@kernel.chat/kbot 2.7.0 → 2.9.0

package/dist/confidence.js

@@ -0,0 +1,693 @@
+// K:BOT Confidence Engine — Self-Awareness for Agent Actions
+//
+// Three systems:
+// 1. CONFIDENCE CALIBRATION — Express uncertainty about responses/actions
+// 2. SKILL BOUNDARIES — Self-model of strengths, weaknesses, unknowns
+// 3. EFFORT ESTIMATION — Predict tool calls, cost, and complexity
+//
+// Persists calibration and effort history to ~/.kbot/ as JSON files.
+// Integrates with the learning engine for pattern/task data.
+import { homedir } from 'node:os';
+import { join } from 'node:path';
+import { existsSync, readFileSync, writeFileSync, mkdirSync } from 'node:fs';
+import { registerTool } from './tools/index.js';
+import { classifyTask, findPattern } from './learning.js';
+import { isLocalProvider, getByokProvider, estimateCost } from './auth.js';
+// ── File Paths ──
+const KBOT_DIR = join(homedir(), '.kbot');
+const CONFIDENCE_FILE = join(KBOT_DIR, 'confidence.json');
+const EFFORT_FILE = join(KBOT_DIR, 'effort-history.json');
+const SKILL_FILE = join(KBOT_DIR, 'skill-profile.json');
+// ── Helpers ──
+function ensureDir() {
+    if (!existsSync(KBOT_DIR))
+        mkdirSync(KBOT_DIR, { recursive: true });
+}
+function loadJSON(path, fallback) {
+    ensureDir();
+    if (!existsSync(path))
+        return fallback;
+    try {
+        return JSON.parse(readFileSync(path, 'utf-8'));
+    }
+    catch {
+        return fallback;
+    }
+}
+function saveJSON(path, data) {
+    ensureDir();
+    try {
+        writeFileSync(path, JSON.stringify(data, null, 2), 'utf-8');
+    }
+    catch { /* silently fail — non-critical persistence */ }
+}
+function loadCalibration() {
+    return loadJSON(CONFIDENCE_FILE, { entries: [], avgError: 0 });
+}
+function saveCalibration(data) {
+    // Keep only last 200 entries to avoid unbounded growth
+    if (data.entries.length > 200) {
+        data.entries = data.entries.slice(-200);
+    }
+    saveJSON(CONFIDENCE_FILE, data);
+}
+/** Compute calibration bias from historical data for a domain */
+function getCalibrationBias(domain) {
+    const data = loadCalibration();
+    const domainEntries = data.entries.filter(e => e.domain === domain);
+    if (domainEntries.length < 3)
+        return 0;
+    // Average (predicted - actual): positive = overconfident, negative = underconfident
+    const totalBias = domainEntries.reduce((sum, e) => sum + (e.predicted - e.actual), 0);
+    return totalBias / domainEntries.length;
+}
+/** Get historical success rate for a domain from calibration data */
+function getHistoricalSuccessRate(domain) {
+    const data = loadCalibration();
+    const entries = data.entries.filter(e => e.domain === domain);
+    if (entries.length === 0)
+        return { rate: 0.5, count: 0 };
+    const avgActual = entries.reduce((sum, e) => sum + e.actual, 0) / entries.length;
+    return { rate: avgActual, count: entries.length };
+}
+/** Detect the primary domain from task text */
+function detectDomain(task) {
+    const lower = task.toLowerCase();
+    // Check file extensions
+    const extMatch = lower.match(/\.(ts|tsx|js|jsx|py|rs|go|java|rb|cpp|c|cs|swift|kt|lua|zig|sql|css|html|md|yaml|yml|json|toml|sh|bash|zsh)/);
