agentic-qe 3.11.5 → 3.12.1

package/dist/validation/quality-gate-runner.d.ts

@@ -0,0 +1,49 @@
+/**
+ * Quality-gate orchestrator (ADR-119 / ADR-120).
+ *
+ * The ONE code path both the `aqe quality-gate` CLI subcommand and the
+ * `qe/quality/gate` MCP tool call, so the two surfaces stay in exact parity
+ * (the CLI/MCP divergence CLAUDE.md warns about). It:
+ *
+ *   1. loads the pinned, hash-checked ADR-117 anchor via `loadAnchorSet` (which
+ *      throws on drift — the frozen checklist can't be silently edited),
+ *   2. selects the constant-denominator checklist for `checklistId`,
+ *   3. runs the two-gate, three-valued `computeQualityVerdict` with the injected
+ *      frontier judge.
+ *
+ * The oracle result and the judge are injected — this module owns no LLM and no
+ * mutation runner, so it is deterministic and unit-testable with a fake judge.
+ */
+import { type Judge, type QualityVerdictResult, type RequirementChecklist } from './quality-verdict.js';
+import type { OracleResult } from './oracle-eval.js';
+/** Default frozen anchor location, relative to the project root (ADR-117). */
+export declare const DEFAULT_ANCHOR_RELATIVE_PATH = "verification/anchors/qe-anchor-v1.json";
+export interface QualityGateRequest {
+    /**
+     * Mechanical-gate result from the ADR-113 oracle. `null` ⇒ the oracle did not
+     * run at all — treated as a non-executed test ⇒ mechanical fail.
+     */
+    oracleResult: Pick<OracleResult, 'passed' | 'baselinePassed'> | null;
+    /** The artifact under judgement (produced test source / spec output). */
+    artifact: string;
+    /** Anchor item id whose pinned checklist to grade against (e.g. "A1-inRange"). */
+    checklistId: string;
+    /** The frontier judge (injected — ALWAYS frontier-tier, ADR-111). */
+    judge: Judge;
+    /** Override the frozen anchor path (absolute, or relative to project root). */
+    anchorPath?: string;
+}
+/**
+ * Run the two-gate quality gate for one artifact against one pinned checklist.
+ * Throws only for caller error (unknown checklist id, unloadable/drifted anchor);
+ * every judgement outcome is returned as a three-valued `QualityVerdictResult`.
+ */
+export declare function runQualityGate(req: QualityGateRequest): Promise<QualityVerdictResult>;
+/**
+ * Load the pinned checklist for `checklistId` from the frozen anchor. Exposed so
+ * callers (CLI/MCP) can list valid ids or pre-validate before building a judge.
+ */
+export declare function loadChecklist(checklistId: string, anchorPath?: string): RequirementChecklist;
+/** List every checklist id in the frozen anchor (for CLI/MCP discoverability). */
+export declare function listChecklistIds(anchorPath?: string): string[];
+//# sourceMappingURL=quality-gate-runner.d.ts.map

