tryassay 0.3.0 → 0.11.0

package/dist/runtime/confidence-calibrator.js

@@ -0,0 +1,95 @@
+// ============================================================
+// Assay Verified Agent Runtime — Confidence Calibrator
+// Compares stated confidence vs actual outcomes per domain.
+// Adjusts effective confidence to reduce overconfidence.
+// ============================================================
+const BUCKET_RANGES = [
+    [0.0, 0.2],
+    [0.2, 0.4],
+    [0.4, 0.6],
+    [0.6, 0.8],
+    [0.8, 1.0],
+];
+export class ConfidenceCalibrator {
+    profiles = new Map();
+    constructor(initial) {
+        if (initial) {
+            for (const p of initial) {
+                this.profiles.set(p.domain, p);
+            }
+        }
+    }
+    calibrate(domain, statedConfidence) {
+        const profile = this.profiles.get(domain);
+        if (!profile || profile.adjustmentFactor >= 1.0) {
+            return statedConfidence; // no adjustment needed or no data
+        }
+        // Only adjust downward (overconfidence correction)
+        return Math.min(statedConfidence, statedConfidence * profile.adjustmentFactor);
+    }
+    async update(experience) {
+        const domain = experience.domain;
+        const confidence = experience.decision.confidence;
+        const succeeded = experience.outcome === 'success';
+        let profile = this.profiles.get(domain);
+        if (!profile) {
+            profile = {
+                domain,
+                buckets: BUCKET_RANGES.map(range => ({
+                    confidenceRange: range,
+                    totalDecisions: 0,
+                    actualSuccessRate: 0,
+                    gap: 0,
+                })),
+                expectedCalibrationError: 0,
+                adjustmentFactor: 1.0,
+                lastUpdated: new Date().toISOString(),
+            };
+        }
+        // Find the right bucket
+        const bucketIndex = BUCKET_RANGES.findIndex(([low, high]) => confidence >= low && confidence < high);
+        // Handle confidence === 1.0 edge case
+        const idx = bucketIndex === -1 ? BUCKET_RANGES.length - 1 : bucketIndex;
+        const bucket = profile.buckets[idx];
+        // Update bucket stats
+        const prevSuccesses = bucket.actualSuccessRate * bucket.totalDecisions;
+        bucket.totalDecisions++;
+        bucket.actualSuccessRate = (prevSuccesses + (succeeded ? 1 : 0)) / bucket.totalDecisions;
+        // Gap = midpoint of range - actual success rate (positive = overconfident)
+        const midpoint = (bucket.confidenceRange[0] + bucket.confidenceRange[1]) / 2;
+        bucket.gap = midpoint - bucket.actualSuccessRate;
+        // Compute ECE (Expected Calibration Error)
+        const totalDecisions = profile.buckets.reduce((s, b) => s + b.totalDecisions, 0);
+        let ece = 0;
+        for (const b of profile.buckets) {
+            if (b.totalDecisions === 0)
+                continue;
+            const mid = (b.confidenceRange[0] + b.confidenceRange[1]) / 2;
+            ece += (b.totalDecisions / totalDecisions) * Math.abs(mid - b.actualSuccessRate);
+        }
+        profile.expectedCalibrationError = ece;
+        // Compute adjustment factor from high-confidence bucket
+        // If agent says 0.8-1.0 confidence but only succeeds 60% of the time,
+        // adjustment = 0.60 / 0.90 ≈ 0.67
+        const highBucket = profile.buckets[profile.buckets.length - 1];
+        if (highBucket.totalDecisions >= 5) {
+            const highMidpoint = (highBucket.confidenceRange[0] + highBucket.confidenceRange[1]) / 2;
+            const ratio = highBucket.actualSuccessRate / highMidpoint;
+            profile.adjustmentFactor = Math.min(1.0, ratio); // never adjust upward
+        }
+        profile.lastUpdated = new Date().toISOString();
+        this.profiles.set(domain, profile);
+    }
+    getProfile(domain) {
+        return this.profiles.get(domain) ?? null;
+    }
+    getAllProfiles() {
+        return Array.from(this.profiles.values());
+    }
+    load(profiles) {
+        for (const p of profiles) {
+            this.profiles.set(p.domain, p);
+        }
+    }
+}
+//# sourceMappingURL=confidence-calibrator.js.map