+    if (extMatch) {
+        const extMap = {
+            ts: 'typescript', tsx: 'typescript', js: 'javascript', jsx: 'javascript',
+            py: 'python', rs: 'rust', go: 'go', java: 'java', rb: 'ruby',
+            cpp: 'cpp', c: 'c', cs: 'csharp', swift: 'swift', kt: 'kotlin',
+            lua: 'lua', zig: 'zig', sql: 'sql', css: 'css', html: 'html',
+            md: 'writing', yaml: 'devops', yml: 'devops', json: 'config',
+            toml: 'config', sh: 'devops', bash: 'devops', zsh: 'devops',
+        };
+        return extMap[extMatch[1]] || 'general';
+    }
+    // Check language/domain keywords
+    const domainKeywords = {
+        typescript: ['typescript', 'ts ', 'tsx', 'type ', 'interface ', 'enum '],
+        javascript: ['javascript', 'js ', 'jsx', 'node', 'npm', 'yarn', 'pnpm', 'bun'],
+        python: ['python', 'pip', 'conda', 'django', 'flask', 'fastapi', 'pytorch'],
+        rust: ['rust', 'cargo', 'crate', 'impl ', 'fn ', 'struct '],
+        go: ['golang', 'go ', 'goroutine', 'go mod'],
+        devops: ['docker', 'kubernetes', 'k8s', 'terraform', 'ansible', 'ci/cd', 'pipeline', 'deploy', 'nginx', 'aws', 'gcp', 'azure'],
+        database: ['sql', 'postgres', 'mysql', 'mongo', 'redis', 'supabase', 'prisma', 'database', 'migration'],
+        writing: ['write', 'blog', 'article', 'documentation', 'readme', 'explain', 'describe'],
+        testing: ['test', 'spec', 'coverage', 'vitest', 'jest', 'pytest', 'assert'],
+        security: ['security', 'vulnerability', 'auth', 'permission', 'encrypt', 'cve'],
+        design: ['css', 'style', 'layout', 'responsive', 'animation', 'color', 'font'],
+        react: ['react', 'component', 'hook', 'jsx', 'tsx', 'next.js', 'nextjs', 'remix'],
+        git: ['git', 'commit', 'branch', 'merge', 'rebase', 'pr ', 'pull request'],
+    };
+    for (const [domain, keywords] of Object.entries(domainKeywords)) {
+        if (keywords.some(kw => lower.includes(kw)))
+            return domain;
+    }
+    return 'general';
+}
+/** Count complexity signals in a task description */
+function assessComplexity(task) {
+    const lower = task.toLowerCase();
+    // Estimate number of files involved
+    const fileRefs = (task.match(/\b\w+\.\w{1,5}\b/g) || []).length;
+    const dirRefs = (task.match(/\b\w+\//g) || []).length;
+    const fileCount = Math.max(1, fileRefs + dirRefs);
+    // Ambiguity — vague words that signal unclear intent
+    const vagueTerms = ['maybe', 'somehow', 'might', 'possibly', 'not sure', 'something like',
+        'kind of', 'sort of', 'whatever', 'figure out', 'if possible'];
+    const ambiguity = vagueTerms.filter(t => lower.includes(t)).length / vagueTerms.length;
+    // Multi-step — signals that the task has multiple phases
+    const multiStepTerms = ['then', 'after that', 'also', 'and then', 'next', 'finally',
+        'first', 'second', 'step', 'multi', 'several', 'all', 'each'];
+    const multiStep = multiStepTerms.some(t => lower.includes(t));
+    return { fileCount, ambiguity, multiStep };
+}
+/**
+ * Estimate confidence for a task before execution.
+ *
+ * Considers task complexity, past success with similar tasks,
+ * provider capability, and available context.