package/dist/validation/quality-gate-runner.js

@@ -0,0 +1,62 @@
+/**
+ * Quality-gate orchestrator (ADR-119 / ADR-120).
+ *
+ * The ONE code path both the `aqe quality-gate` CLI subcommand and the
+ * `qe/quality/gate` MCP tool call, so the two surfaces stay in exact parity
+ * (the CLI/MCP divergence CLAUDE.md warns about). It:
+ *
+ *   1. loads the pinned, hash-checked ADR-117 anchor via `loadAnchorSet` (which
+ *      throws on drift — the frozen checklist can't be silently edited),
+ *   2. selects the constant-denominator checklist for `checklistId`,
+ *   3. runs the two-gate, three-valued `computeQualityVerdict` with the injected
+ *      frontier judge.
+ *
+ * The oracle result and the judge are injected — this module owns no LLM and no
+ * mutation runner, so it is deterministic and unit-testable with a fake judge.
+ */
+import path from 'node:path';
+import { loadAnchorSet } from './anchor-set.js';
+import { computeQualityVerdict, } from './quality-verdict.js';
+import { findProjectRoot } from '../kernel/unified-memory.js';
+/** Default frozen anchor location, relative to the project root (ADR-117). */
+export const DEFAULT_ANCHOR_RELATIVE_PATH = 'verification/anchors/qe-anchor-v1.json';
+/**
+ * Run the two-gate quality gate for one artifact against one pinned checklist.
+ * Throws only for caller error (unknown checklist id, unloadable/drifted anchor);
+ * every judgement outcome is returned as a three-valued `QualityVerdictResult`.
+ */
+export async function runQualityGate(req) {
+    const checklist = loadChecklist(req.checklistId, req.anchorPath);
+    return computeQualityVerdict({
+        oracle: req.oracleResult,
+        artifact: req.artifact,
+        checklist,
+        judge: req.judge,
+    });
+}
+/**
+ * Load the pinned checklist for `checklistId` from the frozen anchor. Exposed so
+ * callers (CLI/MCP) can list valid ids or pre-validate before building a judge.
+ */
+export function loadChecklist(checklistId, anchorPath) {
+    const resolved = resolveAnchorPath(anchorPath);
+    const anchor = loadAnchorSet(resolved); // throws on hash drift (ADR-117 §2)
+    const item = anchor.items.find((it) => it.id === checklistId);
+    if (!item) {
+        const available = anchor.items.map((it) => it.id).join(', ');
+        throw new Error(`Unknown checklist id "${checklistId}" in anchor ${resolved}. Available: ${available}`);
+    }
+    return { id: item.id, requirements: item.requirements };
+}
+/** List every checklist id in the frozen anchor (for CLI/MCP discoverability). */
+export function listChecklistIds(anchorPath) {
+    const anchor = loadAnchorSet(resolveAnchorPath(anchorPath));
+    return anchor.items.map((it) => it.id);
+}
+function resolveAnchorPath(anchorPath) {
+    if (anchorPath && path.isAbsolute(anchorPath))
+        return anchorPath;
+    const root = findProjectRoot();
+    return path.resolve(root, anchorPath ?? DEFAULT_ANCHOR_RELATIVE_PATH);
+}
+//# sourceMappingURL=quality-gate-runner.js.map