package/dist/runtime/confidence-calibrator.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"confidence-calibrator.js","sourceRoot":"","sources":["../../src/runtime/confidence-calibrator.ts"],"names":[],"mappings":"AAAA,+DAA+D;AAC/D,uDAAuD;AACvD,4DAA4D;AAC5D,yDAAyD;AACzD,+DAA+D;AAI/D,MAAM,aAAa,GAAuB;IACxC,CAAC,GAAG,EAAE,GAAG,CAAC;IACV,CAAC,GAAG,EAAE,GAAG,CAAC;IACV,CAAC,GAAG,EAAE,GAAG,CAAC;IACV,CAAC,GAAG,EAAE,GAAG,CAAC;IACV,CAAC,GAAG,EAAE,GAAG,CAAC;CACX,CAAC;AAEF,MAAM,OAAO,oBAAoB;IACvB,QAAQ,GAAoC,IAAI,GAAG,EAAE,CAAC;IAE9D,YAAY,OAA8B;QACxC,IAAI,OAAO,EAAE,CAAC;YACZ,KAAK,MAAM,CAAC,IAAI,OAAO,EAAE,CAAC;gBACxB,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;YACjC,CAAC;QACH,CAAC;IACH,CAAC;IAED,SAAS,CAAC,MAAc,EAAE,gBAAwB;QAChD,MAAM,OAAO,GAAG,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC;QAC1C,IAAI,CAAC,OAAO,IAAI,OAAO,CAAC,gBAAgB,IAAI,GAAG,EAAE,CAAC;YAChD,OAAO,gBAAgB,CAAC,CAAC,kCAAkC;QAC7D,CAAC;QACD,mDAAmD;QACnD,OAAO,IAAI,CAAC,GAAG,CAAC,gBAAgB,EAAE,gBAAgB,GAAG,OAAO,CAAC,gBAAgB,CAAC,CAAC;IACjF,CAAC;IAED,KAAK,CAAC,MAAM,CAAC,UAAsB;QACjC,MAAM,MAAM,GAAG,UAAU,CAAC,MAAM,CAAC;QACjC,MAAM,UAAU,GAAG,UAAU,CAAC,QAAQ,CAAC,UAAU,CAAC;QAClD,MAAM,SAAS,GAAG,UAAU,CAAC,OAAO,KAAK,SAAS,CAAC;QAEnD,IAAI,OAAO,GAAG,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC;QACxC,IAAI,CAAC,OAAO,EAAE,CAAC;YACb,OAAO,GAAG;gBACR,MAAM;gBACN,OAAO,EAAE,aAAa,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;oBACnC,eAAe,EAAE,KAAK;oBACtB,cAAc,EAAE,CAAC;oBACjB,iBAAiB,EAAE,CAAC;oBACpB,GAAG,EAAE,CAAC;iBACP,CAAC,CAAC;gBACH,wBAAwB,EAAE,CAAC;gBAC3B,gBAAgB,EAAE,GAAG;gBACrB,WAAW,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;aACtC,CAAC;QACJ,CAAC;QAED,wBAAwB;QACxB,MAAM,WAAW,GAAG,aAAa,CAAC,SAAS,CACzC,CAAC,CAAC,GAAG,EAAE,IAAI,CAAC,EAAE,EAAE,CAAC,UAAU,IAAI,GAAG,IAAI,UAAU,GAAG,IAAI,CACxD,CAAC;QACF,sCAAsC;QACtC,MAAM,GAAG,GAAG,WAAW,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,aAAa,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,WAAW,CAAC;QACxE,MAAM,MAAM,GAAG,OAAO,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC;QAEpC,sBAAsB;QACtB,MAAM,aAAa,GAAG,MAAM,CAAC,iBAAiB,GAAG,MAAM,CAAC,cAAc,CAAC;QACvE,MAAM,CAAC,cAAc,EAAE,CAAC;QACxB,MAAM,CAAC,iBAAiB,GAAG,CAAC,aAAa,GAAG,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,MAAM,CAAC,cAAc,CAAC;QAEzF,2EAA2E;QAC3E,MAAM,QAAQ,GAAG,CAAC,MAAM,CAAC,eAAe,CAAC,CAAC,CAAC,GAAG,MAAM,CAAC,eAAe,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC;QAC7E,MAAM,CAAC,GAAG,GAAG,QAAQ,GAAG,MAAM,CAAC,iBAAiB,CAAC;QAEjD,2CAA2C;QAC3C,MAAM,cAAc,GAAG,OAAO,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,cAAc,EAAE,CAAC,CAAC,CAAC;QACjF,IAAI,GAAG,GAAG,CAAC,CAAC;QACZ,KAAK,MAAM,CAAC,IAAI,OAAO,CAAC,OAAO,EAAE,CAAC;YAChC,IAAI,CAAC,CAAC,cAAc,KAAK,CAAC;gBAAE,SAAS;YACrC,MAAM,GAAG,GAAG,CAAC,CAAC,CAAC,eAAe,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,eAAe,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC;YAC9D,GAAG,IAAI,CAAC,CAAC,CAAC,cAAc,GAAG,cAAc,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,GAAG,GAAG,CAAC,CAAC,iBAAiB,CAAC,CAAC;QACnF,CAAC;QACD,OAAO,CAAC,wBAAwB,GAAG,GAAG,CAAC;QAEvC,wDAAwD;QACxD,sEAAsE;QACtE,kCAAkC;QAClC,MAAM,UAAU,GAAG,OAAO,CAAC,OAAO,CAAC,OAAO,CAAC,OAAO,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;QAC/D,IAAI,UAAU,CAAC,cAAc,IAAI,CAAC,EAAE,CAAC;YACnC,MAAM,YAAY,GAAG,CAAC,UAAU,CAAC,eAAe,CAAC,CAAC,CAAC,GAAG,UAAU,CAAC,eAAe,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC;YACzF,MAAM,KAAK,GAAG,UAAU,CAAC,iBAAiB,GAAG,YAAY,CAAC;YAC1D,OAAO,CAAC,gBAAgB,GAAG,IAAI,CAAC,GAAG,CAAC,GAAG,EAAE,KAAK,CAAC,CAAC,CAAC,sBAAsB;QACzE,CAAC;QAED,OAAO,CAAC,WAAW,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;QAC/C,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IACrC,CAAC;IAED,UAAU,CAAC,MAAc;QACvB,OAAO,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,MAAM,CAAC,IAAI,IAAI,CAAC;IAC3C,CAAC;IAED,cAAc;QACZ,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,QAAQ,CAAC,MAAM,EAAE,CAAC,CAAC;IAC5C,CAAC;IAED,IAAI,CAAC,QAA8B;QACjC,KAAK,MAAM,CAAC,IAAI,QAAQ,EAAE,CAAC;YACzB,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;QACjC,CAAC;IACH,CAAC;CACF"}

package/dist/runtime/domain-coverage-analyzer.d.ts

@@ -0,0 +1,24 @@
+import type { AgentDefinition, ToolDefinition } from './types.js';
+export interface CoverageAnalysis {
+    readonly claimed_scope: string;
+    readonly verified_scope: string;
+    readonly gaps: readonly string[];
+    readonly overreach: readonly string[];
+    readonly tool_access: {
+        readonly justified: readonly string[];
+        readonly questionable: readonly string[];
+        readonly missing: readonly string[];
+    };
+}
+export declare class DomainCoverageAnalyzer {
+    /**
+     * Analyze an agent definition for coverage gaps and overreach.
+     * Compares the agent's declared domain against known patterns
+     * and the available tool registry.
+     */
+    analyze(agent: AgentDefinition, availableTools: readonly ToolDefinition[]): CoverageAnalysis;
+    private matchDomain;
+    private isLanguageRelated;
+    private analyzeToolAccess;
+    private summarizeVerifiedScope;
+}

