@hongmaple0820/scale-engine 0.18.0 → 0.20.0

package/dist/context/ContextBudget.js

@@ -0,0 +1,322 @@
+import { existsSync, mkdirSync, readdirSync, readFileSync, statSync, writeFileSync } from 'node:fs';
+import { extname, isAbsolute, join, relative, resolve } from 'node:path';
+const CATEGORIES = ['always', 'on-demand', 'evidence', 'archive', 'generated'];
+const DEFAULT_MAX_ALWAYS_TOKENS = 2500;
+const DEFAULT_MAX_TASK_TOKENS = 8000;
+const IGNORED_DIRS = new Set([
+    '.git',
+    'node_modules',
+    'dist',
+    'coverage',
+    'tmp',
+    '.worktrees',
+]);
+const SCANNED_EXTENSIONS = new Set(['.md', '.json', '.yml', '.yaml', '.html', '.txt']);
+export function estimateTokens(content) {
+    if (content.length === 0)
+        return 0;
+    return Math.max(1, Math.ceil(content.length / 4));
+}
+export function scanContextBudget(options = {}) {
+    const projectDir = resolve(options.projectDir ?? process.cwd());
+    const scaleDir = options.scaleDir ?? '.scale';
+    const scaleRoot = resolveScaleRoot(projectDir, scaleDir);
+    const entries = discoverContextFiles(projectDir, scaleDir)
+        .map(file => contextEntry(projectDir, scaleRoot, file))
+        .sort((a, b) => categoryIndex(a.category) - categoryIndex(b.category) || b.estimatedTokens - a.estimatedTokens);
+    const summary = summarize(entries);
+    const thresholds = {
+        maxAlwaysTokens: options.maxAlwaysTokens ?? DEFAULT_MAX_ALWAYS_TOKENS,
+        maxTaskTokens: options.maxTaskTokens ?? DEFAULT_MAX_TASK_TOKENS,
+    };
+    return {
+        projectDir,
+        scaleDir,
+        generatedAt: new Date().toISOString(),
+        thresholds,
+        entries,
+        summary,
+        recommendations: recommendations(summary, thresholds, entries),
+    };
+}
+export function writeContextBudgetReport(report) {
+    const scaleRoot = resolveScaleRoot(report.projectDir, report.scaleDir);
+    mkdirSync(scaleRoot, { recursive: true });
+    const target = join(scaleRoot, 'context-budget.json');
+    writeFileSync(target, JSON.stringify(report, null, 2), 'utf-8');
+    return target;
+}
+export function doctorContextBudget(options = {}) {
+    const report = scanContextBudget(options);
+    const checks = [];
+    checks.push({
+        name: 'Always-loaded context',
+        status: report.summary.alwaysTokens <= report.thresholds.maxAlwaysTokens ? 'pass' : 'fail',
+        message: `Always-loaded context is ${report.summary.alwaysTokens} tokens; limit is ${report.thresholds.maxAlwaysTokens}.`,
+    });
+    checks.push({
+        name: 'Task context budget',
+        status: report.summary.totalTokens <= report.thresholds.maxTaskTokens ? 'pass' : 'warn',
+        message: `Scanned context is ${report.summary.totalTokens} tokens; task pack limit is ${report.thresholds.maxTaskTokens}.`,
+    });
+    checks.push({
+        name: 'Generated artifacts',
+        status: report.summary.byCategory.generated.files > 0 ? 'warn' : 'pass',
+        message: report.summary.byCategory.generated.files > 0
+            ? `${report.summary.byCategory.generated.files} generated artifacts should stay manifest-only unless explicitly requested.`
+            : 'No generated context artifacts detected.',
+    });
+    return {
+        ok: checks.every(check => check.status !== 'fail'),
+        checks,
+        report,
+    };
+}
+export function buildContextPack(options) {
+    const budget = options.budget ?? DEFAULT_MAX_TASK_TOKENS;
+    const report = scanContextBudget({ ...options, maxTaskTokens: budget });
+    const task = options.task;
+    const files = options.files ?? [];
+    const activations = activationRules(task, files, options.level ?? 'M');
+    const sections = [];
+    const alwaysEntries = report.entries.filter(entry => entry.category === 'always');
+    sections.push(section('always-core', 'always', true, 'Core agent entry rules and governance policy.', alwaysEntries));
+    for (const activation of activations) {
+        const activatedEntries = report.entries.filter(entry => activation.matches(entry));
+        if (activatedEntries.length === 0) {
+            sections.push({
+                id: activation.id,
+                category: activation.category,