package/dist/validation/quality-verdict.d.ts

@@ -0,0 +1,94 @@
+/**
+ * Two-gate, three-valued quality verdict (ADR-119).
+ *
+ * A single, two-valued quality verdict lies in three ways: it can pass tests
+ * that never executed, a single LLM judge is noisy (retort measured Haiku
+ * swinging 0.33↔1.0 on identical code), and pass/fail can't tell a genuine gap
+ * from an inconclusive run. This composes retort's design:
+ *
+ *   1. Mechanical gate — reuse the ADR-113 oracle. Tests that did not execute
+ *      ⇒ `fail`. No judge is consulted until the mechanical gate is satisfied.
+ *   2. Spec gate — a FRONTIER-tier judge (never the cheap writer lane, ADR-111)
+ *      scores the artifact against the ADR-117 pinned, constant-denominator
+ *      checklist; two-attempt second opinion; pass only at coverage 1.0.
+ *   3. Judge preflight — refuse to report success if the judge never ran.
+ *
+ * Verdict ∈ {pass, fail, inconclusive}. `inconclusive` (retort's `None`) keeps
+ * an infra hiccup from masquerading as a spec failure — ADR-118's accept()
+ * records neither a promotion nor a rejection on `inconclusive`.
+ *
+ * The judge and checklist are injected so this logic is testable without a live
+ * frontier model or the frozen anchor file (which is authored under ADR-117).
+ */
+import type { OracleResult } from './oracle-eval.js';
+export type QualityVerdict = 'pass' | 'fail' | 'inconclusive';
+/**
+ * A pinned, constant-denominator requirement checklist (ADR-117 anchor).
+ * The denominator (requirement count) is authored ONCE per artifact type so
+ * coverage scores are comparable across runs — retort's fix for non-comparable
+ * per-run extraction.
+ */
+export interface RequirementChecklist {
+    /** Stable id of the pinned checklist (e.g. the ADR-117 anchor id). */
+    id: string;
+    /** The requirements — the constant denominator. Must be non-empty. */
+    requirements: string[];
+}
+/** One judge attempt's opinion. */
+export interface JudgeOpinion {
+    /**
+     * Whether the judge tooling actually ran and produced a real grade. `false`
+     * means usage-limit / timeout / tooling failure — NOT a real opinion, so it
+     * can never contribute to a `fail`.
+     */
+    ran: boolean;
+    /** Fraction of checklist requirements satisfied, in [0, 1]. */
+    coverage: number;
+    /** Requirements judged unmet (for reporting; empty when coverage is 1.0). */
+    unmet: string[];
+}
+/**
+ * The frontier judge. ALWAYS top-tier (ADR-111 — the oracle is the one place
+ * you never economize). Implementations wrap the frontier model + credentials.
+ */
+export interface Judge {
+    /** ADR-119 §4: verify the judge tooling/credentials before grading. */
+    preflight(): Promise<boolean> | boolean;
+    /** Grade the artifact against the pinned checklist. */
+    grade(artifact: string, checklist: RequirementChecklist): Promise<JudgeOpinion> | JudgeOpinion;
+}
+export interface QualityVerdictInput {
+    /**
+     * Mechanical-gate result from the ADR-113 oracle. `null` means the oracle did
+     * not run at all — treated as a non-executed test ⇒ mechanical fail.
+     */
+    oracle: Pick<OracleResult, 'passed' | 'baselinePassed'> | null;
+    /** The artifact under judgement (e.g. produced test source / spec output). */
+    artifact: string;
+    /** The pinned checklist to grade against. */
+    checklist: RequirementChecklist;
+    /** The frontier judge (injected). */
+    judge: Judge;
+}
+export interface QualityVerdictResult {
+    verdict: QualityVerdict;
+    /** Mechanical gate outcome (execution). */
+    mechanical: 'pass' | 'fail';
+    /** Best spec-gate coverage observed across attempts (null if judge never ran). */
+    specCoverage: number | null;
+    /** Unmet requirements from the best real attempt (empty on pass). */
+    unmet: string[];
+    /** Number of spec-gate attempts actually made. */
+    attempts: number;
+    reason: string;
+}
+/**
+ * Compute the two-gate, three-valued quality verdict.
+ *
+ * Order is strict: mechanical gate first (a non-executing test is `fail`
+ * regardless of judge opinion), then judge preflight, then the two-attempt spec
+ * gate. `fail` requires TWO real short opinions; fewer than two real opinions ⇒
+ * `inconclusive` (never a silent pass, never a spec-fail from infra).
+ */
+export declare function computeQualityVerdict(input: QualityVerdictInput): Promise<QualityVerdictResult>;
+//# sourceMappingURL=quality-verdict.d.ts.map

