
@@ -974,6 +1213,22 @@ class LegionCommand {
         let finalResult = null;
         let stepsTotal = 0;
         let stepsDone = 0;
+        let finalStatus = 'failed';
+        let failedStepId = '';
+        let failedWorker = '';
+        const streamEvents = [];
+        // Mid-run budget checkpoint: accumulate estimated spend per completed role.
+        const roleQuoteIndex = new Map((cortexExecution?.quote || []).map((q) => [q.role, q.estCostUsd]));
+        const marginPctMidrun = Number.isFinite(Number.parseFloat(String(process.env.VIGTHORIA_CORTEX_MARGIN_PCT || '10')))
+            ? Math.max(0, Number.parseFloat(String(process.env.VIGTHORIA_CORTEX_MARGIN_PCT || '10')))
+            : 10;
+        const baseEstimateUsd = (cortexExecution?.quote || []).reduce((s, q) => s + q.estCostUsd, 0);
+        const budgetCheckpointThreshold = cortexExecution
+            ? baseEstimateUsd * (1 + marginPctMidrun / 100) * 0.70
+            : Infinity;
+        let accumulatedEstUsd = 0;
+        let budgetCheckpointFired = false;
+        const completedRoleSummaries = [];
         try {
             const reader = response.body.getReader();
             while (true) {
@@ -982,11 +1237,11 @@ class LegionCommand {
                     break;
                 buffer += decoder.decode(value, { stream: true });
                 const lines = buffer.split('\n');
-                buffer = lines.pop() ?? ''; // keep incomplete last line
+                buffer = lines.pop() ?? '';
                 for (const line of lines) {
                     const trimmed = line.trim();
                     if (!trimmed || trimmed.startsWith(':'))
-                        continue; // keep-alive or comment
+                        continue;
                     if (!trimmed.startsWith('data:'))
                         continue;
                     const jsonStr = trimmed.slice(5).trim();
@@ -997,6 +1252,7 @@ class LegionCommand {
                     catch {
                         continue;
                     }
+                    streamEvents.push(evt);
                     switch (evt.event) {
                         case 'plan':
                             stepsTotal = evt.steps_total || 0;
@@ -1009,17 +1265,87 @@ class LegionCommand {
                         case 'step_complete': {
                             stepsDone = Number(evt.steps_done) || 0;
                             const icon = evt.status === 'completed' ? chalk_1.default.green(logger_js_1.CH.success) : chalk_1.default.red(logger_js_1.CH.error);
-                            const summary = evt.summary ? chalk_1.default.gray(` — ${String(evt.summary).slice(0, 120)}`) : '';
-                            console.log(`    ${icon} ${chalk_1.default.white(String(evt.step_id))} ${chalk_1.default.gray('[' + evt.worker + ']')}${summary}`);
+                            const stepSummaryRaw = String(evt.summary || '');
+                            const summarySnip = stepSummaryRaw ? chalk_1.default.gray(` — ${stepSummaryRaw.slice(0, 120)}`) : '';
+                            console.log(`    ${icon} ${chalk_1.default.white(String(evt.step_id))} ${chalk_1.default.gray('[' + evt.worker + ']')}${summarySnip}`);
+                            if (evt.status !== 'completed' && !failedStepId) {
+                                failedStepId = String(evt.step_id || '');
+                                failedWorker = String(evt.worker || '');
+                            }
+                            // Track spend and trigger mid-run budget checkpoint.