+ */
+export function estimateConfidence(task, context) {
+    const domain = detectDomain(task);
+    const taskType = classifyTask(task);
+    const complexity = assessComplexity(task);
+    const cachedPattern = findPattern(task);
+    const historical = getHistoricalSuccessRate(domain);
+    const calibrationBias = getCalibrationBias(domain);
+    // ── Base scores ──
+    // Factual confidence — higher if we have context, patterns, or history
+    let factual = 0.5;
+    if (context.length > 200)
+        factual += 0.15; // good context available
+    if (context.length > 1000)
+        factual += 0.1; // rich context
+    if (cachedPattern)
+        factual += 0.15; // we've done this before
+    if (historical.count > 5) {
+        factual += (historical.rate - 0.5) * 0.2; // historical performance adjustment
+    }
+    // Approach confidence — higher for known task types, lower for ambiguity
+    let approach = 0.6;
+    if (taskType !== 'general')
+        approach += 0.1; // recognized task type
+    if (cachedPattern && cachedPattern.successRate > 0.8)
+        approach += 0.2; // proven pattern
+    approach -= complexity.ambiguity * 0.3; // ambiguity reduces confidence
+    if (complexity.multiStep)
+        approach -= 0.1; // multi-step = more room for error
+    // Completeness — how sure we cover everything
+    let completeness = 0.6;
+    if (complexity.fileCount > 5)
+        completeness -= 0.15; // many files = easy to miss things
+    if (complexity.multiStep)
+        completeness -= 0.1; // multi-step = might skip a step
+    if (context.includes('repo-map') || context.includes('project structure'))
+        completeness += 0.1;
+    if (cachedPattern)
+        completeness += 0.1;
+    // Provider adjustment — local models get a penalty on complex tasks
+    try {
+        const provider = getByokProvider();
+        if (isLocalProvider(provider)) {
+            const localPenalty = complexity.multiStep ? 0.2 : 0.1;
+            factual -= localPenalty;
+            approach -= localPenalty;
+            completeness -= localPenalty * 0.5;
+        }
+    }
+    catch { /* no provider configured — skip adjustment */ }
+    // Apply calibration correction — if we're historically overconfident, reduce scores
+    factual -= calibrationBias * 0.5;
+    approach -= calibrationBias * 0.5;
+    completeness -= calibrationBias * 0.3;
+    // Clamp all scores to [0.05, 0.99]
+    factual = Math.max(0.05, Math.min(0.99, factual));
+    approach = Math.max(0.05, Math.min(0.99, approach));
+    completeness = Math.max(0.05, Math.min(0.99, completeness));
+    // Overall — weighted average
+    const overall = Math.round((factual * 0.3 + approach * 0.4 + completeness * 0.3) * 100) / 100;
+    // Build reasoning
+    const reasons = [];
+    if (cachedPattern)
+        reasons.push('have a proven pattern for this');
+    if (historical.count > 5 && historical.rate > 0.7)
+        reasons.push(`strong track record in ${domain}`);
+    if (historical.count > 5 && historical.rate < 0.4)
+        reasons.push(`historically weak in ${domain}`);
+    if (complexity.ambiguity > 0.2)
+        reasons.push('request is somewhat ambiguous');
+    if (complexity.multiStep)
+        reasons.push('multi-step task');
+    if (context.length < 100)
+        reasons.push('limited context available');
+    try {
+        if (isLocalProvider(getByokProvider()))
+            reasons.push('using local model (lower capability)');
+    }
+    catch { /* skip */ }
+    const pct = Math.round(overall * 100);
+    const reasoning = reasons.length > 0
+        ? `~${pct}% confident — ${reasons.join(', ')}`
+        : `~${pct}% confident — standard assessment for ${domain} ${taskType} task`;
+    return {
+        overall,
+        factual: Math.round(factual * 100) / 100,
+        approach: Math.round(approach * 100) / 100,
+        completeness: Math.round(completeness * 100) / 100,
+        reasoning,
+    };
+}
+/**
+ * Format a confidence score as a human-readable string.
+ */
+export function reportConfidence(score) {
+    const pct = Math.round(score.overall * 100);
+    const level = pct >= 80 ? 'high' : pct >= 50 ? 'moderate' : 'low';
+    const lines = [
+        `Confidence: ${pct}% (${level})`,
+        `  Factual:      ${Math.round(score.factual * 100)}%`,
+        `  Approach:     ${Math.round(score.approach * 100)}%`,
+        `  Completeness: ${Math.round(score.completeness * 100)}%`,
+        `  ${score.reasoning}`,
+    ];
+    return lines.join('\n');
+}
+/**
+ * Record a calibration entry — predicted vs actual (from self-eval or user feedback).
+ * Called after a task completes to improve future predictions.
+ */
+export function recordCalibration(task, predicted, actual) {
+    const domain = detectDomain(task);
+    const data = loadCalibration();
+    data.entries.push({
+        task: task.slice(0, 200), // truncate for storage
+        predicted,
+        actual,
+        domain,
+        timestamp: new Date().toISOString(),
+    });
+    // Recompute running average error
+    if (data.entries.length > 0) {
+        data.avgError = data.entries.reduce((sum, e) => sum + Math.abs(e.predicted - e.actual), 0) / data.entries.length;
+    }
+    saveCalibration(data);
+}
+function loadSkillData() {
+    return loadJSON(SKILL_FILE, { skills: {} });
+}
+function saveSkillData(data) {
+    saveJSON(SKILL_FILE, data);
+}
+/**
+ * Build a skill profile from stored skill data and calibration history.