+                included: false,
+                estimatedTokens: 0,
+                reason: `${activation.reason}; no matching artifacts found.`,
+                paths: [],
+            });
+            continue;
+        }
+        sections.push(section(activation.id, activation.category, true, activation.reason, activatedEntries));
+    }
+    const unique = dedupeSections(sections);
+    let total = 0;
+    const packed = [];
+    const omitted = [];
+    for (const item of unique) {
+        if (item.included && total + item.estimatedTokens > budget) {
+            packed.push({ ...item, included: false, reason: `${item.reason} Omitted because it would exceed budget ${budget}.` });
+            omitted.push({ id: item.id, reason: 'budget-exceeded', estimatedTokens: item.estimatedTokens });
+            continue;
+        }
+        if (item.included)
+            total += item.estimatedTokens;
+        packed.push(item);
+    }
+    return {
+        task: {
+            taskId: options.taskId,
+            task,
+            level: options.level ?? 'M',
+            files,
+            budget,
+        },
+        generatedAt: new Date().toISOString(),
+        totalEstimatedTokens: total,
+        lazyLoaded: packed
+            .filter(item => item.included && item.id !== 'always-core')
+            .map(item => ({ id: item.id, reason: item.reason })),
+        omitted,
+        sections: packed,
+    };
+}
+function discoverContextFiles(projectDir, scaleDir) {
+    const files = [];
+    const scaleRoot = resolveScaleRoot(projectDir, scaleDir);
+    const roots = [projectDir, join(projectDir, 'docs'), scaleRoot];
+    for (const root of roots) {
+        const absolute = root;
+        if (!existsSync(absolute))
+            continue;
+        if (statSync(absolute).isFile()) {
+            files.push(absolute);
+            continue;
+        }
+        walk(absolute, projectDir, files);
+    }
+    return Array.from(new Set(files));
+}
+function walk(dir, projectDir, files) {
+    const rel = normalizePath(relative(projectDir, dir));
+    const base = rel.split('/').pop() ?? '';
+    if (IGNORED_DIRS.has(base))
+        return;
+    for (const entry of readdirSync(dir, { withFileTypes: true })) {
+        const absolute = join(dir, entry.name);
+        if (entry.isDirectory()) {
+            if (!IGNORED_DIRS.has(entry.name))
+                walk(absolute, projectDir, files);
+            continue;
+        }
+        if (!entry.isFile())
+            continue;
+        const ext = extname(entry.name).toLowerCase();
+        const relativePath = normalizePath(relative(projectDir, absolute));
+        if (SCANNED_EXTENSIONS.has(ext) || ['AGENTS.md', 'CLAUDE.md', 'GEMINI.md', '.cursorrules'].includes(relativePath)) {
+            files.push(absolute);
+        }
+    }
+}
+function contextEntry(projectDir, scaleRoot, file) {
+    const content = safeRead(file);
+    const path = logicalContextPath(projectDir, scaleRoot, file);
+    const classification = classifyContextPath(path);
+    return {
+        path,
+        category: classification.category,
+        bytes: Buffer.byteLength(content, 'utf-8'),
+        estimatedTokens: estimateTokens(content),
+        reason: classification.reason,
+    };
+}
+export function classifyContextPath(path) {
+    const normalized = normalizePath(path);
+    const lower = normalized.toLowerCase();
+    if (['agents.md', 'claude.md', 'gemini.md', '.cursorrules'].includes(lower)) {
+        return { category: 'always', reason: 'agent entrypoint loaded at session start' };
+    }
+    if (/^\.scale\/(verification|skills|workflow|resource-policy|tool-policy|workspace|context-budget)\.json$/i.test(normalized)) {
+        return { category: 'always', reason: 'governance source-of-truth configuration' };
+    }
+    if (lower.includes('/evidence/') || lower.includes('/events/') || lower.startsWith('docs/worklog/tasks/')) {
+        return { category: 'evidence', reason: 'task or runtime evidence should be summarized and referenced' };
+    }
+    if (lower.endsWith('.html') || lower.includes('graphify-out/') || lower.includes('/resource-reports/')) {