+                            const stepRole = String(evt.step_id || '');
+                            accumulatedEstUsd += roleQuoteIndex.get(stepRole) || 0;
+                            completedRoleSummaries.push({ role: stepRole, status: String(evt.status || ''), summary: stepSummaryRaw.slice(0, 200) });
+                            if (!budgetCheckpointFired
+                                && cortexExecution
+                                && accumulatedEstUsd >= budgetCheckpointThreshold
+                                && budgetCheckpointThreshold < Infinity
+                                && process.stdin.isTTY
+                                && process.stdout.isTTY) {
+                                budgetCheckpointFired = true;
+                                const remainingRoles = (cortexExecution.quote || [])
+                                    .filter((q) => !completedRoleSummaries.some((c) => c.role === q.role))
+                                    .map((q) => q.role);
+                                const remainingEstUsd = (cortexExecution.quote || [])
+                                    .filter((q) => remainingRoles.includes(q.role))
+                                    .reduce((s, q) => s + q.estCostUsd, 0);
+                                console.log();
+                                console.log(chalk_1.default.bold.yellow('  ━━━ Mid-Run Budget Checkpoint ━━━'));
+                                console.log(chalk_1.default.gray(`  Consumed so far (estimated):  $${accumulatedEstUsd.toFixed(4)}`));
+                                console.log(chalk_1.default.gray(`  Estimated remaining:          $${remainingEstUsd.toFixed(4)}`));
+                                console.log();
+                                console.log(chalk_1.default.white('  Roles completed:'));
+                                for (const c of completedRoleSummaries) {
+                                    const roleIcon = c.status === 'completed' ? chalk_1.default.green('✔') : chalk_1.default.red('✘');
+                                    const roleSumSnip = c.summary ? ` — ${c.summary.slice(0, 100)}` : '';
+                                    console.log(chalk_1.default.gray(`    ${roleIcon} ${c.role.padEnd(11)}${roleSumSnip}`));
+                                }
+                                console.log();
+                                console.log(chalk_1.default.white('  Remaining roles: ') + chalk_1.default.gray(remainingRoles.join(', ') || 'none'));
+                                console.log();
+                                const checkpointRl = readline.createInterface({ input: process.stdin, output: process.stdout });
+                                let continueRun = false;
+                                try {
+                                    const ans = (await checkpointRl.question('  Budget checkpoint — continue execution? (y/N): ')).trim().toLowerCase();
+                                    continueRun = ans === 'y' || ans === 'yes';
+                                }
+                                finally {
+                                    checkpointRl.close();
+                                }
+                                if (!continueRun) {
+                                    console.log(chalk_1.default.yellow('  Cortex stopped by user at budget checkpoint.'));
+                                    console.log(chalk_1.default.gray(`  State: ${stepsDone} of ${stepsTotal} steps completed.`));
+                                    finalStatus = 'failed';
+                                    failedStepId = failedStepId || 'user_checkpoint_stop';
+                                    // Break out of the for-of lines loop; the outer while(true) will end when reader.read() drains.
+                                    break;
+                                }
+                                console.log(chalk_1.default.green('  Continuing execution…'));
+                                console.log();
+                            }
                             break;
                         }
                         case 'complete':
-                            finalResult = evt.result || null;
+                            finalResult = evt.result || finalResult;
+                            if (Number.isFinite(Number(evt.steps_done))) {
+                                stepsDone = Number(evt.steps_done) || stepsDone;
+                            }
+                            if (Number.isFinite(Number(evt.steps_total))) {
+                                stepsTotal = Number(evt.steps_total) || stepsTotal;
+                            }
+                            if (!failedStepId && evt.failed_step_id) {
+                                failedStepId = String(evt.failed_step_id || '');
+                            }
+                            if (!failedWorker && evt.failed_worker) {
+                                failedWorker = String(evt.failed_worker || '');
+                            }
                             if (evt.status === 'completed') {
+                                finalStatus = 'completed';
                                 console.log();
                                 console.log(chalk_1.default.green(`  ${logger_js_1.CH.success} Legion completed successfully`));
                             }
                             else if (evt.status === 'failed') {
+                                finalStatus = 'failed';
                                 console.log();
                                 console.log(chalk_1.default.red(`  ${logger_js_1.CH.error} Legion execution failed`));
                                 if (evt.error)
@@ -1027,6 +1353,7 @@ class LegionCommand {
                             }
                             break;
                         case 'error':
+                            finalStatus = 'failed';
                             console.log(chalk_1.default.red(`  ${logger_js_1.CH.error} Stream error: ${String(evt.error).slice(0, 300)}`));