package/dist/runtime/domain-coverage-analyzer.js

@@ -0,0 +1,178 @@
+// ============================================================
+// Assay Verified Agent Runtime — Domain Coverage Analyzer
+// Analyzes agent definitions for domain coverage gaps and overreach.
+// ============================================================
+const DOMAIN_EXPECTATIONS = {
+    kubernetes: {
+        languages: ['yaml', 'helm'],
+        frameworks: ['kubernetes', 'docker', 'helm'],
+        expected_tools: ['cli_wrapper'],
+        expected_claims: ['security', 'correctness', 'completeness'],
+    },
+    database: {
+        languages: ['sql', 'typescript', 'python'],
+        frameworks: ['postgresql', 'mysql', 'mongodb', 'prisma', 'drizzle'],
+        expected_tools: ['cli_wrapper', 'data_transform'],
+        expected_claims: ['security', 'correctness', 'performance'],
+    },
+    frontend: {
+        languages: ['typescript', 'javascript', 'css', 'html'],
+        frameworks: ['react', 'vue', 'svelte', 'next.js', 'angular'],
+        expected_tools: ['cli_wrapper'],
+        expected_claims: ['correctness', 'completeness', 'type-safety'],
+    },
+    api: {
+        languages: ['typescript', 'python', 'go'],
+        frameworks: ['express', 'fastify', 'flask', 'gin'],
+        expected_tools: ['api_client', 'cli_wrapper'],
+        expected_claims: ['security', 'correctness', 'error-handling'],
+    },
+    security: {
+        languages: ['typescript', 'python', 'go', 'rust'],
+        frameworks: [],
+        expected_tools: ['verification_helper', 'cli_wrapper'],
+        expected_claims: ['security', 'correctness'],
+    },
+    infrastructure: {
+        languages: ['yaml', 'hcl', 'terraform'],
+        frameworks: ['terraform', 'cloudformation', 'pulumi', 'ansible'],
+        expected_tools: ['cli_wrapper'],
+        expected_claims: ['security', 'correctness', 'completeness'],
+    },
+};
+// ── Domain Coverage Analyzer ───────────────────────────────
+export class DomainCoverageAnalyzer {
+    /**
+     * Analyze an agent definition for coverage gaps and overreach.
+     * Compares the agent's declared domain against known patterns
+     * and the available tool registry.
+     */
+    analyze(agent, availableTools) {
+        const gaps = [];
+        const overreach = [];
+        // Match against known domain expectations
+        const domainKey = this.matchDomain(agent.domain);
+        const expectations = domainKey ? DOMAIN_EXPECTATIONS[domainKey] : null;
+        // Check language coverage
+        if (expectations) {
+            for (const lang of expectations.languages) {
+                if (!agent.domain.languages.includes(lang)) {
+                    gaps.push(`Domain "${domainKey}" typically handles "${lang}" but agent doesn't declare it`);
+                }
+            }
+            // Check for overreach: languages not expected for this domain
+            for (const lang of agent.domain.languages) {
+                const isExpected = expectations.languages.includes(lang);
+                const isRelated = this.isLanguageRelated(lang, expectations.languages);
+                if (!isExpected && !isRelated) {
+                    overreach.push(`Agent declares "${lang}" which is unusual for domain "${domainKey}"`);
+                }
+            }
+            // Check claim category coverage
+            for (const cat of expectations.expected_claims) {
+                if (!agent.domain.claim_categories.includes(cat)) {
+                    gaps.push(`Domain "${domainKey}" typically covers "${cat}" claims but agent doesn't declare it`);
+                }
+            }
+        }
+        // Analyze tool access
+        const toolAnalysis = this.analyzeToolAccess(agent, availableTools, expectations);
+        // Check for overly broad domain description
+        if (agent.domain.description.length < 20) {
+            gaps.push('Domain description is too brief to determine scope');
+        }
+        if (agent.domain.description.toLowerCase().includes('everything') ||
+            agent.domain.description.toLowerCase().includes('all code') ||
+            agent.domain.description.toLowerCase().includes('any language')) {
+            overreach.push('Domain description is overly broad (claims to handle "everything" or "all")');
+        }
+        // Check constraints consistency
+        if (agent.trust_level === 'sandboxed' && agent.constraints.can_spawn_agents) {
+            overreach.push('Sandboxed agents cannot spawn other agents');
+        }
+        if (agent.trust_level === 'standard' && agent.constraints.can_spawn_agents) {
+            overreach.push('Standard trust agents cannot spawn other agents');
+        }
+        return {
+            claimed_scope: agent.domain.description,
+            verified_scope: this.summarizeVerifiedScope(agent, expectations),
+            gaps,
+            overreach,
+            tool_access: toolAnalysis,
+        };
+    }
+    // ── Private helpers ──────────────────────────────────────
+    matchDomain(domain) {
+        const desc = domain.description.toLowerCase();
+        const frameworks = domain.frameworks.map(f => f.toLowerCase());
+        for (const [key, _exp] of Object.entries(DOMAIN_EXPECTATIONS)) {
+            if (desc.includes(key))
+                return key;
+            if (frameworks.some(f => _exp.frameworks.includes(f)))
+                return key;
+        }
+        return null;
+    }
+    isLanguageRelated(lang, expectedLangs) {
+        const related = {
+            typescript: ['javascript'],
+            javascript: ['typescript'],
+            yaml: ['json'],
+            json: ['yaml'],
+            hcl: ['terraform'],
+        };
+        return (related[lang] ?? []).some(r => expectedLangs.includes(r));
+    }
+    analyzeToolAccess(agent, availableTools, expectations) {
+        const justified = [];
+        const questionable = [];
+        const missing = [];
+        // Check each requested tool
+        for (const toolId of agent.tools) {
+            const tool = availableTools.find(t => t.id === toolId);
+            if (!tool) {
+                questionable.push(`${toolId} (not found in registry)`);
+                continue;
+            }
+            // Check if tool type is expected for this domain
+            if (expectations && expectations.expected_tools.includes(tool.type)) {
+                justified.push(toolId);
+            }
+            else if (!expectations) {
+                justified.push(toolId); // Can't determine, assume justified
+            }
+            else {
+                questionable.push(`${toolId} (type "${tool.type}" not typical for domain)`);
+            }
+        }
+        // Check for missing tools
+        if (expectations) {
+            for (const expectedType of expectations.expected_tools) {
+                const hasType = agent.tools.some(id => {
+                    const tool = availableTools.find(t => t.id === id);
+                    return tool?.type === expectedType;
+                });
+                if (!hasType) {
+                    missing.push(`No "${expectedType}" tool — typically needed for this domain`);
+                }
+            }
+        }
+        return { justified, questionable, missing };
+    }
+    summarizeVerifiedScope(agent, expectations) {
+        const parts = [];
+        parts.push(`Languages: ${agent.domain.languages.join(', ')}`);
+        if (agent.domain.frameworks.length > 0) {
+            parts.push(`Frameworks: ${agent.domain.frameworks.join(', ')}`);
+        }
+        parts.push(`Claims: ${agent.domain.claim_categories.join(', ')}`);
+        if (expectations) {
+            parts.push(`Matched domain: known pattern`);
+        }
+        else {
+            parts.push(`Domain: no known pattern match (custom)`);
+        }
+        return parts.join('. ');
+    }
+}
+//# sourceMappingURL=domain-coverage-analyzer.js.map