+        return { category: 'generated', reason: 'generated artifact should stay manifest-only by default' };
+    }
+    if (lower.startsWith('docs/plans/') || lower.startsWith('docs/superpowers/') || lower.includes('roadmap') || lower.includes('optimization_plan') || lower.includes('changelog')) {
+        return { category: 'archive', reason: 'planning or historical context should be loaded only when requested' };
+    }
+    if (lower.includes('memory') || lower.includes('skill') || lower.includes('tool') || lower.includes('runtime') || lower.includes('resource')) {
+        return { category: 'on-demand', reason: 'domain-specific governance context loaded by trigger' };
+    }
+    return { category: 'on-demand', reason: 'project documentation loaded by task relevance' };
+}
+function summarize(entries) {
+    const byCategory = Object.fromEntries(CATEGORIES.map(category => [category, { files: 0, tokens: 0 }]));
+    for (const entry of entries) {
+        byCategory[entry.category].files += 1;
+        byCategory[entry.category].tokens += entry.estimatedTokens;
+    }
+    return {
+        totalFiles: entries.length,
+        totalTokens: entries.reduce((sum, entry) => sum + entry.estimatedTokens, 0),
+        byCategory,
+        alwaysTokens: byCategory.always.tokens,
+        onDemandTokens: byCategory['on-demand'].tokens,
+    };
+}
+function recommendations(summary, thresholds, entries) {
+    const output = [];
+    if (summary.alwaysTokens > thresholds.maxAlwaysTokens) {
+        output.push(`Reduce Always-loaded context from ${summary.alwaysTokens} to <= ${thresholds.maxAlwaysTokens} tokens.`);
+    }
+    const largest = entries.slice().sort((a, b) => b.estimatedTokens - a.estimatedTokens).slice(0, 3);
+    for (const entry of largest) {
+        if (entry.estimatedTokens > 1000 && entry.category !== 'archive') {
+            output.push(`Consider summarizing or demoting ${entry.path} (${entry.estimatedTokens} tokens).`);
+        }
+    }
+    if (summary.byCategory.generated.files > 0) {
+        output.push('Keep generated artifacts out of prompt context; load their manifests or summaries instead.');
+    }
+    return output;
+}
+function activationRules(task, files, level) {
+    const haystack = `${task} ${files.join(' ')} ${level}`.toLowerCase();
+    const rules = [
+        {
+            id: 'runtime-evidence',
+            category: 'evidence',
+            reason: 'Evidence is needed for completion and verification claims.',
+            matches: entry => entry.category === 'evidence',
+        },
+    ];
+    if (/skill|mcp|cli|browser|desktop|e2e|ui|ux|frontend/.test(haystack)) {
+        rules.push({
+            id: 'tool-and-skill-governance',
+            category: 'on-demand',
+            reason: 'Task mentions tool, skill, browser, desktop, E2E, or UI work.',
+            matches: entry => entry.category === 'on-demand' && /skill|tool|runtime|resource|engineering/i.test(entry.path),
+        });
+    }
+    if (/memory|knowledge|learn|contradiction|resume/.test(haystack)) {
+        rules.push({
+            id: 'memory-governance',
+            category: 'on-demand',
+            reason: 'Task needs project memory, knowledge recall, or contradiction handling.',
+            matches: entry => entry.category === 'on-demand' && /memory|knowledge/i.test(entry.path),
+        });
+    }
+    if (/plan|roadmap|architecture|review|critical|release/.test(haystack) || level.toUpperCase() === 'L' || level.toUpperCase() === 'CRITICAL') {
+        rules.push({
+            id: 'planning-archive',
+            category: 'archive',
+            reason: 'Task is high level or critical; archived plans may be relevant but remain budgeted.',
+            matches: entry => entry.category === 'archive',
+        });
+    }
+    return rules;
+}
+function section(id, category, included, reason, entries) {
+    return {
+        id,
+        category,
+        included,
+        estimatedTokens: entries.reduce((sum, entry) => sum + entry.estimatedTokens, 0),
+        reason,
+        paths: entries.map(entry => entry.path),
+    };
+}
+function dedupeSections(sections) {
+    const seenPaths = new Set();
+    return sections.map(section => {
+        const paths = section.paths.filter(path => {