                             break;
                     }
@@ -1035,6 +1362,7 @@ class LegionCommand {
         }
         catch (streamErr) {
             this.logger.error(`Legion stream read error: ${streamErr?.message || streamErr}`);
+            finalStatus = 'failed';
         }
         const elapsedSec = ((Date.now() - startTime) / 1000).toFixed(1);
         if (stepsTotal > 0 && stepsDone < stepsTotal) {
@@ -1042,17 +1370,107 @@ class LegionCommand {
         }
         console.log();
         console.log(chalk_1.default.gray(`  Time: ${elapsedSec}s`));
-        // Show final output from architect/last step
-        if (finalResult) {
-            const lastStepResult = finalResult.final_output;
-            const summary = lastStepResult?.result?.summary || lastStepResult?.summary || '';
-            if (summary) {
-                console.log();
-                console.log(chalk_1.default.white('  Final output:'));
-                console.log(chalk_1.default.gray(`  ${String(summary).slice(0, 600)}`));
-            }
+        const lastStepResult = finalResult?.final_output;
+        const summary = String(lastStepResult?.result?.summary || lastStepResult?.summary || '');
+        if (summary) {
+            console.log();
+            console.log(chalk_1.default.white('  Final output:'));
+            console.log(chalk_1.default.gray(`  ${summary.slice(0, 600)}`));
+        }
+        if (cortexExecution) {
+            const report = this.buildCortexRunReport({
+                generatedAt: new Date().toISOString(),
+                status: finalStatus,
+                elapsedSeconds: Number(elapsedSec),
+                workspace: cortexExecution.workspace,
+                request: cortexExecution.originalRequest,
+                plannedSteps: stepsTotal,
+                completedSteps: stepsDone,
+                modifiedFiles: this.extractModifiedFiles(finalResult),
+                workers: [],
+                finalSummary: summary,
+            }, streamEvents);
+            const files = this.writeCortexSummaryReport(report);
+            console.log(chalk_1.default.gray('  Cortex summary: ' + files.markdownPath));
+            console.log(chalk_1.default.gray('  Cortex report JSON: ' + files.jsonPath));
         }
         console.log();
+        return {
+            status: finalStatus,
+            plannedSteps: stepsTotal,
+            completedSteps: stepsDone,
+            failedStepId: failedStepId || undefined,
+            failedWorker: failedWorker || undefined,
+            finalSummary: summary || undefined,
+        };
+    }
+    buildCortexRunReport(base, streamEvents) {
+        const workers = streamEvents
+            .filter((evt) => evt && evt.event === 'step_complete')
+            .map((evt) => ({
+            stepId: String(evt.step_id || ''),
+            worker: String(evt.worker || ''),
+            status: String(evt.status || ''),
+            summary: String(evt.summary || ''),
+        }));
+        return { ...base, workers };
+    }
+    extractModifiedFiles(finalResult) {
+        if (!finalResult || typeof finalResult !== 'object')
+            return [];
+        const candidates = [];
+        const finalOutput = (finalResult.final_output && typeof finalResult.final_output === 'object') ? finalResult.final_output : null;
+        const result = (finalOutput?.result && typeof finalOutput.result === 'object') ? finalOutput.result : null;
+        for (const key of ['modified_files', 'changed_files', 'files_changed', 'touched_files']) {
+            const top = finalResult[key];
+            const res = result ? result[key] : undefined;
+            if (Array.isArray(top))
+                candidates.push(...top.map((x) => String(x || '').trim()));
+            if (Array.isArray(res))
+                candidates.push(...res.map((x) => String(x || '').trim()));
+        }
+        return Array.from(new Set(candidates.filter((x) => x.length > 0)));
+    }
+    writeCortexSummaryReport(report) {
+        const reportsDir = path.join(report.workspace, '.vigthoria', 'reports');
+        fs.mkdirSync(reportsDir, { recursive: true });
+        const stamp = report.generatedAt.replace(/[:.]/g, '-');
+        const markdownPath = path.join(reportsDir, 'CORTEX_SUMMARY_' + stamp + '.md');
+        const jsonPath = path.join(reportsDir, 'CORTEX_SUMMARY_' + stamp + '.json');
+        const workerLines = report.workers.length > 0
+            ? report.workers.map((w) => '- ' + w.stepId + ' [' + w.worker + '] ' + w.status + (w.summary ? ': ' + w.summary : '')).join('\n')
+            : '- No worker step summaries captured';
+        const modifiedLines = report.modifiedFiles.length > 0
+            ? report.modifiedFiles.map((f) => '- ' + f).join('\n')
+            : '- No modified files reported by backend outputs';
+        const markdown = [
+            '# Cortex Run Summary',
+            '',
+            'Generated: ' + report.generatedAt,
+            'Status: ' + report.status,
+            'Elapsed: ' + report.elapsedSeconds + 's',
+            'Workspace: ' + report.workspace,
+            '',
+            '## Request',
+            report.request,
+            '',
+            '## Execution',
+            '- Planned steps: ' + report.plannedSteps,
+            '- Completed steps: ' + report.completedSteps,
+            '',
+            '## Worker Results',
+            workerLines,
+            '',
+            '## Modified Files',
+            modifiedLines,
+            '',
+            '## Final Summary',
+            report.finalSummary || 'No final summary reported by backend',
+            '',
+        ].join('\n');
+        fs.writeFileSync(markdownPath, markdown, 'utf8');
+        fs.writeFileSync(jsonPath, JSON.stringify(report, null, 2) + '\n', 'utf8');
+        return { markdownPath, jsonPath };
     }
     formatLegionError(context, err) {
         const message = err?.message || String(err || 'Unknown error');