package/dist/validation/quality-verdict.js

@@ -0,0 +1,120 @@
+/**
+ * Two-gate, three-valued quality verdict (ADR-119).
+ *
+ * A single, two-valued quality verdict lies in three ways: it can pass tests
+ * that never executed, a single LLM judge is noisy (retort measured Haiku
+ * swinging 0.33↔1.0 on identical code), and pass/fail can't tell a genuine gap
+ * from an inconclusive run. This composes retort's design:
+ *
+ *   1. Mechanical gate — reuse the ADR-113 oracle. Tests that did not execute
+ *      ⇒ `fail`. No judge is consulted until the mechanical gate is satisfied.
+ *   2. Spec gate — a FRONTIER-tier judge (never the cheap writer lane, ADR-111)
+ *      scores the artifact against the ADR-117 pinned, constant-denominator
+ *      checklist; two-attempt second opinion; pass only at coverage 1.0.
+ *   3. Judge preflight — refuse to report success if the judge never ran.
+ *
+ * Verdict ∈ {pass, fail, inconclusive}. `inconclusive` (retort's `None`) keeps
+ * an infra hiccup from masquerading as a spec failure — ADR-118's accept()
+ * records neither a promotion nor a rejection on `inconclusive`.
+ *
+ * The judge and checklist are injected so this logic is testable without a live
+ * frontier model or the frozen anchor file (which is authored under ADR-117).
+ */
+/** Coverage at or above this is a spec-gate pass (pinned checklist ⇒ pass only at 1.0). */
+const PASS_COVERAGE = 1.0;
+/** ADR-119 §3: at most two attempts (second opinion). */
+const MAX_ATTEMPTS = 2;
+/**
+ * Compute the two-gate, three-valued quality verdict.
+ *
+ * Order is strict: mechanical gate first (a non-executing test is `fail`
+ * regardless of judge opinion), then judge preflight, then the two-attempt spec
+ * gate. `fail` requires TWO real short opinions; fewer than two real opinions ⇒
+ * `inconclusive` (never a silent pass, never a spec-fail from infra).
+ */
+export async function computeQualityVerdict(input) {
+    if (input.checklist.requirements.length === 0) {
+        throw new Error('RequirementChecklist must have a non-empty, constant denominator (ADR-117).');
+    }
+    // 1. Mechanical gate (ADR-113). Non-execution ⇒ fail, no judge consulted.
+    const mechanicalPass = input.oracle != null && input.oracle.baselinePassed && input.oracle.passed;
+    if (!mechanicalPass) {
+        const why = input.oracle == null
+            ? 'oracle did not run'
+            : !input.oracle.baselinePassed
+                ? 'tests did not execute against the reference implementation'
+                : 'mutation score below threshold';
+        return {
+            verdict: 'fail',
+            mechanical: 'fail',
+            specCoverage: null,
+            unmet: [],
+            attempts: 0,
+            reason: `mechanical gate failed: ${why}`,
+        };
+    }
+    // 2. Judge preflight (ADR-119 §4). If the judge tooling can't run, the verdict
+    // is inconclusive — never a silent pass.
+    const ready = await input.judge.preflight();
+    if (!ready) {
+        return {
+            verdict: 'inconclusive',
+            mechanical: 'pass',
+            specCoverage: null,
+            unmet: [],
+            attempts: 0,
+            reason: 'judge preflight failed: judge tooling did not run — refusing to report success',
+        };
+    }
+    // 3. Two-attempt spec gate (retort protocol). Pass on the first attempt that
+    // reaches 1.0; otherwise collect real opinions.
+    const realOpinions = [];
+    let bestCoverage = null;
+    let bestUnmet = [];
+    let attempts = 0;
+    for (let i = 0; i < MAX_ATTEMPTS; i++) {
+        attempts++;
+        const op = await input.judge.grade(input.artifact, input.checklist);
+        if (!op.ran) {
+            continue; // not a real opinion — cannot contribute to a fail
+        }
+        realOpinions.push(op);
+        if (bestCoverage == null || op.coverage > bestCoverage) {
+            bestCoverage = op.coverage;
+            bestUnmet = op.unmet;
+        }
+        if (op.coverage >= PASS_COVERAGE) {
+            return {
+                verdict: 'pass',
+                mechanical: 'pass',
+                specCoverage: op.coverage,
+                unmet: [],
+                attempts,
+                reason: `spec gate passed at coverage ${op.coverage.toFixed(3)} on attempt ${attempts}`,
+            };
+        }
+    }
+    // No attempt reached 1.0. Fail ONLY on two real short opinions; otherwise
+    // inconclusive (fewer than two real opinions = usage limit / timeout).
+    if (realOpinions.length >= 2) {
+        return {
+            verdict: 'fail',
+            mechanical: 'pass',
+            specCoverage: bestCoverage,
+            unmet: bestUnmet,
+            attempts,
+            reason: `spec gate failed: ${realOpinions.length} real opinions, best coverage `
+                + `${(bestCoverage ?? 0).toFixed(3)} < ${PASS_COVERAGE} (${bestUnmet.length} unmet)`,
+        };
+    }
+    return {
+        verdict: 'inconclusive',
+        mechanical: 'pass',
+        specCoverage: bestCoverage,
+        unmet: bestUnmet,
+        attempts,
+        reason: `spec gate inconclusive: only ${realOpinions.length} real opinion(s) obtainable `
+            + `(need 2 to fail) — retry later, record neither pass nor fail`,
+    };
+}
+//# sourceMappingURL=quality-verdict.js.map