+            if (seenPaths.has(path))
+                return false;
+            seenPaths.add(path);
+            return true;
+        });
+        return {
+            ...section,
+            paths,
+            estimatedTokens: paths.length === section.paths.length ? section.estimatedTokens : Math.ceil(section.estimatedTokens * (paths.length / Math.max(1, section.paths.length))),
+        };
+    }).filter(section => section.paths.length > 0 || section.id !== 'always-core');
+}
+function categoryIndex(category) {
+    return CATEGORIES.indexOf(category);
+}
+function safeRead(path) {
+    try {
+        return readFileSync(path, 'utf-8');
+    }
+    catch {
+        return '';
+    }
+}
+function normalizePath(path) {
+    return path.replace(/\\/g, '/');
+}
+function resolveScaleRoot(projectDir, scaleDir) {
+    return isAbsolute(scaleDir) ? scaleDir : resolve(projectDir, scaleDir);
+}
+function logicalContextPath(projectDir, scaleRoot, file) {
+    if (isWithin(projectDir, file))
+        return normalizePath(relative(projectDir, file));
+    if (isWithin(scaleRoot, file))
+        return normalizePath(join('.scale', relative(scaleRoot, file)));
+    return normalizePath(file);
+}
+function isWithin(root, target) {
+    const rel = relative(root, target);
+    return rel === '' || (!!rel && !rel.startsWith('..') && !isAbsolute(rel));
+}
+//# sourceMappingURL=ContextBudget.js.map

package/dist/context/ContextBudget.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"ContextBudget.js","sourceRoot":"","sources":["../../src/context/ContextBudget.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAE,SAAS,EAAE,WAAW,EAAE,YAAY,EAAE,QAAQ,EAAE,aAAa,EAAE,MAAM,SAAS,CAAA;AACnG,OAAO,EAAE,OAAO,EAAE,UAAU,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,MAAM,WAAW,CAAA;AA+DxE,MAAM,UAAU,GAA4B,CAAC,QAAQ,EAAE,WAAW,EAAE,UAAU,EAAE,SAAS,EAAE,WAAW,CAAC,CAAA;AAEvG,MAAM,yBAAyB,GAAG,IAAI,CAAA;AACtC,MAAM,uBAAuB,GAAG,IAAI,CAAA;AAEpC,MAAM,YAAY,GAAG,IAAI,GAAG,CAAC;IAC3B,MAAM;IACN,cAAc;IACd,MAAM;IACN,UAAU;IACV,KAAK;IACL,YAAY;CACb,CAAC,CAAA;AAEF,MAAM,kBAAkB,GAAG,IAAI,GAAG,CAAC,CAAC,KAAK,EAAE,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,OAAO,EAAE,MAAM,CAAC,CAAC,CAAA;AAiBtF,MAAM,UAAU,cAAc,CAAC,OAAe;IAC5C,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,CAAC,CAAA;IAClC,OAAO,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAA;AACnD,CAAC;AAED,MAAM,UAAU,iBAAiB,CAAC,UAAgC,EAAE;IAClE,MAAM,UAAU,GAAG,OAAO,CAAC,OAAO,CAAC,UAAU,IAAI,OAAO,CAAC,GAAG,EAAE,CAAC,CAAA;IAC/D,MAAM,QAAQ,GAAG,OAAO,CAAC,QAAQ,IAAI,QAAQ,CAAA;IAC7C,MAAM,SAAS,GAAG,gBAAgB,CAAC,UAAU,EAAE,QAAQ,CAAC,CAAA;IACxD,MAAM,OAAO,GAAG,oBAAoB,CAAC,UAAU,EAAE,QAAQ,CAAC;SACvD,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC,YAAY,CAAC,UAAU,EAAE,SAAS,EAAE,IAAI,CAAC,CAAC;SACtD,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,aAAa,CAAC,CAAC,CAAC,QAAQ,CAAC,GAAG,aAAa,CAAC,CAAC,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC,eAAe,GAAG,CAAC,CAAC,eAAe,CAAC,CAAA;IAEjH,MAAM,OAAO,GAAG,SAAS,CAAC,OAAO,CAAC,CAAA;IAClC,MAAM,UAAU,GAAG;QACjB,eAAe,EAAE,OAAO,CAAC,eAAe,IAAI,yBAAyB;QACrE,aAAa,EAAE,OAAO,CAAC,aAAa,IAAI,uBAAuB;KAChE,CAAA;IAED,OAAO;QACL,UAAU;QACV,QAAQ;QACR,WAAW,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;QACrC,UAAU;QACV,OAAO;QACP,OAAO;QACP,eAAe,EAAE,eAAe,CAAC,OAAO,EAAE,UAAU,EAAE,OAAO,CAAC;KAC/D,CAAA;AACH,CAAC;AAED,MAAM,UAAU,wBAAwB,CAAC,MAA2B;IAClE,MAAM,SAAS,GAAG,gBAAgB,CAAC,MAAM,CAAC,UAAU,EAAE,MAAM,CAAC,QAAQ,CAAC,CAAA;IACtE,SAAS,CAAC,SAAS,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAA;IACzC,MAAM,MAAM,GAAG,IAAI,CAAC,SAAS,EAAE,qBAAqB,CAAC,CAAA;IACrD,aAAa,CAAC,MAAM,EAAE,IAAI,CAAC,SAAS,CAAC,MAAM,EAAE,IAAI,EAAE,CAAC,CAAC,EAAE,OAAO,CAAC,CAAA;IAC/D,OAAO,MAAM,CAAA;AACf,CAAC;AAED,MAAM,UAAU,mBAAmB,CAAC,UAAgC,EAAE;IACpE,MAAM,MAAM,GAAG,iBAAiB,CAAC,OAAO,CAAC,CAAA;IACzC,MAAM,MAAM,GAAwC,EAAE,CAAA;IACtD,MAAM,CAAC,IAAI,CAAC;QACV,IAAI,EAAE,uBAAuB;QAC7B,MAAM,EAAE,MAAM,CAAC,OAAO,CAAC,YAAY,IAAI,MAAM,CAAC,UAAU,CAAC,eAAe,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM;QAC1F,OAAO,EAAE,4BAA4B,MAAM,CAAC,OAAO,CAAC,YAAY,qBAAqB,MAAM,CAAC,UAAU,CAAC,eAAe,GAAG;KAC1H,CAAC,CAAA;IACF,MAAM,CAAC,IAAI,CAAC;QACV,IAAI,EAAE,qBAAqB;QAC3B,MAAM,EAAE,MAAM,CAAC,OAAO,CAAC,WAAW,IAAI,MAAM,CAAC,UAAU,CAAC,aAAa,