+ */
+export function getSkillProfile() {
+    const skillData = loadSkillData();
+    const calibration = loadCalibration();
+    // Merge skill data with calibration entries for domains not in skill data
+    const domainSet = new Set(Object.keys(skillData.skills));
+    for (const entry of calibration.entries) {
+        domainSet.add(entry.domain);
+    }
+    const allDomains = Array.from(domainSet);
+    const entries = [];
+    for (const domain of allDomains) {
+        const stored = skillData.skills[domain];
+        const calEntries = calibration.entries.filter(e => e.domain === domain);
+        let successRate;
+        let avgConfidence;
+        let sampleSize;
+        let lastAttempt;
+        if (stored) {
+            sampleSize = stored.sampleSize;
+            successRate = sampleSize > 0 ? stored.successCount / sampleSize : 0.5;
+            avgConfidence = sampleSize > 0 ? stored.totalConfidence / sampleSize : 0.5;
+            lastAttempt = stored.lastAttempt;
+        }
+        else if (calEntries.length > 0) {
+            sampleSize = calEntries.length;
+            successRate = calEntries.reduce((s, e) => s + e.actual, 0) / sampleSize;
+            avgConfidence = calEntries.reduce((s, e) => s + e.predicted, 0) / sampleSize;
+            lastAttempt = calEntries[calEntries.length - 1].timestamp;
+        }
+        else {
+            continue;
+        }
+        entries.push({
+            domain,
+            successRate: Math.round(successRate * 100) / 100,
+            avgConfidence: Math.round(avgConfidence * 100) / 100,
+            sampleSize,
+            lastAttempt,
+        });
+    }
+    // All known domains that could exist but have no data
+    const allKnownDomains = [
+        'typescript', 'javascript', 'python', 'rust', 'go', 'java', 'ruby',
+        'cpp', 'c', 'csharp', 'swift', 'kotlin', 'devops', 'database', 'sql',
+        'testing', 'security', 'design', 'writing', 'react', 'git', 'config',
+        'html', 'css', 'lua', 'zig',
+    ];
+    const strengths = entries
+        .filter(e => e.successRate >= 0.7 && e.sampleSize >= 3)
+        .sort((a, b) => b.successRate - a.successRate);
+    const weaknesses = entries
+        .filter(e => e.successRate < 0.5 && e.sampleSize >= 3)
+        .sort((a, b) => a.successRate - b.successRate);
+    const assessedDomains = new Set(entries.map(e => e.domain));
+    const unknown = allKnownDomains.filter(d => {
+        const entry = entries.find(e => e.domain === d);
+        return !entry || entry.sampleSize < 3;
+    });
+    return { strengths, weaknesses, unknown };
+}
+/**
+ * Assess whether the agent is suitable for a given task.
+ */
+export function assessSkillForTask(task) {
+    const domain = detectDomain(task);
+    const profile = getSkillProfile();
+    // Check if this is a known strength
+    const strength = profile.strengths.find(s => s.domain === domain);
+    if (strength) {
+        return {
+            canDo: true,
+            confidence: strength.avgConfidence,
+        };
+    }
+    // Check if this is a known weakness
+    const weakness = profile.weaknesses.find(w => w.domain === domain);
+    if (weakness) {
+        return {
+            canDo: true,
+            confidence: weakness.avgConfidence,
+            suggestion: `Historical success rate in ${domain} is ${Math.round(weakness.successRate * 100)}% — consider breaking this into smaller steps or using a specialized tool.`,
+        };
+    }
+    // Check if unknown domain
+    if (profile.unknown.includes(domain)) {
+        return {
+            canDo: true,
+            confidence: 0.5,
+            suggestion: `Limited experience with ${domain} tasks — proceeding with caution.`,
+        };
+    }
+    // General domain — no strong signal either way
+    return {
+        canDo: true,
+        confidence: 0.6,
+    };
+}
+/**
+ * Update the skill profile after completing a task.