package/dist/runtime/domain-coverage-analyzer.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"domain-coverage-analyzer.js","sourceRoot":"","sources":["../../src/runtime/domain-coverage-analyzer.ts"],"names":[],"mappings":"AAAA,+DAA+D;AAC/D,0DAA0D;AAC1D,qEAAqE;AACrE,+DAA+D;AA8B/D,MAAM,mBAAmB,GAAsC;IAC7D,UAAU,EAAE;QACV,SAAS,EAAE,CAAC,MAAM,EAAE,MAAM,CAAC;QAC3B,UAAU,EAAE,CAAC,YAAY,EAAE,QAAQ,EAAE,MAAM,CAAC;QAC5C,cAAc,EAAE,CAAC,aAAa,CAAC;QAC/B,eAAe,EAAE,CAAC,UAAU,EAAE,aAAa,EAAE,cAAc,CAAC;KAC7D;IACD,QAAQ,EAAE;QACR,SAAS,EAAE,CAAC,KAAK,EAAE,YAAY,EAAE,QAAQ,CAAC;QAC1C,UAAU,EAAE,CAAC,YAAY,EAAE,OAAO,EAAE,SAAS,EAAE,QAAQ,EAAE,SAAS,CAAC;QACnE,cAAc,EAAE,CAAC,aAAa,EAAE,gBAAgB,CAAC;QACjD,eAAe,EAAE,CAAC,UAAU,EAAE,aAAa,EAAE,aAAa,CAAC;KAC5D;IACD,QAAQ,EAAE;QACR,SAAS,EAAE,CAAC,YAAY,EAAE,YAAY,EAAE,KAAK,EAAE,MAAM,CAAC;QACtD,UAAU,EAAE,CAAC,OAAO,EAAE,KAAK,EAAE,QAAQ,EAAE,SAAS,EAAE,SAAS,CAAC;QAC5D,cAAc,EAAE,CAAC,aAAa,CAAC;QAC/B,eAAe,EAAE,CAAC,aAAa,EAAE,cAAc,EAAE,aAAa,CAAC;KAChE;IACD,GAAG,EAAE;QACH,SAAS,EAAE,CAAC,YAAY,EAAE,QAAQ,EAAE,IAAI,CAAC;QACzC,UAAU,EAAE,CAAC,SAAS,EAAE,SAAS,EAAE,OAAO,EAAE,KAAK,CAAC;QAClD,cAAc,EAAE,CAAC,YAAY,EAAE,aAAa,CAAC;QAC7C,eAAe,EAAE,CAAC,UAAU,EAAE,aAAa,EAAE,gBAAgB,CAAC;KAC/D;IACD,QAAQ,EAAE;QACR,SAAS,EAAE,CAAC,YAAY,EAAE,QAAQ,EAAE,IAAI,EAAE,MAAM,CAAC;QACjD,UAAU,EAAE,EAAE;QACd,cAAc,EAAE,CAAC,qBAAqB,EAAE,aAAa,CAAC;QACtD,eAAe,EAAE,CAAC,UAAU,EAAE,aAAa,CAAC;KAC7C;IACD,cAAc,EAAE;QACd,SAAS,EAAE,CAAC,MAAM,EAAE,KAAK,EAAE,WAAW,CAAC;QACvC,UAAU,EAAE,CAAC,WAAW,EAAE,gBAAgB,EAAE,QAAQ,EAAE,SAAS,CAAC;QAChE,cAAc,EAAE,CAAC,aAAa,CAAC;QAC/B,eAAe,EAAE,CAAC,UAAU,EAAE,aAAa,EAAE,cAAc,CAAC;KAC7D;CACF,CAAC;AAEF,8DAA8D;AAE9D,MAAM,OAAO,sBAAsB;IACjC;;;;OAIG;IACH,OAAO,CACL,KAAsB,EACtB,cAAyC;QAEzC,MAAM,IAAI,GAAa,EAAE,CAAC;QAC1B,MAAM,SAAS,GAAa,EAAE,CAAC;QAE/B,0CAA0C;QAC1C,MAAM,SAAS,GAAG,IAAI,CAAC,WAAW,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;QACjD,MAAM,YAAY,GAAG,SAAS,CAAC,CAAC,CAAC,mBAAmB,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;QAEvE,0BAA0B;QAC1B,IAAI,YAAY,EAAE,CAAC;YACjB,KAAK,MAAM,IAAI,IAAI,YAAY,CAAC,SAAS,EAAE,CAAC;gBAC1C,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,SAAS,CAAC,QAAQ,CAAC,IAAI,CAAC,EAAE,CAAC;oBAC3C,IAAI,CAAC,IAAI,CAAC,WAAW,SAAS,wBAAwB,IAAI,gCAAgC,CAAC,CAAC;gBAC9F,CAAC;YACH,CAAC;YAED,8DAA8D;YAC9D,KAAK,MAAM,IAAI,IAAI,KAAK,CAAC,MAAM,CAAC,SAAS,EAAE,CAAC;gBAC1C,MAAM,UAAU,GAAG,YAAY,CAAC,SAAS,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC;gBACzD,MAAM,SAAS,GAAG,IAAI,CAAC,iBAAiB,CAAC,IAAI,EAAE,YAAY,CAAC,SAAS,CAAC,CAAC;gBACvE,IAAI,CAAC,UAAU,IAAI,CAAC,SAAS,EAAE,CAAC;oBAC9B,SAAS,CAAC,IAAI,CAAC,mBAAmB,IAAI,kCAAkC,SAAS,GAAG,CAAC,CAAC;gBACxF,CAAC;YACH,CAAC;YAED,gCAAgC;YAChC,KAAK,MAAM,GAAG,IAAI,YAAY,CAAC,eAAe,EAAE,CAAC;gBAC/C,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,gBAAgB,CAAC,QAAQ,CAAC,GAAG,CAAC,EAAE,CAAC;oBACjD,IAAI,CAAC,IAAI,CAAC,WAAW,SAAS,uBAAuB,GAAG,uCAAuC,CAAC,CAAC;gBACnG,CAAC;YACH,CAAC;QACH,CAAC;QAED,sBAAsB;QACtB,MAAM,YAAY,GAAG,IAAI,CAAC,iBAAiB,CAAC,KAAK,EAAE,cAAc,EAAE,YAAY,CAAC,CAAC;QAEjF,4CAA4C;QAC5C,IAAI,KAAK,CAAC,MAAM,CAAC,WAAW,CAAC,MAAM,GAAG,EAAE,EAAE,CAAC;YACzC,IAAI,CAAC,IAAI,CAAC,oDAAoD,CAAC,CAAC;QAClE,CAAC;QACD,IAAI,KAAK,CAAC,MAAM,CAAC,WAAW,CAAC,WAAW,EAAE,CAAC,QAAQ,CAAC,YAAY,CAAC;YAC7D,KAAK,CAAC,MAAM,CAAC,WAAW,CAAC,WAAW,EAAE,CAAC,QAAQ,CAAC,UAAU,CAAC;YAC3D,KAAK,CAAC,MAAM,CAAC,WAAW,CAAC,WAAW,EAAE,CAAC,QAAQ,CAAC,cAAc,CAAC,EAAE,CAAC;YACpE,SAAS,CAAC,IAAI,CAAC,6EAA6E,CAAC,CAAC;QAChG,CAAC;QAED,gCAAgC;QAChC,IAAI,KAAK,CAAC,WAAW,KAAK,WAAW,IAAI,KAAK,CAAC,WAAW,CAAC,gBAAgB,EAAE,CAAC;YAC5E,SAAS,CAAC,IAAI,CAAC,4CAA4C,CAAC,CAAC;QAC/D,CAAC;QACD,IAAI,KAAK,CAAC,WAAW,KAAK,UAAU,IAAI,KAAK,CAAC,WAAW,CAAC,gBAAgB,EAAE,CAAC;YAC3E,SAAS,CAAC,IAAI,CAAC,iDAAiD,CAAC,CAAC;QACpE,CAAC;QAED,OAAO;YACL,aAAa,EAAE,KAAK,CAAC,MAAM,CAAC,WAAW;YACvC,cAAc,EAAE,IAAI,CAAC,sBAAsB,CAAC,KAAK,EAAE,YAAY,CAAC;YAChE,IAAI;YACJ,SAAS;YACT,WAAW,EAAE,YAAY;SAC1B,CAAC;IACJ,CAAC;IAED,4DAA4D;IAEpD,WAAW,CAAC,MAAiC;QACnD,MAAM,IAAI,GAAG,MAAM,CAAC,WAAW,CAAC,WAAW,EAAE,CAAC;QAC9C,MAAM,UAAU,GAAG,MAAM,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,WAAW,EAAE,CAAC,CAAC;QAE/D,KAAK,MAAM,CAAC,GAAG,EAAE,IAAI,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,mBAAmB,CAAC,EAAE,CAAC;YAC9D,IAAI,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC;gBAAE,OAAO,GAAG,CAAC;YACnC,IAAI,UAAU,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,IAAI,CAAC,UAAU,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC;gBAAE,OAAO,GAAG,CAAC;QACpE,CAAC;QACD,OAAO,IAAI,CAAC;IACd,CAAC;IAEO,iBAAiB,CAAC,IAAY,EAAE,aAAuB;QAC7D,MAAM,OAAO,GAA6B;YACxC,UAAU,EAAE,CAAC,YAAY,CAAC;YAC1B,UAAU,EAAE,CAAC,YAAY,CAAC;YAC1B,IAAI,EAAE,CAAC,MAAM,CAAC;YACd,IAAI,EAAE,CAAC,MAAM,CAAC;YACd,GAAG,EAAE,CAAC,WAAW,CAAC;SACnB,CAAC;QACF,OAAO,CAAC,OAAO,CAAC,IAAI,CAAC,IAAI,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,aAAa,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC;IACpE,CAAC;IAEO,iBAAiB,CACvB,KAAsB,EACtB,cAAyC,EACzC,YAAsC;QAEtC,MAAM,SAAS,GAAa,EAAE,CAAC;QAC/B,MAAM,YAAY,GAAa,EAAE,CAAC;QAClC,MAAM,OAAO,GAAa,EAAE,CAAC;QAE7B,4BAA4B;QAC5B,KAAK,MAAM,MAAM,IAAI,KAAK,CAAC,KAAK,EAAE,CAAC;YACjC,MAAM,IAAI,GAAG,cAAc,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,KAAK,MAAM,CAAC,CAAC;YACvD,IAAI,CAAC,IAAI,EAAE,CAAC;gBACV,YAAY,CAAC,IAAI,CAAC,GAAG,MAAM,0BAA0B,CAAC,CAAC;gBACvD,SAAS;YACX,CAAC;YAED,iDAAiD;YACjD,IAAI,YAAY,IAAI,YAAY,CAAC,cAAc,CAAC,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC;gBACpE,SAAS,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;YACzB,CAAC;iBAAM,IAAI,CAAC,YAAY,EAAE,CAAC;gBACzB,SAAS,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,oCAAoC;YAC9D,CAAC;iBAAM,CAAC;gBACN,YAAY,CAAC,IAAI,CAAC,GAAG,MAAM,WAAW,IAAI,CAAC,IAAI,2BAA2B,CAAC,CAAC;YAC9E,CAAC;QACH,CAAC;QAED,0BAA0B;QAC1B,IAAI,YAAY,EAAE,CAAC;YACjB,KAAK,MAAM,YAAY,IAAI,YAAY,CAAC,cAAc,EAAE,CAAC;gBACvD,MAAM,OAAO,GAAG,KAAK,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,EAAE;oBACpC,MAAM,IAAI,GAAG,cAAc,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,KAAK,EAAE,CAAC,CAAC;oBACnD,OAAO,IAAI,EAAE,IAAI,KAAK,YAAY,CAAC;gBACrC,CAAC,CAAC,CAAC;gBACH,IAAI,CAAC,OAAO,EAAE,CAAC;oBACb,OAAO,CAAC,IAAI,CAAC,OAAO,YAAY,2CAA2C,CAAC,CAAC;gBAC/E,CAAC;YACH,CAAC;QACH,CAAC;QAED,OAAO,EAAE,SAAS,EAAE,YAAY,EAAE,OAAO,EAAE,CAAC;IAC9C,CAAC;IAEO,sBAAsB,CAC5B,KAAsB,EACtB,YAAsC;QAEtC,MAAM,KAAK,GAAa,EAAE,CAAC;QAC3B,KAAK,CAAC,IAAI,CAAC,cAAc,KAAK,CAAC,MAAM,CAAC,SAAS,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QAC9D,IAAI,KAAK,CAAC,MAAM,CAAC,UAAU,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACvC,KAAK,CAAC,IAAI,CAAC,eAAe,KAAK,CAAC,MAAM,CAAC,UAAU,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QAClE,CAAC;QACD,KAAK,CAAC,IAAI,CAAC,WAAW,KAAK,CAAC,MAAM,CAAC,gBAAgB,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QAClE,IAAI,YAAY,EAAE,CAAC;YACjB,KAAK,CAAC,IAAI,CAAC,+BAA+B,CAAC,CAAC;QAC9C,CAAC;aAAM,CAAC;YACN,KAAK,CAAC,IAAI,CAAC,yCAAyC,CAAC,CAAC;QACxD,CAAC;QACD,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAC1B,CAAC;CACF"}