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM;QACvF,OAAO,EAAE,sBAAsB,MAAM,CAAC,OAAO,CAAC,WAAW,+BAA+B,MAAM,CAAC,UAAU,CAAC,aAAa,GAAG;KAC3H,CAAC,CAAA;IACF,MAAM,CAAC,IAAI,CAAC;QACV,IAAI,EAAE,qBAAqB;QAC3B,MAAM,EAAE,MAAM,CAAC,OAAO,CAAC,UAAU,CAAC,SAAS,CAAC,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM;QACvE,OAAO,EAAE,MAAM,CAAC,OAAO,CAAC,UAAU,CAAC,SAAS,CAAC,KAAK,GAAG,CAAC;YACpD,CAAC,CAAC,GAAG,MAAM,CAAC,OAAO,CAAC,UAAU,CAAC,SAAS,CAAC,KAAK,6EAA6E;YAC3H,CAAC,CAAC,0CAA0C;KAC/C,CAAC,CAAA;IACF,OAAO;QACL,EAAE,EAAE,MAAM,CAAC,KAAK,CAAC,KAAK,CAAC,EAAE,CAAC,KAAK,CAAC,MAAM,KAAK,MAAM,CAAC;QAClD,MAAM;QACN,MAAM;KACP,CAAA;AACH,CAAC;AAED,MAAM,UAAU,gBAAgB,CAAC,OAA2B;IAC1D,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,IAAI,uBAAuB,CAAA;IACxD,MAAM,MAAM,GAAG,iBAAiB,CAAC,EAAE,GAAG,OAAO,EAAE,aAAa,EAAE,MAAM,EAAE,CAAC,CAAA;IACvE,MAAM,IAAI,GAAG,OAAO,CAAC,IAAI,CAAA;IACzB,MAAM,KAAK,GAAG,OAAO,CAAC,KAAK,IAAI,EAAE,CAAA;IACjC,MAAM,WAAW,GAAG,eAAe,CAAC,IAAI,EAAE,KAAK,EAAE,OAAO,CAAC,KAAK,IAAI,GAAG,CAAC,CAAA;IACtE,MAAM,QAAQ,GAAyB,EAAE,CAAA;IAEzC,MAAM,aAAa,GAAG,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,EAAE,CAAC,KAAK,CAAC,QAAQ,KAAK,QAAQ,CAAC,CAAA;IACjF,QAAQ,CAAC,IAAI,CAAC,OAAO,CAAC,aAAa,EAAE,QAAQ,EAAE,IAAI,EAAE,+CAA+C,EAAE,aAAa,CAAC,CAAC,CAAA;IAErH,KAAK,MAAM,UAAU,IAAI,WAAW,EAAE,CAAC;QACrC,MAAM,gBAAgB,GAAG,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,EAAE,CAAC,UAAU,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,CAAA;QAClF,IAAI,gBAAgB,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YAClC,QAAQ,CAAC,IAAI,CAAC;gBACZ,EAAE,EAAE,UAAU,CAAC,EAAE;gBACjB,QAAQ,EAAE,UAAU,CAAC,QAAQ;gBAC7B,QAAQ,EAAE,KAAK;gBACf,eAAe,EAAE,CAAC;gBAClB,MAAM,EAAE,GAAG,UAAU,CAAC,MAAM,gCAAgC;gBAC5D,KAAK,EAAE,EAAE;aACV,CAAC,CAAA;YACF,SAAQ;QACV,CAAC;QACD,QAAQ,CAAC,IAAI,CAAC,OAAO,CAAC,UAAU,CAAC,EAAE,EAAE,UAAU,CAAC,QAAQ,EAAE,IAAI,EAAE,UAAU,CAAC,MAAM,EAAE,gBAAgB,CAAC,CAAC,CAAA;IACvG,CAAC;IAED,MAAM,MAAM,GAAG,cAAc,CAAC,QAAQ,CAAC,CAAA;IACvC,IAAI,KAAK,GAAG,CAAC,CAAA;IACb,MAAM,MAAM,GAAyB,EAAE,CAAA;IACvC,MAAM,OAAO,GAA2B,EAAE,CAAA;IAC1C,KAAK,MAAM,IAAI,IAAI,MAAM,EAAE,CAAC;QAC1B,IAAI,IAAI,CAAC,QAAQ,IAAI,KAAK,GAAG,IAAI,CAAC,eAAe,GAAG,MAAM,EAAE,CAAC;YAC3D,MAAM,CAAC,IAAI,CAAC,EAAE,GAAG,IAAI,EAAE,QAAQ,EAAE,KAAK,EAAE,MAAM,EAAE,GAAG,IAAI,CAAC,MAAM,2CAA2C,MAAM,GAAG,EAAE,CAAC,CAAA;YACrH,OAAO,CAAC,IAAI,CAAC,EAAE,EAAE,EAAE,IAAI,CAAC,EAAE,EAAE,MAAM,EAAE,iBAAiB,EAAE,eAAe,EAAE,IAAI,CAAC,eAAe,EAAE,CAAC,CAAA;YAC/F,SAAQ;QACV,CAAC;QACD,IAAI,IAAI,CAAC,QAAQ;YAAE,KAAK,IAAI,IAAI,CAAC,eAAe,CAAA;QAChD,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,CAAA;IACnB,CAAC;IAED,OAAO;QACL,IAAI,EAAE;YACJ,MAAM,EAAE,OAAO,CAAC,MAAM;YACtB,IAAI;YACJ,KAAK,EAAE,OAAO,CAAC,KAAK,IAAI,GAAG;YAC3B,KAAK;YACL,MAAM;SACP;QACD,WAAW,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;QACrC,oBAAoB,EAAE,KAAK;QAC3B,UAAU,EAAE,MAAM;aACf,MAAM,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,CAAC,QAAQ,IAAI,IAAI,CAAC,EAAE,KAAK,aAAa,CAAC;aAC1D,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,EAAE,EAAE,EAAE,IAAI,CAAC,EAAE,EAAE,MAAM,EAAE,IAAI,CAAC,MAAM,EAAE,CAAC,CAAC;QACtD,OAAO;QACP,QAAQ,EAAE,MAAM;KACjB,CAAA;AACH,CAAC;AAED,SAAS,oBAAoB,CAAC,UAAkB,EAAE,QAAgB;IAChE,MAAM,KAAK,GAAa,EAAE,CAAA;IAC1B,MAAM,SAAS,GAAG,gBAAgB,CAAC,UAAU,EAAE,QAAQ,CAAC,CAAA;IACxD,MAAM,KAAK,GAAG,CAAC,UAAU,EAAE,IAAI,CAAC,UAAU,EAAE,MAAM,CAAC,EAAE,SAAS,CAAC,CAAA;IAC/D,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;QACzB,MAAM,QAAQ,GAAG,IAAI,CAAA;QACrB,IAAI,CAAC,UAAU,CAAC,QAAQ,CAAC;YAAE,SAAQ;QACnC,IAAI,QAAQ,CAAC,QAAQ,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC;YAChC,KAAK,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAA;YACpB,SAAQ;QACV,CAAC;QACD,IAAI,CAAC,QAAQ,EAAE,UAAU,EAAE,KAAK,CAAC,CAAA;IACnC,CAAC;IACD,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,GAAG,CAAC,KAAK,CAAC,CAAC,CAAA;AACnC,CAAC;AAED,SAAS,IAAI,CAAC,GAAW,EAAE,UAAkB,EAAE,KAAe;IAC5D,MAAM,GAAG,GAAG,aAAa,CAAC,QAAQ,CAAC,UAAU,EAAE,GAAG,CAAC,CAAC,CAAA;IACpD,MAAM,IAAI,GAAG,GAAG,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,IAAI,EAAE,CAAA;IACvC,IAAI,YAAY,CAAC,GAAG,CAAC,IAAI,CAAC;QAAE,OAAM;IAClC,KAAK,MAAM,KAAK,IAAI,WAAW,CAAC,GAAG,EAAE,EAAE,aAAa,EAAE,IAAI,EAAE,CAAC,EAAE,CAAC;QAC9D,MAAM,QAAQ,GAAG,IAAI,CAAC,GAAG,EAAE,KAAK,CAAC,IAAI,CAAC,CAAA;QACtC,IAAI,KAAK,CAAC,WAAW,EAAE,EAAE,CAAC;YACxB,IAAI,CAAC,YAAY,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC;gBAAE,IAAI,CAAC,QAAQ,EAAE,UAAU,EAAE,KAAK,CAAC,CAAA;YACpE,SAAQ;QACV,CAAC;QACD,IAAI,CAAC,KAAK,CAAC,MAAM,EAAE;YAAE,SAAQ;QAC7B,MAAM,GAAG,GAAG,OAAO,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,WAAW,EAAE,CAAA;QAC7C,MAAM,YAAY,GAAG,aAAa,CAAC,QAAQ,CAAC,UAAU,EAAE,QAAQ,CAAC,CAAC,CAAA;QAClE,IAAI,kBAAkB,CAAC,GAAG,CAAC,GAAG,CAAC,IAAI,CAAC,WAAW,EAAE,WAAW,EAAE,WAAW,EAAE,cAAc,CAAC,CAAC,QAAQ,CAAC,YAAY,CAAC,EAAE,CAAC;YAClH,KAAK,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAA;QACtB,CAAC;IACH,CAAC;AACH,CAAC;AAED,SAAS,YAAY,CAAC,UAAkB,EAAE,SAAiB,EAAE,IAAY;IACvE,MAAM,OAAO,GAAG,QAAQ,CAAC,IAAI,CAAC,CAAA;IAC9B,MAAM,IAAI,GAAG,kBAAkB,CAAC,UAAU,EAAE,SAAS,EAAE,IAAI,CAAC,CAAA;IAC5D,MAAM,cAAc,GAAG,mBAAmB,CAAC,IAAI,CAAC,CAAA;IAChD,OAAO;QACL,IAAI;QACJ,QAAQ,EAAE,cAAc,CAAC,QAAQ;QACjC,KAAK,EAAE,MAAM,CAAC,UAAU,CAAC,OAAO,EAAE,OAAO,CAAC;QAC1C,eAAe,EAAE,cAAc,CAAC,OAAO,CAAC;QACxC,MAAM,EAAE,cAAc,CAAC,MAAM;KAC9B,CAAA;AACH,CAAC;AAED,MAAM,UAAU,mBAAmB,CAAC,IAAY;IAC9C,MAAM,UAAU,GAAG,aAAa,CAAC,IAAI,CAAC,CAAA;IACtC,MAAM,KAAK,GAAG,UAAU,CAAC,WAAW,EAAE,CAAA;IACtC,IAAI,CAAC,WAAW,EAAE,WAAW,EAAE,WAAW,EAAE,cAAc,CAAC,CAAC,QAAQ,CAAC,KAAK,CAAC,EAAE,CAAC;QAC5E,OAAO,EAAE,QAAQ,EAAE,QAAQ,EAAE,MAAM,EAAE,0CAA0C,EAAE,CAAA;IACnF,CAAC;IACD,IAAI,uGAAuG,CAAC,IAAI,CAAC,UAAU,CAAC,EAAE,CAAC;QAC7H,OAAO,EAAE,QAAQ,EAAE,QAAQ,EAAE,MAAM,EAAE,0CAA0C,EAAE,CAAA;IACnF,CAAC;IACD,IAAI,KAAK,CAAC,QAAQ,CAAC,YAAY,CAAC,IAAI,KAAK,CAAC,QAAQ,CAAC,UAAU,CAAC,IAAI,KAAK,CAAC,UAAU,CAAC,qBAAqB,CAAC,EAAE,CAAC;QAC1G,OAAO,EAAE,QAAQ,EAAE,UAAU,EAAE,MAAM,EAAE,8DAA8D,EAAE,CAAA;IACzG,CAAC;IACD,IAAI,KAAK,CAAC,QAAQ,CAAC,OAAO,CAAC,IAAI,KAAK,CAAC,QAAQ,CAAC,eAAe,CAAC,IAAI,KAAK,CAAC,QAAQ,CAAC,oBAAoB,CAAC,EAAE,CAAC;QACvG,OAAO,EAAE,QAAQ,EAAE,WAAW,EAAE,MAAM,EAAE,yDAAyD,EAAE,CAAA;IACrG,CAAC;IACD,IAAI,KAAK,CAAC,UAAU,CAAC,aAAa,CAAC,IAAI,KAAK,CAAC,UAAU,CAAC,mBAAmB,CAAC,IAAI,KAAK,CAAC,QAAQ,CAAC,SAAS,CAAC,IAAI,KAAK,CAAC,QAAQ,CAAC,mBAAmB,CAAC,IAAI,KAAK,CAAC,QAAQ,CAAC,WAAW,CAAC,EAAE,CAAC;QAChL,OAAO,EAAE,QAAQ,EAAE,SAAS,EAAE,MAAM,EAAE,qEAAqE,EAAE,CAAA;IAC/G,CAAC;IACD,IAAI,KAAK,CAAC,QAAQ,CAAC,QAAQ,CAAC,IAAI,KAAK,CAAC,QAAQ,CAAC,OAAO,CAAC,IAAI,KAAK,CAAC,QAAQ,CAAC,MAAM,CAAC,IAAI,KAAK,CAAC,QAAQ,CAAC,SAAS,CAAC,IAAI,KAAK,CAAC,QAAQ,CAAC,UAAU,CAAC,EAAE,CAAC;QAC7I,OAAO,EAAE,QAAQ,EAAE,WAAW,EAAE,MAAM,EAAE,sDAAsD,EAAE,CAAA;IAClG,CAAC;IACD,OAAO,EAAE,QAAQ,EAAE,WAAW,EAAE,MAAM,EAAE,gDAAgD,EAAE,CAAA;AAC5F,CAAC;AAED,SAAS,SAAS,CAAC,OAA6B;IAC9C,MAAM,UAAU,GAAG,MAAM,CAAC,WAAW,CAAC,UAAU,CAAC,GAAG,CAAC,QAAQ,CAAC,EAAE,CAAC,CAAC,QAAQ,EAAE,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,CAAC,CAAC,CAAuC,CAAA;IAC5I,KAAK,MAAM,KAAK,IAAI,OAAO,EAAE,CAAC;QAC5B,UAAU,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC,KAAK,IAAI,CAAC,CAAA;QACrC,UAAU,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC,MAAM,IAAI,KAAK,CAAC,eAAe,CAAA;IAC5D,CAAC;IACD,OAAO;QACL,UAAU,EAAE,OAAO,CAAC,MAAM;QAC1B,WAAW,EAAE,OAAO,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,KAAK,EAAE,EAAE,CAAC,GAAG,GAAG,KAAK,CAAC,eAAe,EAAE,CAAC,CAAC;QAC3E,UAAU;QACV,YAAY,EAAE,UAAU,CAAC,MAAM,CAAC,MAAM;QACtC,cAAc,EAAE,UAAU,CAAC,WAAW,CAAC,CAAC,MAAM;KAC/C,CAAA;AACH,CAAC;AAED,SAAS,eAAe,CAAC,OAA6B,EAAE,UAA6C,EAAE,OAA6B;IAClI,MAAM,MAAM,GAAa,EAAE,CAAA;IAC3B,IAAI,OAAO,CAAC,YAAY,GAAG,UAAU,CAAC,eAAe,EAAE,CAAC;QACtD,MAAM,CAAC,IAAI,CAAC,qCAAqC,OAAO,CAAC,YAAY,UAAU,UAAU,CAAC,eAAe,UAAU,CAAC,CAAA;IACtH,CAAC;IACD,MAAM,OAAO,GAAG,OAAO,CAAC,KAAK,EAAE,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,eAAe,GAAG,CAAC,CAAC,eAAe,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAA;IACjG,KAAK,MAAM,KAAK,IAAI,OAAO,EAAE,CAAC;QAC5B,IAAI,KAAK,CAAC,eAAe,GAAG,IAAI,IAAI,KAAK,CAAC,QAAQ,KAAK,SAAS,EAAE,CAAC;YACjE,MAAM,CAAC,IAAI,CAAC,oCAAoC,KAAK,CAAC,IAAI,KAAK,KAAK,CAAC,eAAe,WAAW,CAAC,CAAA;QAClG,CAAC;IACH,CAAC;IACD,IAAI,OAAO,CAAC,UAAU,CAAC,SAAS,CAAC,KAAK,GAAG,CAAC,EAAE,CAAC;QAC3C,MAAM,CAAC,IAAI,CAAC,4FAA4F,CAAC,CAAA;IAC3G,CAAC;IACD,OAAO,MAAM,CAAA;AACf,CAAC;AASD,SAAS,eAAe,CAAC,IAAY,EAAE,KAAe,EAAE,KAAa;IACnE,MAAM,QAAQ,GAAG,GAAG,IAAI,IAAI,KAAK,CAAC,IAAI,CAAC,GAAG,CAAC,IAAI,KAAK,EAAE,CAAC,WAAW,EAAE,CAAA;IACpE,MAAM,KAAK,GAAqB;QAC9B;YACE,EAAE,EAAE,kBAAkB;YACtB,QAAQ,EAAE,UAAU;YACpB,MAAM,EAAE,4DAA4D;YACpE,OAAO,EAAE,KAAK,CAAC,EAAE,CAAC,KAAK,CAAC,QAAQ,KAAK,UAAU;SAChD;KACF,CAAA;IACD,IAAI,kDAAkD,CAAC,IAAI,CAAC,QAAQ,CAAC,EAAE,CAAC;QACtE,KAAK,CAAC,IAAI,CAAC;YACT,EAAE,EAAE,2BAA2B;YAC/B,QAAQ,EAAE,WAAW;YACrB,MAAM,EAAE,+DAA+D;YACvE,OAAO,EAAE,KAAK,CAAC,EAAE,CAAC,KAAK,CAAC,QAAQ,KAAK,WAAW,IAAI,0CAA0C,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC;SAChH,CAAC,CAAA;IACJ,CAAC;IACD,IAAI,6CAA6C,CAAC,IAAI,CAAC,QAAQ,CAAC,EAAE,CAAC;QACjE,KAAK,CAAC,IAAI,CAAC;YACT,EAAE,EAAE,mBAAmB;YACvB,QAAQ,EAAE,WAAW;YACrB,MAAM,EAAE,yEAAyE;YACjF,OAAO,EAAE,KAAK,CAAC,EAAE,CAAC,KAAK,CAAC,QAAQ,KAAK,WAAW,IAAI,mBAAmB,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC;SACzF,CAAC,CAAA;IACJ,CAAC;IACD,IAAI,mDAAmD,CAAC,IAAI,CAAC,QAAQ,CAAC,IAAI,KAAK,CAAC,WAAW,EAAE,KAAK,GAAG,IAAI,KAAK,CAAC,WAAW,EAAE,KAAK,UAAU,EAAE,CAAC;QAC5I,KAAK,CAAC,IAAI,CAAC;YACT,EAAE,EAAE,kBAAkB;YACtB,QAAQ,EAAE,SAAS;YACnB,MAAM,EAAE,qFAAqF;YAC7F,OAAO,EAAE,KAAK,CAAC,EAAE,CAAC,KAAK,CAAC,QAAQ,KAAK,SAAS;SAC/C,CAAC,CAAA;IACJ,CAAC;IACD,OAAO,KAAK,CAAA;AACd,CAAC;AAED,SAAS,OAAO,CAAC,EAAU,EAAE,QAA+B,EAAE,QAAiB,EAAE,MAAc,EAAE,OAA6B;IAC5H,OAAO;QACL,EAAE;QACF,QAAQ;QACR,QAAQ;QACR,eAAe,EAAE,OAAO,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,KAAK,EAAE,EAAE,CAAC,GAAG,GAAG,KAAK,CAAC,eAAe,EAAE,CAAC,CAAC;QAC/E,MAAM;QACN,KAAK,EAAE,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC,KAAK,CAAC,IAAI,CAAC;KACxC,CAAA;AACH,CAAC;AAED,SAAS,cAAc,CAAC,QAA8B;IACpD,MAAM,SAAS,GAAG,IAAI,GAAG,EAAU,CAAA;IACnC,OAAO,QAAQ,CAAC,GAAG,CAAC,OAAO,CAAC,EAAE;QAC5B,MAAM,KAAK,GAAG,OAAO,CAAC,KAAK,CAAC,MAAM,CAAC,IAAI,CAAC,EAAE;YACxC,IAAI,SAAS,CAAC,GAAG,CAAC,IAAI,CAAC;gBAAE,OAAO,KAAK,CAAA;YACrC,SAAS,CAAC,GAAG,CAAC,IAAI,CAAC,CAAA;YACnB,OAAO,IAAI,CAAA;QACb,CAAC,CAAC,CAAA;QACF,OAAO;YACL,GAAG,OAAO;YACV,KAAK;YACL,eAAe,EAAE,KAAK,CAAC,MAAM,KAAK,OAAO,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,OAAO,CAAC,eAAe,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,eAAe,GAAG,CAAC,KAAK,CAAC,MAAM,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,OAAO,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC;SAC3K,CAAA;IACH,CAAC,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,IAAI,OAAO,CAAC,EAAE,KAAK,aAAa,CAAC,CAAA;AAChF,CAAC;AAED,SAAS,aAAa,CAAC,QAA+B;IACpD,OAAO,UAAU,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAA;AACrC,CAAC;AAED,SAAS,QAAQ,CAAC,IAAY;IAC5B,IAAI,CAAC;QACH,OAAO,YAAY,CAAC,IAAI,EAAE,OAAO,CAAC,CAAA;IACpC,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,EAAE,CAAA;IACX,CAAC;AACH,CAAC;AAED,SAAS,aAAa,CAAC,IAAY;IACjC,OAAO,IAAI,CAAC,OAAO,CAAC,KAAK,EAAE,GAAG,CAAC,CAAA;AACjC,CAAC;AAED,SAAS,gBAAgB,CAAC,UAAkB,EAAE,QAAgB;IAC5D,OAAO,UAAU,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,OAAO,CAAC,UAAU,EAAE,QAAQ,CAAC,CAAA;AACxE,CAAC;AAED,SAAS,kBAAkB,CAAC,UAAkB,EAAE,SAAiB,EAAE,IAAY;IAC7E,IAAI,QAAQ,CAAC,UAAU,EAAE,IAAI,CAAC;QAAE,OAAO,aAAa,CAAC,QAAQ,CAAC,UAAU,EAAE,IAAI,CAAC,CAAC,CAAA;IAChF,IAAI,QAAQ,CAAC,SAAS,EAAE,IAAI,CAAC;QAAE,OAAO,aAAa,CAAC,IAAI,CAAC,QAAQ,EAAE,QAAQ,CAAC,SAAS,EAAE,IAAI,CAAC,CAAC,CAAC,CAAA;IAC9F,OAAO,aAAa,CAAC,IAAI,CAAC,CAAA;AAC5B,CAAC;AAED,SAAS,QAAQ,CAAC,IAAY,EAAE,MAAc;IAC5C,MAAM,GAAG,GAAG,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC,CAAA;IAClC,OAAO,GAAG,KAAK,EAAE,IAAI,CAAC,CAAC,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,UAAU,CAAC,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC,CAAA;AAC3E,CAAC"}

package/dist/eval/WorkflowEval.d.ts

@@ -0,0 +1,161 @@
+export type WorkflowEvalCaseType = 'bugfix' | 'feature' | 'refactor' | 'security' | 'frontend' | 'release' | 'resource';
+export type FailureReplayCategory = 'wrong-exploration-path' | 'hallucinated-project-fact' | 'missing-codegraph-or-graph-fallback' | 'over-broad-context-load' | 'bad-skill-recommendation' | 'missing-verification-evidence' | 'failed-security-or-resource-gate' | 'human-correction-after-agent-confidence' | 'command-failure' | 'unknown';
+export interface WorkflowEvalAttempt {
+    id?: string;
+    command: string;
+    expectedExitCode?: number;
+    outputContains?: string;
+    timeoutMs?: number;
+}
+export interface WorkflowEvalCase {
+    id: string;
+    type: WorkflowEvalCaseType;
+    title: string;
+    task: string;
+    phase?: string;
+    successCriteria?: string[];
+    attempts: WorkflowEvalAttempt[];
+    expectedFailureCategory?: FailureReplayCategory;
+    humanCorrections?: number;