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "agentic-qe",
-  "version": "3.11.5",
+  "version": "3.12.1",
   "description": "Agentic Quality Engineering V3 - Domain-Driven Design Architecture with 13 Bounded Contexts, O(log n) coverage analysis, ReasoningBank learning, 60 specialized QE agents, mathematical Coherence verification, deep Claude Flow integration",
   "type": "module",
   "main": "./dist/index.js",

package/dist/cli/chunks/adapter-KNM5R47J.js

@@ -1,2 +0,0 @@
-import{createRequire as __cr}from"module";const require=__cr(import.meta.url);if(process.argv.includes('--version')||process.argv.includes('-v')){console.log("3.11.5");process.exit(0)}
-import{a,b,c}from"./chunk-7P6X4BFO.js";import"./chunk-3WR35GFF.js";import"./chunk-TEIWJ3GJ.js";import"./chunk-CKOF27L7.js";import"./chunk-RKXZLTWR.js";import"./chunk-C4EZNOWK.js";export{a as AgentBoosterAdapter,b as createAgentBoosterAdapter,c as createAgentBoosterAdapterSync};

package/dist/cli/chunks/aqe-learning-engine-MXPJLVR3.js

@@ -1,2 +0,0 @@
-import{createRequire as __cr}from"module";const require=__cr(import.meta.url);if(process.argv.includes('--version')||process.argv.includes('-v')){console.log("3.11.5");process.exit(0)}
-import{a,b,c,d,e,f}from"./chunk-Y6DWJKHU.js";import"./chunk-J6PFPY7V.js";import"./chunk-EEKG35K5.js";import"./chunk-UORTQXJ2.js";import"./chunk-4XJ2A3HY.js";import"./chunk-PW72M2FE.js";import"./chunk-UBI2EOS3.js";import"./chunk-MCHEUSY4.js";import"./chunk-YGVDP6X7.js";import"./chunk-JBKZ5TUM.js";import"./chunk-W7TYKPLY.js";import"./chunk-IBDQWKI3.js";import"./chunk-53YOYZW5.js";import"./chunk-TRNBQFFG.js";import"./chunk-LPR2R3AN.js";import"./chunk-NCSXLMDL.js";import"./chunk-IU5RSN6M.js";import"./chunk-ZMPAM4DY.js";import"./chunk-Y25PO65W.js";import"./chunk-YFCULQBP.js";import"./chunk-CKOF27L7.js";import"./chunk-4QOI5JS7.js";import"./chunk-RJN2RM76.js";import"./chunk-OS62Y77J.js";import"./chunk-TTJJ4DQE.js";import"./chunk-5WMU7XCF.js";import"./chunk-RKXZLTWR.js";import"./chunk-K4I7UIU5.js";import"./chunk-FHTYEAQ3.js";import"./chunk-WHHQIL4F.js";import"./chunk-3JF5A72M.js";import"./chunk-CDJP6COA.js";import"./chunk-FIPKIIMS.js";import"./chunk-2XQHG56O.js";import"./chunk-2XPMATEK.js";import"./chunk-JDSHUECG.js";import"./chunk-GODAWJMZ.js";import"./chunk-C4EZNOWK.js";export{b as AQELearningEngine,a as DEFAULT_ENGINE_CONFIG,c as DEFAULT_META_LEARNING_CONFIG,d as MetaLearningEngine,e as createAQELearningEngine,f as createDefaultLearningEngine};

package/dist/cli/chunks/base-X4EXQ2AS.js

@@ -1,2 +0,0 @@
-import{createRequire as __cr}from"module";const require=__cr(import.meta.url);if(process.argv.includes('--version')||process.argv.includes('-v')){console.log("3.11.5");process.exit(0)}
-import{a,b,c,d,e,f,g,h,i,j,k,l}from"./chunk-6T65OKUM.js";import"./chunk-I3NWKPXJ.js";import"./chunk-NCSXLMDL.js";import"./chunk-IU5RSN6M.js";import"./chunk-ZMPAM4DY.js";import"./chunk-Y25PO65W.js";import"./chunk-YFCULQBP.js";import"./chunk-CKOF27L7.js";import"./chunk-TTJJ4DQE.js";import"./chunk-RKXZLTWR.js";import"./chunk-K4I7UIU5.js";import"./chunk-SUL3I6QZ.js";import"./chunk-FHTYEAQ3.js";import"./chunk-WHHQIL4F.js";import"./chunk-3JF5A72M.js";import"./chunk-CDJP6COA.js";import"./chunk-FIPKIIMS.js";import"./chunk-2XQHG56O.js";import"./chunk-2XPMATEK.js";import"./chunk-JDSHUECG.js";import"./chunk-GODAWJMZ.js";import"./chunk-C4EZNOWK.js";export{l as MCPToolBase,k as defaultToolLogger,j as getKernel,i as getLLMRouter,e as getMemoryBackend,g as getSharedLLMRouter,c as getSharedMemoryBackend,a as isMemoryOnlyMode,b as registerRvfResetFn,h as resetSharedLLMRouter,d as resetSharedMemoryBackend,f as setSharedLLMRouter};