package/dist/runtime/enriched-prompt-builder.d.ts

@@ -0,0 +1,25 @@
+import type { Observation, Decision, AgentConfig, ReasoningContext, PlanningContext } from './types.js';
+import type { ExperienceStoreInterface } from './reflector.js';
+import type { ConfidenceCalibrator } from './confidence-calibrator.js';
+import type { StrategyLibrary } from './strategy-library.js';
+import type { ExtractionResult } from './pattern-extractor.js';
+export declare class EnrichedPromptBuilder {
+    private store;
+    private calibrator;
+    private strategyLibrary;
+    private lastExtraction;
+    private tokenBudget;
+    constructor(opts: {
+        store: ExperienceStoreInterface;
+        calibrator: ConfidenceCalibrator;
+        strategyLibrary: StrategyLibrary;
+        tokenBudget?: number;
+    });
+    setLastExtraction(result: ExtractionResult): void;
+    forReasoning(observation: Observation, agentConfig: AgentConfig): Promise<ReasoningContext>;
+    forPlanning(decision: Decision, agentConfig: AgentConfig): Promise<PlanningContext>;
+    private buildReasoningPrompt;
+    private buildPlanningPrompt;
+    private inferDomain;
+    private inferDomainFromDecision;
+}

package/dist/runtime/enriched-prompt-builder.js

@@ -0,0 +1,173 @@
+// ============================================================
+// Assay Verified Agent Runtime — Enriched Prompt Builder
+// Assembles structured learning context for Reasoner and Planner.
+// Replaces raw experience injection with distilled knowledge.
+// ============================================================
+export class EnrichedPromptBuilder {
+    store;
+    calibrator;
+    strategyLibrary;
+    lastExtraction = null;
+    tokenBudget;
+    constructor(opts) {
+        this.store = opts.store;
+        this.calibrator = opts.calibrator;
+        this.strategyLibrary = opts.strategyLibrary;
+        this.tokenBudget = opts.tokenBudget ?? 2000;
+    }
+    setLastExtraction(result) {
+        this.lastExtraction = result;
+    }
+    async forReasoning(observation, agentConfig) {
+        const contextString = `${observation.source} ${JSON.stringify(observation.payload).slice(0, 200)}`;
+        // Get relevant experiences
+        let relevantExperiences;
+        if ('search' in this.store && typeof this.store.search === 'function') {
+            relevantExperiences = await this.store.search(contextString, 3);
+        }
+        else {
+            const exps = await this.store.getRelevantExperiences(contextString, 3);
+            relevantExperiences = exps.map((e, i) => ({
+                experience: e,
+                similarity: 1 - i * 0.1,
+            }));
+        }
+        // Infer domain from observation
+        const domain = this.inferDomain(observation);
+        // Get skill profile
+        const skillProfile = this.lastExtraction?.skillProfiles.find(sp => sp.domain === domain) ?? null;
+        // Get patterns and anti-patterns
+        const patterns = (this.lastExtraction?.patterns ?? []).filter(p => p.domain === domain).slice(0, 3);
+        const antiPatterns = (this.lastExtraction?.antiPatterns ?? []).filter(ap => ap.domain === domain).slice(0, 3);
+        // Get calibration
+        const calibration = this.calibrator.getProfile(domain);
+        // Build prompt section
+        const promptSection = this.buildReasoningPrompt(domain, skillProfile, patterns, antiPatterns, calibration, relevantExperiences);
+        return {
+            relevantExperiences,
+            skillProfile,
+            patterns,
+            antiPatterns,
+            calibration,
+            promptSection,
+        };
+    }
+    async forPlanning(decision, agentConfig) {
+        const domain = this.inferDomainFromDecision(decision);
+        // Get operation types from proposed actions
+        const opTypes = new Set(decision.proposedActions.map(a => a.operationType));
+        // Get ranked strategies for each operation type
+        const allStrategies = [];
+        for (const opType of opTypes) {
+            const ranked = await this.strategyLibrary.rank(domain, opType);
+            allStrategies.push(...ranked.slice(0, 3));
+        }
+        // Get hotspots
+        const hotspots = (this.lastExtraction?.verificationHotspots ?? [])
+            .filter(h => h.failRate > 0.1)
+            .sort((a, b) => b.failRate - a.failRate)
+            .slice(0, 5);
+        const promptSection = this.buildPlanningPrompt(domain, allStrategies, hotspots);
+        return {
+            rankedStrategies: allStrategies,
+            hotspots,
+            promptSection,
+        };
+    }
+    buildReasoningPrompt(domain, skillProfile, patterns, antiPatterns, calibration, experiences) {
+        const sections = [];
+        // Calibration (highest priority, smallest)
+        if (calibration) {
+            const tendency = calibration.adjustmentFactor < 0.9
+                ? `overconfident by ~${Math.round((1 - calibration.adjustmentFactor) * 100)}%`
+                : calibration.adjustmentFactor < 1.0
+                    ? `slightly overconfident`
+                    : `well-calibrated`;
+            sections.push(`**Confidence calibration (${domain}):** You tend to be ${tendency}. ECE: ${calibration.expectedCalibrationError.toFixed(3)}`);
+        }
+        // Anti-patterns (second priority)
+        if (antiPatterns.length > 0) {
+            sections.push(`**What fails in ${domain}:**`);
+            for (const ap of antiPatterns) {
+                sections.push(`- ${ap.description} (${ap.claimCategory}, seen ${ap.occurrenceCount}x)`);
+            }
+        }
+        // Patterns (third priority)
+        if (patterns.length > 0) {
+            sections.push(`**What works in ${domain}:**`);
+            for (const p of patterns) {
+                sections.push(`- ${p.description} (verified ${p.verificationCount}x)`);
+            }
+        }
+        // Experiences (fourth priority)
+        if (experiences.length > 0) {
+            sections.push(`**Similar past experiences:**`);
+            for (const se of experiences) {
+                const e = se.experience;
+                sections.push(`- [${e.outcome}] ${e.lessons.slice(0, 2).join('; ')} (similarity: ${se.similarity.toFixed(2)})`);
+            }
+        }
+        // Skill profile (fifth priority)
+        if (skillProfile) {
+            sections.push(`**Track record (${domain}):** ${(skillProfile.successRate * 100).toFixed(0)}% success across ${skillProfile.totalExperiences} experiences. First-pass rate: ${(skillProfile.firstPassRate * 100).toFixed(0)}%`);
+        }
+        // Trim to token budget (rough estimate: 4 chars per token)
+        const joined = sections.join('\n');
+        const charBudget = this.tokenBudget * 4;
+        if (joined.length > charBudget) {
+            return joined.slice(0, charBudget) + '\n[...truncated]';
+        }
+        return joined;
+    }
+    buildPlanningPrompt(domain, strategies, hotspots) {
+        const sections = [];
+        if (strategies.length > 0) {
+            sections.push(`## Available Strategies\n`);
+            sections.push(`For ${domain}, these strategies have worked:`);
+            for (let i = 0; i < strategies.length; i++) {
+                const s = strategies[i];
+                sections.push(`${i + 1}. "${s.name}" — ${(s.successRate * 100).toFixed(0)}% success, used ${s.timesUsed}x`);
+                sections.push(`   ${s.description}`);
+            }
+        }
+        if (hotspots.length > 0) {
+            sections.push(`\n## Verification Hotspots (avoid these)\n`);
+            for (const h of hotspots) {
+                sections.push(`- ${h.claimCategory}: ${(h.failRate * 100).toFixed(0)}% fail rate. Common cause: ${h.topFailureReasons[0] ?? 'unknown'}`);
+                sections.push(`  → Ensure your plan addresses this explicitly.`);
+            }
+        }
+        return sections.join('\n');
+    }
+    inferDomain(observation) {
+        const payload = JSON.stringify(observation.payload).toLowerCase();
+        if (payload.includes('test'))
+            return 'testing';
+        if (payload.includes('security') || payload.includes('auth'))
+            return 'security';
+        if (payload.includes('deploy') || payload.includes('ci'))
+            return 'devops';
+        if (payload.includes('database') || payload.includes('sql'))
+            return 'database';
+        if (payload.includes('api') || payload.includes('endpoint'))
+            return 'api';
+        if (observation.source === 'filesystem')
+            return 'filesystem';
+        return 'general';
+    }
+    inferDomainFromDecision(decision) {
+        const reasoning = decision.reasoning.toLowerCase();
+        if (reasoning.includes('test'))
+            return 'testing';
+        if (reasoning.includes('security') || reasoning.includes('auth'))
+            return 'security';
+        if (reasoning.includes('deploy') || reasoning.includes('ci'))
+            return 'devops';
+        if (reasoning.includes('database') || reasoning.includes('sql'))
+            return 'database';
+        if (reasoning.includes('api') || reasoning.includes('endpoint'))
+            return 'api';
+        return 'general';
+    }
+}
+//# sourceMappingURL=enriched-prompt-builder.js.map