+    estimatedContextTokens?: number;
+}
+export interface WorkflowEvalSuite {
+    version: string;
+    id: string;
+    name: string;
+    cases: WorkflowEvalCase[];
+}
+export interface WorkflowEvalAttemptResult {
+    id: string;
+    command: string;
+    expectedExitCode: number;
+    exitCode: number;
+    passed: boolean;
+    durationMs: number;
+    outputSummary: string;
+    redactionApplied: boolean;
+}
+export interface WorkflowEvalCaseResult {
+    id: string;
+    type: WorkflowEvalCaseType;
+    title: string;
+    task: string;
+    passed: boolean;
+    passAt1: boolean;
+    passAt3: boolean;
+    fixIterations: number;
+    humanCorrections: number;
+    estimatedTokens: number;
+    toolCalls: number;
+    attempts: WorkflowEvalAttemptResult[];
+    failureReplayIds: string[];
+}
+export interface WorkflowEvalMetrics {
+    total: number;
+    passed: number;
+    failed: number;
+    passAt1: number;
+    passAt3: number;
+    passAt1Rate: number;
+    passAt3Rate: number;
+    averageFixIterations: number;
+    totalToolCalls: number;
+    estimatedTokens: number;
+    humanCorrections: number;
+    failureReplayCount: number;
+}
+export interface WorkflowEvalRun {
+    id: string;
+    suiteId: string;
+    generatedAt: string;
+    projectDir: string;
+    ok: boolean;
+    cases: WorkflowEvalCaseResult[];
+    metrics: WorkflowEvalMetrics;
+    failureReplayIds: string[];
+}
+export interface FailureReplayRecord {
+    id: string;
+    taskId: string;
+    suiteId: string;
+    caseId: string;
+    generatedAt: string;
+    category: FailureReplayCategory;
+    phase: string;
+    task: string;
+    wrongTurn: string;
+    evidence: string;
+    correction: string;
+    prevention: string;
+    replayCommand?: string;
+    status: 'open' | 'promoted' | 'accepted-risk' | 'closed';
+    redactionApplied: boolean;
+}
+export interface FailureImprovementCandidate {
+    id: string;
+    failureId: string;
+    createdAt: string;
+    category: FailureReplayCategory;
+    title: string;
+    recommendation: string;
+    evidencePath: string;
+    status: 'candidate';
+}
+export interface WorkflowEvalComparison {
+    baseline: Pick<WorkflowEvalRun, 'id' | 'suiteId' | 'metrics'>;
+    candidate: Pick<WorkflowEvalRun, 'id' | 'suiteId' | 'metrics'>;
+    delta: {
+        passAt1Rate: number;
+        passAt3Rate: number;
+        averageFixIterations: number;
+        totalToolCalls: number;
+        estimatedTokens: number;
+        humanCorrections: number;
+    };
+    recommendation: 'improved' | 'regressed' | 'mixed' | 'same';
+}
+export interface WorkflowEvalStoreOptions {
+    projectDir?: string;
+    scaleDir?: string;
+}
+export declare function defaultWorkflowEvalSuite(): WorkflowEvalSuite;
+export declare class WorkflowEvalStore {
+    readonly projectDir: string;
+    readonly scaleRoot: string;
+    readonly evalRoot: string;
+    readonly suitesDir: string;
+    readonly runsDir: string;
+    readonly failuresDir: string;
+    readonly improvementsDir: string;
+    constructor(options?: WorkflowEvalStoreOptions);
+    initSuite(suiteId?: string, force?: boolean): {
+        path: string;
+        written: boolean;
+        suite: WorkflowEvalSuite;
+    };
+    loadSuite(suiteIdOrPath?: string): WorkflowEvalSuite;
+    saveRun(run: WorkflowEvalRun): string;
+    saveFailure(record: FailureReplayRecord): string;
+    listFailures(query?: {
+        taskId?: string;
+        sinceDays?: number;
+    }): FailureReplayRecord[];
+    getFailure(id: string): FailureReplayRecord | null;
+    loadRun(idOrPath: string): WorkflowEvalRun;
+    promoteFailure(id: string): FailureImprovementCandidate;
+    suitePath(suiteId: string): string;
+    private resolveSuitePath;
+    private resolveRunPath;
+}
+export declare function runWorkflowEvalSuite(options?: WorkflowEvalStoreOptions & {
+    suite?: string;
+}): Promise<{
+    run: WorkflowEvalRun;
+    runPath: string;
+    failurePaths: string[];
+}>;
+export declare function compareWorkflowEvalRuns(options: WorkflowEvalStoreOptions & {
+    baseline: string;
+    candidate: string;
+}): WorkflowEvalComparison;
+export declare function renderWorkflowEvalReport(run: WorkflowEvalRun): string;