package/dist/cli/chunks/browser-workflow-4HX6RP5F.js

@@ -1,2 +0,0 @@
-import{createRequire as __cr}from"module";const require=__cr(import.meta.url);if(process.argv.includes('--version')||process.argv.includes('-v')){console.log("3.11.5");process.exit(0)}
-import{a}from"./chunk-KVVJVMDM.js";import"./chunk-6T65OKUM.js";import"./chunk-I3NWKPXJ.js";import"./chunk-NCSXLMDL.js";import"./chunk-IU5RSN6M.js";import"./chunk-ZMPAM4DY.js";import"./chunk-Y25PO65W.js";import"./chunk-YFCULQBP.js";import"./chunk-CKOF27L7.js";import"./chunk-TTJJ4DQE.js";import"./chunk-RKXZLTWR.js";import"./chunk-K4I7UIU5.js";import"./chunk-SUL3I6QZ.js";import"./chunk-FHTYEAQ3.js";import"./chunk-WHHQIL4F.js";import"./chunk-3JF5A72M.js";import"./chunk-CDJP6COA.js";import"./chunk-FIPKIIMS.js";import"./chunk-2XQHG56O.js";import"./chunk-2XPMATEK.js";import"./chunk-JDSHUECG.js";import"./chunk-GODAWJMZ.js";import"./chunk-C4EZNOWK.js";export{a as BrowserWorkflowTool};

package/dist/cli/chunks/chunk-FHTYEAQ3.js

@@ -1,2 +0,0 @@
-import{createRequire as __cr}from"module";const require=__cr(import.meta.url);if(process.argv.includes('--version')||process.argv.includes('-v')){console.log("3.11.5");process.exit(0)}
-import{a as t,q as r}from"./chunk-WHHQIL4F.js";import{c as l}from"./chunk-C4EZNOWK.js";function m(s,e){let o=e?.readonly??!1,n=e?.fileMustExist??!1,i=e?.busyTimeout??5e3,b=e?.walMode??!o,a=new t(s,{readonly:o,fileMustExist:n});return a.pragma(`busy_timeout = ${i}`),b&&a.pragma("journal_mode = WAL"),a}var u=l(()=>{r()});export{m as a,u as b};