package/dist/runtime/enriched-prompt-builder.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"enriched-prompt-builder.js","sourceRoot":"","sources":["../../src/runtime/enriched-prompt-builder.ts"],"names":[],"mappings":"AAAA,+DAA+D;AAC/D,yDAAyD;AACzD,kEAAkE;AAClE,8DAA8D;AAC9D,+DAA+D;AAsB/D,MAAM,OAAO,qBAAqB;IACxB,KAAK,CAA2B;IAChC,UAAU,CAAuB;IACjC,eAAe,CAAkB;IACjC,cAAc,GAA4B,IAAI,CAAC;IAC/C,WAAW,CAAS;IAE5B,YAAY,IAKX;QACC,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC;QACxB,IAAI,CAAC,UAAU,GAAG,IAAI,CAAC,UAAU,CAAC;QAClC,IAAI,CAAC,eAAe,GAAG,IAAI,CAAC,eAAe,CAAC;QAC5C,IAAI,CAAC,WAAW,GAAG,IAAI,CAAC,WAAW,IAAI,IAAI,CAAC;IAC9C,CAAC;IAED,iBAAiB,CAAC,MAAwB;QACxC,IAAI,CAAC,cAAc,GAAG,MAAM,CAAC;IAC/B,CAAC;IAED,KAAK,CAAC,YAAY,CAChB,WAAwB,EACxB,WAAwB;QAExB,MAAM,aAAa,GAAG,GAAG,WAAW,CAAC,MAAM,IAAI,IAAI,CAAC,SAAS,CAAC,WAAW,CAAC,OAAO,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,CAAC;QAEnG,2BAA2B;QAC3B,IAAI,mBAAuC,CAAC;QAC5C,IAAI,QAAQ,IAAI,IAAI,CAAC,KAAK,IAAI,OAAQ,IAAI,CAAC,KAAiC,CAAC,MAAM,KAAK,UAAU,EAAE,CAAC;YACnG,mBAAmB,GAAG,MAAO,IAAI,CAAC,KAAiC,CAAC,MAAM,CAAC,aAAa,EAAE,CAAC,CAAC,CAAC;QAC/F,CAAC;aAAM,CAAC;YACN,MAAM,IAAI,GAAG,MAAM,IAAI,CAAC,KAAK,CAAC,sBAAsB,CAAC,aAAa,EAAE,CAAC,CAAC,CAAC;YACvE,mBAAmB,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC;gBACxC,UAAU,EAAE,CAAC;gBACb,UAAU,EAAE,CAAC,GAAG,CAAC,GAAG,GAAG;aACxB,CAAC,CAAC,CAAC;QACN,CAAC;QAED,gCAAgC;QAChC,MAAM,MAAM,GAAG,IAAI,CAAC,WAAW,CAAC,WAAW,CAAC,CAAC;QAE7C,oBAAoB;QACpB,MAAM,YAAY,GAAG,IAAI,CAAC,cAAc,EAAE,aAAa,CAAC,IAAI,CAAC,EAAE,CAAC,EAAE,CAAC,EAAE,CAAC,MAAM,KAAK,MAAM,CAAC,IAAI,IAAI,CAAC;QAEjG,iCAAiC;QACjC,MAAM,QAAQ,GAAG,CAAC,IAAI,CAAC,cAAc,EAAE,QAAQ,IAAI,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,MAAM,KAAK,MAAM,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;QACpG,MAAM,YAAY,GAAG,CAAC,IAAI,CAAC,cAAc,EAAE,YAAY,IAAI,EAAE,CAAC,CAAC,MAAM,CAAC,EAAE,CAAC,EAAE,CAAC,EAAE,CAAC,MAAM,KAAK,MAAM,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;QAE9G,kBAAkB;QAClB,MAAM,WAAW,GAAG,IAAI,CAAC,UAAU,CAAC,UAAU,CAAC,MAAM,CAAC,CAAC;QAEvD,uBAAuB;QACvB,MAAM,aAAa,GAAG,IAAI,CAAC,oBAAoB,CAC7C,MAAM,EACN,YAAY,EACZ,QAAQ,EACR,YAAY,EACZ,WAAW,EACX,mBAAmB,CACpB,CAAC;QAEF,OAAO;YACL,mBAAmB;YACnB,YAAY;YACZ,QAAQ;YACR,YAAY;YACZ,WAAW;YACX,aAAa;SACd,CAAC;IACJ,CAAC;IAED,KAAK,CAAC,WAAW,CACf,QAAkB,EAClB,WAAwB;QAExB,MAAM,MAAM,GAAG,IAAI,CAAC,uBAAuB,CAAC,QAAQ,CAAC,CAAC;QAEtD,4CAA4C;QAC5C,MAAM,OAAO,GAAG,IAAI,GAAG,CAAC,QAAQ,CAAC,eAAe,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,aAAa,CAAC,CAAC,CAAC;QAE5E,gDAAgD;QAChD,MAAM,aAAa,GAAe,EAAE,CAAC;QACrC,KAAK,MAAM,MAAM,IAAI,OAAO,EAAE,CAAC;YAC7B,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,eAAe,CAAC,IAAI,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;YAC/D,aAAa,CAAC,IAAI,CAAC,GAAG,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC;QAC5C,CAAC;QAED,eAAe;QACf,MAAM,QAAQ,GAAG,CAAC,IAAI,CAAC,cAAc,EAAE,oBAAoB,IAAI,EAAE,CAAC;aAC/D,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,QAAQ,GAAG,GAAG,CAAC;aAC7B,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,GAAG,CAAC,CAAC,QAAQ,CAAC;aACvC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;QAEf,MAAM,aAAa,GAAG,IAAI,CAAC,mBAAmB,CAAC,MAAM,EAAE,aAAa,EAAE,QAAQ,CAAC,CAAC;QAEhF,OAAO;YACL,gBAAgB,EAAE,aAAa;YAC/B,QAAQ;YACR,aAAa;SACd,CAAC;IACJ,CAAC;IAEO,oBAAoB,CAC1B,MAAc,EACd,YAAiC,EACjC,QAAwB,EACxB,YAAgC,EAChC,WAAsC,EACtC,WAA+B;QAE/B,MAAM,QAAQ,GAAa,EAAE,CAAC;QAE9B,2CAA2C;QAC3C,IAAI,WAAW,EAAE,CAAC;YAChB,MAAM,QAAQ,GAAG,WAAW,CAAC,gBAAgB,GAAG,GAAG;gBACjD,CAAC,CAAC,qBAAqB,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,GAAG,WAAW,CAAC,gBAAgB,CAAC,GAAG,GAAG,CAAC,GAAG;gBAC9E,CAAC,CAAC,WAAW,CAAC,gBAAgB,GAAG,GAAG;oBACpC,CAAC,CAAC,wBAAwB;oBAC1B,CAAC,CAAC,iBAAiB,CAAC;YACtB,QAAQ,CAAC,IAAI,CAAC,6BAA6B,MAAM,uBAAuB,QAAQ,UAAU,WAAW,CAAC,wBAAwB,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;QAC/I,CAAC;QAED,kCAAkC;QAClC,IAAI,YAAY,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YAC5B,QAAQ,CAAC,IAAI,CAAC,mBAAmB,MAAM,KAAK,CAAC,CAAC;YAC9C,KAAK,MAAM,EAAE,IAAI,YAAY,EAAE,CAAC;gBAC9B,QAAQ,CAAC,IAAI,CAAC,KAAK,EAAE,CAAC,WAAW,KAAK,EAAE,CAAC,aAAa,UAAU,EAAE,CAAC,eAAe,IAAI,CAAC,CAAC;YAC1F,CAAC;QACH,CAAC;QAED,4BAA4B;QAC5B,IAAI,QAAQ,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACxB,QAAQ,CAAC,IAAI,CAAC,mBAAmB,MAAM,KAAK,CAAC,CAAC;YAC9C,KAAK,MAAM,CAAC,IAAI,QAAQ,EAAE,CAAC;gBACzB,QAAQ,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,WAAW,cAAc,CAAC,CAAC,iBAAiB,IAAI,CAAC,CAAC;YACzE,CAAC;QACH,CAAC;QAED,gCAAgC;QAChC,IAAI,WAAW,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YAC3B,QAAQ,CAAC,IAAI,CAAC,+BAA+B,CAAC,CAAC;YAC/C,KAAK,MAAM,EAAE,IAAI,WAAW,EAAE,CAAC;gBAC7B,MAAM,CAAC,GAAG,EAAE,CAAC,UAAU,CAAC;gBACxB,QAAQ,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,OAAO,KAAK,CAAC,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,iBAAiB,EAAE,CAAC,UAAU,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC;YAClH,CAAC;QACH,CAAC;QAED,iCAAiC;QACjC,IAAI,YAAY,EAAE,CAAC;YACjB,QAAQ,CAAC,IAAI,CAAC,mBAAmB,MAAM,QAAQ,CAAC,YAAY,CAAC,WAAW,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,oBAAoB,YAAY,CAAC,gBAAgB,kCAAkC,CAAC,YAAY,CAAC,aAAa,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC;QACjO,CAAC;QAED,2DAA2D;QAC3D,MAAM,MAAM,GAAG,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QACnC,MAAM,UAAU,GAAG,IAAI,CAAC,WAAW,GAAG,CAAC,CAAC;QACxC,IAAI,MAAM,CAAC,MAAM,GAAG,UAAU,EAAE,CAAC;YAC/B,OAAO,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,UAAU,CAAC,GAAG,kBAAkB,CAAC;QAC1D,CAAC;QACD,OAAO,MAAM,CAAC;IAChB,CAAC;IAEO,mBAAmB,CACzB,MAAc,EACd,UAAsB,EACtB,QAA+B;QAE/B,MAAM,QAAQ,GAAa,EAAE,CAAC;QAE9B,IAAI,UAAU,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YAC1B,QAAQ,CAAC,IAAI,CAAC,2BAA2B,CAAC,CAAC;YAC3C,QAAQ,CAAC,IAAI,CAAC,OAAO,MAAM,iCAAiC,CAAC,CAAC;YAC9D,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,UAAU,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;gBAC3C,MAAM,CAAC,GAAG,UAAU,CAAC,CAAC,CAAC,CAAC;gBACxB,QAAQ,CAAC,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC,IAAI,OAAO,CAAC,CAAC,CAAC,WAAW,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,mBAAmB,CAAC,CAAC,SAAS,GAAG,CAAC,CAAC;gBAC5G,QAAQ,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,WAAW,EAAE,CAAC,CAAC;YACvC,CAAC;QACH,CAAC;QAED,IAAI,QAAQ,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACxB,QAAQ,CAAC,IAAI,CAAC,4CAA4C,CAAC,CAAC;YAC5D,KAAK,MAAM,CAAC,IAAI,QAAQ,EAAE,CAAC;gBACzB,QAAQ,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,aAAa,KAAK,CAAC,CAAC,CAAC,QAAQ,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,8BAA8B,CAAC,CAAC,iBAAiB,CAAC,CAAC,CAAC,IAAI,SAAS,EAAE,CAAC,CAAC;gBACzI,QAAQ,CAAC,IAAI,CAAC,iDAAiD,CAAC,CAAC;YACnE,CAAC;QACH,CAAC;QAED,OAAO,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAC7B,CAAC;IAEO,WAAW,CAAC,WAAwB;QAC1C,MAAM,OAAO,GAAG,IAAI,CAAC,SAAS,CAAC,WAAW,CAAC,OAAO,CAAC,CAAC,WAAW,EAAE,CAAC;QAClE,IAAI,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAC;YAAE,OAAO,SAAS,CAAC;QAC/C,IAAI,OAAO,CAAC,QAAQ,CAAC,UAAU,CAAC,IAAI,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAC;YAAE,OAAO,UAAU,CAAC;QAChF,IAAI,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAC,IAAI,OAAO,CAAC,QAAQ,CAAC,IAAI,CAAC;YAAE,OAAO,QAAQ,CAAC;QAC1E,IAAI,OAAO,CAAC,QAAQ,CAAC,UAAU,CAAC,IAAI,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAC;YAAE,OAAO,UAAU,CAAC;QAC/E,IAAI,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAC,IAAI,OAAO,CAAC,QAAQ,CAAC,UAAU,CAAC;YAAE,OAAO,KAAK,CAAC;QAC1E,IAAI,WAAW,CAAC,MAAM,KAAK,YAAY;YAAE,OAAO,YAAY,CAAC;QAC7D,OAAO,SAAS,CAAC;IACnB,CAAC;IAEO,uBAAuB,CAAC,QAAkB;QAChD,MAAM,SAAS,GAAG,QAAQ,CAAC,SAAS,CAAC,WAAW,EAAE,CAAC;QACnD,IAAI,SAAS,CAAC,QAAQ,CAAC,MAAM,CAAC;YAAE,OAAO,SAAS,CAAC;QACjD,IAAI,SAAS,CAAC,QAAQ,CAAC,UAAU,CAAC,IAAI,SAAS,CAAC,QAAQ,CAAC,MAAM,CAAC;YAAE,OAAO,UAAU,CAAC;QACpF,IAAI,SAAS,CAAC,QAAQ,CAAC,QAAQ,CAAC,IAAI,SAAS,CAAC,QAAQ,CAAC,IAAI,CAAC;YAAE,OAAO,QAAQ,CAAC;QAC9E,IAAI,SAAS,CAAC,QAAQ,CAAC,UAAU,CAAC,IAAI,SAAS,CAAC,QAAQ,CAAC,KAAK,CAAC;YAAE,OAAO,UAAU,CAAC;QACnF,IAAI,SAAS,CAAC,QAAQ,CAAC,KAAK,CAAC,IAAI,SAAS,CAAC,QAAQ,CAAC,UAAU,CAAC;YAAE,OAAO,KAAK,CAAC;QAC9E,OAAO,SAAS,CAAC;IACnB,CAAC;CACF"}