+ *
+ * @param domain - The task domain (auto-detected or overridden)
+ * @param success - Whether the task completed successfully
+ * @param confidence - The confidence score used for this task
+ */
+export function updateSkillProfile(domain, success, confidence) {
+    const data = loadSkillData();
+    const existing = data.skills[domain];
+    if (existing) {
+        if (success)
+            existing.successCount++;
+        else
+            existing.failureCount++;
+        existing.totalConfidence += confidence;
+        existing.sampleSize++;
+        existing.lastAttempt = new Date().toISOString();
+    }
+    else {
+        data.skills[domain] = {
+            successCount: success ? 1 : 0,
+            failureCount: success ? 0 : 1,
+            totalConfidence: confidence,
+            sampleSize: 1,
+            lastAttempt: new Date().toISOString(),
+        };
+    }
+    saveSkillData(data);
+}
+function loadEffortHistory() {
+    return loadJSON(EFFORT_FILE, { entries: [] });
+}
+function saveEffortHistory(data) {
+    // Keep only last 100
+    if (data.entries.length > 100) {
+        data.entries = data.entries.slice(-100);
+    }
+    saveJSON(EFFORT_FILE, data);
+}
+/** Get average actual tool calls for a task type from history */
+function getHistoricalEffort(taskType, domain) {
+    const history = loadEffortHistory();
+    const relevant = history.entries.filter(e => e.actual && (e.taskType === taskType || e.domain === domain));
+    if (relevant.length < 2)
+        return null;
+    const avgToolCalls = relevant.reduce((s, e) => s + (e.actual.toolCalls || 0), 0) / relevant.length;
+    const avgCost = relevant.reduce((s, e) => s + (e.actual.costUsd || 0), 0) / relevant.length;
+    return { avgToolCalls, avgCost, count: relevant.length };
+}
+/** Complexity classification based on signals */
+function classifyComplexity(task) {
+    const complexity = assessComplexity(task);
+    const taskType = classifyTask(task);
+    const wordCount = task.split(/\s+/).length;
+    // Simple heuristics
+    if (wordCount < 8 && !complexity.multiStep && complexity.fileCount <= 1)
+        return 'trivial';
+    if (wordCount < 20 && !complexity.multiStep && complexity.fileCount <= 2)
+        return 'simple';
+    if (complexity.fileCount > 10 || (complexity.multiStep && wordCount > 50))
+        return 'ambitious';
+    if (complexity.multiStep || complexity.fileCount > 5 || wordCount > 40)
+        return 'complex';
+    return 'moderate';
+}
+/** Estimate tool call counts by task type */
+function estimateToolCounts(taskType, complexityLevel, fileCount) {
+    // Base estimates by task type
+    const baseEstimates = {
+        debug: { min: 3, expected: 6, max: 15 },
+        build: { min: 5, expected: 10, max: 25 },
+        refactor: { min: 3, expected: 8, max: 20 },
+        test: { min: 2, expected: 5, max: 12 },
+        deploy: { min: 2, expected: 5, max: 10 },
+        explain: { min: 1, expected: 3, max: 6 },
+        review: { min: 2, expected: 5, max: 12 },
+        search: { min: 1, expected: 3, max: 8 },
+        general: { min: 2, expected: 5, max: 12 },
+    };
+    const base = baseEstimates[taskType] || baseEstimates.general;
+    // Complexity multiplier
+    const complexityMultiplier = {
+        trivial: 0.3,
+        simple: 0.6,
+        moderate: 1.0,
+        complex: 1.8,
+        ambitious: 3.0,
+    };
+    const mult = complexityMultiplier[complexityLevel] || 1.0;
+    // File count adjustment
+    const fileAdj = Math.max(0, (fileCount - 2) * 0.5);
+    return {
+        min: Math.max(1, Math.round(base.min * mult)),
+        expected: Math.max(1, Math.round(base.expected * mult + fileAdj)),
+        max: Math.max(2, Math.round(base.max * mult + fileAdj * 2)),
+    };
+}
+/** Build a human-readable breakdown of expected operations */
+function buildBreakdown(taskType, fileCount, complexityLevel) {
+    const parts = [];
+    // Reads
+    const reads = Math.max(1, Math.ceil(fileCount * 0.8));
+    parts.push(`~${reads} file read${reads > 1 ? 's' : ''}`);
+    // Edits
+    if (['build', 'debug', 'refactor'].includes(taskType)) {
+        const edits = Math.max(1, Math.ceil(fileCount * 0.5));
+        parts.push(`~${edits} edit${edits > 1 ? 's' : ''}`);
+    }
+    // Search
+    if (['debug', 'search', 'review', 'refactor'].includes(taskType)) {
+        parts.push('~1-2 searches');
+    }
+    // Test/build run
+    if (['test', 'build', 'debug', 'deploy'].includes(taskType)) {
+        parts.push('~1 test/build run');
+    }
+    // Git
+    if (['deploy', 'build'].includes(taskType)) {
+        parts.push('~1 git operation');
+    }
+    if (complexityLevel === 'ambitious') {
+        parts.push('may require multiple iterations');
+    }
+    return parts.join(', ');
+}
+/**
+ * Estimate the effort required for a task — tool calls, cost, and complexity.