package/dist/cli/chunks/chunk-MIU5RUBJ.js

@@ -1,149 +0,0 @@
-import{createRequire as __cr}from"module";const require=__cr(import.meta.url);if(process.argv.includes('--version')||process.argv.includes('-v')){console.log("3.11.5");process.exit(0)}
-import{a as _}from"./chunk-AGJOGUP5.js";import{e as E}from"./chunk-MCHEUSY4.js";import{a as b}from"./chunk-ZMPAM4DY.js";import{c as h}from"./chunk-Y25PO65W.js";import{b as p,d as S}from"./chunk-CKOF27L7.js";import{S as A,j as g}from"./chunk-GODAWJMZ.js";S();b();A();var d=h.create("pattern-lifecycle"),D={promotionRewardThreshold:.7,promotionMinOccurrences:E,promotionMinSuccessRate:.7,deprecationFailureThreshold:3,staleDaysThreshold:30,confidenceDecayRate:.01,minActiveConfidence:.3,maxAgeForActivePatterns:90,promotionActivityWindowDays:30,asymmetricLearning:{}},y=5,L=.9,R=2;function v(u,e){return e<y||!Number.isFinite(u)?0:Math.min(Math.max(u,0),1)}var f=class{constructor(e,t={}){this.db=e;this.config={...D,...t},this.asymmetricEngine=new _(this.config.asymmetricLearning),this.ensureSchema()}db;config;asymmetricEngine;_witnessChain=null;set witnessChain(e){this._witnessChain=e}_compressionService=null;set compressionService(e){this._compressionService=e}ensureSchema(){try{this.db.prepare(`
-        SELECT deprecated_at FROM qe_patterns LIMIT 1
-      `).get()}catch(e){d.debug("Adding missing deprecated_at column",{error:e instanceof Error?e.message:String(e)}),this.db.exec(`
-        ALTER TABLE qe_patterns ADD COLUMN deprecated_at TEXT DEFAULT NULL
-      `),console.log("[PatternLifecycle] Added deprecated_at column to qe_patterns")}try{this.db.prepare(`
-        SELECT consecutive_failures FROM qe_patterns LIMIT 1
-      `).get()}catch(e){d.debug("Adding missing consecutive_failures column",{error:e instanceof Error?e.message:String(e)}),this.db.exec(`
-        ALTER TABLE qe_patterns ADD COLUMN consecutive_failures INTEGER DEFAULT 0
-      `),console.log("[PatternLifecycle] Added consecutive_failures column to qe_patterns")}try{this.db.prepare(`
-        SELECT promotion_date FROM qe_patterns LIMIT 1
-      `).get()}catch(e){d.debug("Adding missing promotion_date column",{error:e instanceof Error?e.message:String(e)}),this.db.exec(`
-        ALTER TABLE qe_patterns ADD COLUMN promotion_date TEXT DEFAULT NULL
-      `),console.log("[PatternLifecycle] Added promotion_date column to qe_patterns")}}getRecentExperiences(e={}){let t=e.minReward??this.config.promotionRewardThreshold,n=e.limit??100,r=e.sinceDays??7;if(!this.db.prepare("SELECT name FROM sqlite_master WHERE type='table' AND name='captured_experiences'").get())return d.debug("captured_experiences table not found"),[];let a=new Date(Date.now()-r*864e5).toISOString().replace("T"," ").slice(0,19);return this.db.prepare(`
-      SELECT
-        domain as task_type,
-        COUNT(*) as count,
-        AVG(quality) as avg_reward,
-        MAX(quality) as max_reward,
-        MIN(quality) as min_reward,
-        SUM(CASE WHEN quality >= ? THEN 1 ELSE 0 END) as success_count,
-        GROUP_CONCAT(DISTINCT agent) as actions,
-        MAX(started_at) as latest_at
-      FROM captured_experiences
-      WHERE started_at >= ?
-        AND quality >= ?
-        AND agent != 'cli-hook'
-      GROUP BY domain
-      HAVING COUNT(*) >= ?
-      ORDER BY avg_reward DESC
-      LIMIT ?
-    `).all(t,a,t*.5,this.config.promotionMinOccurrences,n).map(i=>({taskType:i.task_type,domain:this.taskTypeToQEDomain(i.task_type),count:i.count,avgReward:i.avg_reward,maxReward:i.max_reward,minReward:i.min_reward,actions:i.actions?i.actions.split(",").filter(Boolean):[],successCount:i.success_count,latestAt:new Date(i.latest_at)}))}taskTypeToQEDomain(e){let