package/dist/runtime/gap-detector.d.ts

@@ -0,0 +1,6 @@
+import type { VerificationHotspot, RuleCandidate } from './types.js';
+export declare class GapDetector {
+    analyze(hotspots: VerificationHotspot[]): Promise<RuleCandidate[]>;
+    private proposeRule;
+    private validateTestCases;
+}

package/dist/runtime/gap-detector.js

@@ -0,0 +1,111 @@
+// ============================================================
+// Assay Verified Agent Runtime — Gap Detector
+// Identifies verification hotspots where formal rules could
+// replace LLM verification, and proposes rule candidates.
+// ============================================================
+import { randomUUID } from 'node:crypto';
+import { getClient, MODEL } from '../lib/anthropic.js';
+export class GapDetector {
+    async analyze(hotspots) {
+        const candidates = [];
+        const eligibleHotspots = hotspots.filter(h => h.candidateForFormalRule &&
+            h.failRate > 0.2 &&
+            h.formalCoverage < 0.1 &&
+            h.totalClaims >= 10);
+        for (const hotspot of eligibleHotspots) {
+            try {
+                const candidate = await this.proposeRule(hotspot);
+                if (candidate && this.validateTestCases(candidate)) {
+                    candidates.push(candidate);
+                }
+            }
+            catch {
+                // Skip failed proposals
+            }
+        }
+        return candidates;
+    }
+    async proposeRule(hotspot) {
+        const client = getClient();
+        const response = await client.messages.create({
+            model: MODEL,
+            max_tokens: 2048,
+            system: `You design formal verification rules for code. Rules must be regex or AST-based patterns that can be checked deterministically without an LLM.
+
+Respond with ONLY JSON, no markdown fences.`,
+            messages: [{
+                    role: 'user',
+                    content: `This verification hotspot has a ${(hotspot.failRate * 100).toFixed(0)}% failure rate but only ${(hotspot.formalCoverage * 100).toFixed(0)}% formal verification coverage:
+
+Category: ${hotspot.claimCategory}
+Total claims: ${hotspot.totalClaims}
+Failed claims: ${hotspot.failedClaims}
+Top failure reasons: ${hotspot.topFailureReasons.join('; ')}
+
+Propose a formal rule (regex pattern) that could catch the most common failure pattern in this category.
+
+Respond:
+{
+  "description": "what the rule checks",
+  "rationale": "why this should be formal, not LLM",
+  "substrate": "regex",
+  "pattern": "the regex pattern",
+  "language": "typescript",
+  "testCases": [
+    { "input": "code that should FAIL", "language": "typescript", "expectedVerdict": "FAIL", "description": "what this tests" },
+    { "input": "code that should PASS", "language": "typescript", "expectedVerdict": "PASS", "description": "what this tests" }
+  ],
+  "estimatedFormalCoverage": 0.3
+}`,
+                }],
+        });
+        const text = response.content[0].type === 'text' ? response.content[0].text : '';
+        try {
+            let cleaned = text.trim();
+            if (cleaned.startsWith('```')) {
+                cleaned = cleaned.replace(/^```(?:json)?\s*/, '').replace(/\s*```$/, '');
+            }
+            const parsed = JSON.parse(cleaned);
+            return {
+                id: randomUUID(),
+                claimCategory: hotspot.claimCategory,
+                description: parsed.description,
+                rationale: parsed.rationale,
+                substrate: parsed.substrate === 'ast' ? 'ast' : 'regex',
+                pattern: parsed.pattern,
+                language: parsed.language ?? 'typescript',
+                testCases: Array.isArray(parsed.testCases) ? parsed.testCases : [],
+                sourceHotspot: hotspot.claimCategory,
+                supportingExperiences: [],
+                estimatedFormalCoverage: parsed.estimatedFormalCoverage ?? 0,
+                status: 'proposed',
+                proposedAt: new Date().toISOString(),
+            };
+        }
+        catch {
+            return null;
+        }
+    }
+    validateTestCases(candidate) {
+        if (candidate.testCases.length < 2)
+            return false;
+        if (candidate.substrate !== 'regex')
+            return true; // can't validate AST locally
+        try {
+            const regex = new RegExp(candidate.pattern, 'gm');
+            for (const tc of candidate.testCases) {
+                const matches = regex.test(tc.input);
+                regex.lastIndex = 0; // reset stateful regex
+                const formalVerdict = matches ? 'FAIL' : 'PASS';
+                if (formalVerdict !== tc.expectedVerdict) {
+                    return false; // test case doesn't match expected behavior
+                }
+            }
+            return true;
+        }
+        catch {
+            return false; // invalid regex
+        }
+    }
+}
+//# sourceMappingURL=gap-detector.js.map