+ *
+ * @param task - The task description
+ * @param context - Optional context (repo state, file list, etc.)
+ */
+export function estimateEffort(task, context) {
+    const taskType = classifyTask(task);
+    const domain = detectDomain(task);
+    const complexityLevel = classifyComplexity(task);
+    const complexity = assessComplexity(task);
+    // Try historical data first
+    const historical = getHistoricalEffort(taskType, domain);
+    let toolCalls;
+    if (historical && historical.count >= 3) {
+        // Use historical averages with some spread
+        const avg = Math.round(historical.avgToolCalls);
+        toolCalls = {
+            min: Math.max(1, Math.round(avg * 0.5)),
+            expected: avg,
+            max: Math.round(avg * 2),
+        };
+    }
+    else {
+        toolCalls = estimateToolCounts(taskType, complexityLevel, complexity.fileCount);
+    }
+    // Cost estimation — rough approximation based on tool calls
+    // Each tool call ~ 500 input tokens + 200 output tokens on average
+    const tokensPerCall = { input: 500, output: 200 };
+    let costPerCall;
+    try {
+        const provider = getByokProvider();
+        costPerCall = estimateCost(provider, tokensPerCall.input, tokensPerCall.output);
+        if (isLocalProvider(provider))
+            costPerCall = 0;
+    }
+    catch {
+        // Default to Anthropic pricing (~$3/$15 per MTok)
+        costPerCall = (tokensPerCall.input * 3 / 1_000_000) + (tokensPerCall.output * 15 / 1_000_000);
+    }
+    const estimatedCostUsd = {
+        min: Math.round(toolCalls.min * costPerCall * 10000) / 10000,
+        expected: Math.round(toolCalls.expected * costPerCall * 10000) / 10000,
+        max: Math.round(toolCalls.max * costPerCall * 10000) / 10000,
+    };
+    const breakdown = buildBreakdown(taskType, complexity.fileCount, complexityLevel);
+    // Store prediction for later calibration
+    const history = loadEffortHistory();
+    history.entries.push({
+        task: task.slice(0, 200),
+        taskType,
+        domain,
+        predicted: {
+            toolCalls: toolCalls.expected,
+            costUsd: estimatedCostUsd.expected,
+            complexity: complexityLevel,
+        },
+        timestamp: new Date().toISOString(),
+    });
+    saveEffortHistory(history);
+    return {
+        toolCalls,
+        estimatedCostUsd,
+        complexity: complexityLevel,
+        breakdown,
+    };
+}
+/**
+ * Record actual effort after a task completes, for future calibration.
+ */
+export function recordActualEffort(task, actualToolCalls, actualCostUsd) {
+    const history = loadEffortHistory();
+    // Find the most recent prediction for this task
+    const taskSlice = task.slice(0, 200);
+    for (let i = history.entries.length - 1; i >= 0; i--) {
+        if (history.entries[i].task === taskSlice && !history.entries[i].actual) {
+            history.entries[i].actual = { toolCalls: actualToolCalls, costUsd: actualCostUsd };
+            break;
+        }
+    }
+    saveEffortHistory(history);
+}
+// ══════════════════════════════════════════════════════════════════
+// TOOL REGISTRATION
+// ══════════════════════════════════════════════════════════════════
+/**
+ * Register confidence engine tools with the K:BOT tool registry.
+ */
+export function registerConfidenceTools() {
+    registerTool({
+        name: 'confidence_check',
+        description: 'Get a confidence score for a proposed action or task. Returns factual, approach, and completeness scores with reasoning.',
+        parameters: {
+            task: {
+                type: 'string',
+                description: 'Description of the task or action to assess confidence for',
+                required: true,
+            },
+            context: {
+                type: 'string',
+                description: 'Available context (repo state, file contents, memory, etc.)',
+                required: false,
+                default: '',
+            },
+        },
+        tier: 'free',
+        execute: async (args) => {
+            const task = String(args.task || '');
+            const context = String(args.context || '');
+            if (!task)
+                return 'Error: task parameter is required';
+            const score = estimateConfidence(task, context);
+            return reportConfidence(score);
+        },
+    });
+    registerTool({
+        name: 'skill_profile',
+        description: 'Show the agent skill profile — strengths, weaknesses, and untested domains. Optionally assess suitability for a specific task.',
+        parameters: {
+            task: {
+                type: 'string',
+                description: 'Optional task to assess suitability for',
+                required: false,
+            },
+        },
+        tier: 'free',
+        execute: async (args) => {
+            const task = args.task ? String(args.task) : null;
+            const profile = getSkillProfile();
+            const lines = ['=== Skill Profile ===', ''];
+            if (profile.strengths.length > 0) {
+                lines.push('Strengths:');
+                for (const s of profile.strengths) {
+                    lines.push(`  ${s.domain}: ${Math.round(s.successRate * 100)}% success (${s.sampleSize} tasks, avg confidence ${Math.round(s.avgConfidence * 100)}%)`);
+                }
+                lines.push('');
+            }
+            if (profile.weaknesses.length > 0) {
+                lines.push('Weaknesses:');
+                for (const w of profile.weaknesses) {
+                    lines.push(`  ${w.domain}: ${Math.round(w.successRate * 100)}% success (${w.sampleSize} tasks, avg confidence ${Math.round(w.avgConfidence * 100)}%)`);
+                }
+                lines.push('');
+            }
+            if (profile.unknown.length > 0) {
+                lines.push(`Untested domains: ${profile.unknown.join(', ')}`);
+                lines.push('');
+            }
+            if (task) {
+                lines.push('--- Task Assessment ---');
+                const assessment = assessSkillForTask(task);
+                lines.push(`Can do: ${assessment.canDo ? 'yes' : 'no'}`);
+                lines.push(`Confidence: ${Math.round(assessment.confidence * 100)}%`);
+                if (assessment.suggestion)
+                    lines.push(`Note: ${assessment.suggestion}`);
+            }
+            return lines.join('\n');
+        },
+    });
+    registerTool({
+        name: 'effort_estimate',
+        description: 'Predict how many tool calls, cost, and complexity a task will involve. Uses historical data when available.',
+        parameters: {
+            task: {
+                type: 'string',
+                description: 'Description of the task to estimate effort for',
+                required: true,
+            },
+            context: {
+                type: 'string',
+                description: 'Optional context about the current state',
+                required: false,
+            },
+        },
+        tier: 'free',
+        execute: async (args) => {
+            const task = String(args.task || '');
+            const context = args.context ? String(args.context) : undefined;
+            if (!task)
+                return 'Error: task parameter is required';
+            const estimate = estimateEffort(task, context);
+            const lines = [
+                '=== Effort Estimate ===',
+                '',
+                `Complexity: ${estimate.complexity}`,
+                '',
+                `Tool calls:`,
+                `  Min:      ${estimate.toolCalls.min}`,
+                `  Expected: ${estimate.toolCalls.expected}`,
+                `  Max:      ${estimate.toolCalls.max}`,
+                '',
+                `Estimated cost (USD):`,
+                `  Min:      $${estimate.estimatedCostUsd.min.toFixed(4)}`,
+                `  Expected: $${estimate.estimatedCostUsd.expected.toFixed(4)}`,
+                `  Max:      $${estimate.estimatedCostUsd.max.toFixed(4)}`,
+                '',
+                `Breakdown: ${estimate.breakdown}`,
+            ];
+            return lines.join('\n');
+        },
+    });
+}
+//# sourceMappingURL=confidence.js.map