t={generate:"test-generation","test-generation":"test-generation",test:"test-generation",analyze:"coverage-analysis",coverage:"coverage-analysis","coverage-analysis":"coverage-analysis",run:"test-execution",execute:"test-execution","test-execution":"test-execution",report:"quality-assessment",quality:"quality-assessment",assessment:"quality-assessment",security:"security-compliance",sast:"security-compliance",audit:"security-compliance",defect:"defect-intelligence",predict:"defect-intelligence",bug:"defect-intelligence",requirements:"requirements-validation",validation:"requirements-validation",code:"code-intelligence",complexity:"code-intelligence",contract:"contract-testing",api:"contract-testing",visual:"visual-accessibility",a11y:"visual-accessibility",accessibility:"visual-accessibility",chaos:"chaos-resilience",resilience:"chaos-resilience",flaky:"chaos-resilience",learning:"learning-optimization",optimize:"learning-optimization"},n=e.toLowerCase();for(let[r,s]of Object.entries(t))if(n.includes(r))return s;return null}taskTypeToPatternType(e){let t={generate:"test-template",test:"test-template",coverage:"coverage-strategy",analyze:"coverage-strategy",mock:"mock-pattern",stub:"mock-pattern",assert:"assertion-pattern",expect:"assertion-pattern",security:"assertion-pattern",contract:"api-contract",api:"api-contract",visual:"visual-baseline",screenshot:"visual-baseline",a11y:"a11y-check",accessibility:"a11y-check",perf:"perf-benchmark",load:"perf-benchmark",stress:"perf-benchmark",flaky:"flaky-fix",retry:"flaky-fix",refactor:"refactor-safe",error:"error-handling",exception:"error-handling"},n=e.toLowerCase();for(let[r,s]of Object.entries(t))if(n.includes(r))return s;return"test-template"}findPatternCandidates(e){let t=[];for(let n of e){let r=n.avgReward>=this.config.promotionRewardThreshold,s=n.count>=this.config.promotionMinOccurrences,a=n.successCount/n.count,c=a>=this.config.promotionMinSuccessRate;if(r&&s&&c){let i=n.domain||"code-intelligence",o=this.taskTypeToPatternType(n.taskType);if(this.findExistingPattern(n.taskType,i))continue;t.push({name:`${n.taskType}-success-pattern`,domain:i,patternType:o,confidence:Math.min(.95,n.avgReward*.9),sourceExperiences:n.count,avgReward:n.avgReward,successRate:a,actions:n.actions.slice(0,5),templateContent:this.generateTemplateContent(n)})}}return t}findExistingPattern(e,t){let n=this.db.prepare(`
-      SELECT * FROM qe_patterns
-      WHERE name LIKE ? AND qe_domain = ? AND deprecated_at IS NULL
-      LIMIT 1
-    `).get(`%${e}%`,t);return n?this.rowToPattern(n):null}generateTemplateContent(e){let t=e.actions.map((n,r)=>`${r+1}. ${n}`).join(`
-`);return`Task Type: ${e.taskType}
-Domain: ${e.domain||"general"}
-Average Reward: ${e.avgReward.toFixed(3)}
-Success Rate: ${(e.successCount/e.count*100).toFixed(1)}%
-
-Typical Actions:
-${t||"N/A"}
-
-Pattern extracted from ${e.count} successful experiences.`}checkPromotion(e){let t=this.getPattern(e);if(!t)return{shouldPromote:!1,meetsRewardThreshold:!1,meetsOccurrenceThreshold:!1,meetsSuccessRateThreshold:!1,meetsActivityWindow:!1,currentReward:0,currentOccurrences:0,currentSuccessRate:0};let n=t.qualityScore,r=n>=this.config.promotionRewardThreshold,s=t.usageCount>=this.config.promotionMinOccurrences,a=t.successRate>=this.config.promotionMinSuccessRate,c=this.config.promotionActivityWindowDays*24*60*60*1e3,i=t.lastUsedAt?.getTime()??t.createdAt.getTime(),o=Date.now()-i<c;return{shouldPromote:t.tier==="short-term"&&r&&s&&a&&o,meetsRewardThreshold:r,meetsOccurrenceThreshold:s,meetsSuccessRateThreshold:a,meetsActivityWindow:o,currentReward:n,currentOccurrences:t.usageCount,currentSuccessRate:t.successRate}}promotePattern(e){return this.checkPromotion(e).shouldPromote?(this.db.prepare(`
-      UPDATE qe_patterns
-      SET tier = 'long-term',
-          promotion_date = datetime('now'),
-          confidence = MIN(1.0, confidence + 0.1),
-          updated_at = datetime('now')
-      WHERE id = ?
-    `).run(e),console.log(`[PatternLifecycle] Promoted pattern ${e} to long-term`),!0):!1}promoteEligiblePatterns(){let e=this.db.prepare(`
-      SELECT id FROM qe_patterns
-      WHERE tier = 'short-term' AND deprecated_at IS NULL
-    `).all(),t=0;for(let{id:n}of e)this.promotePattern(n)&&t++;return{promoted:t,checked:e.length}}runPromotionSweep(){return this.promoteEligiblePatterns()}checkDeprecation(e){let t=this.getPattern(e);if(!t)return{shouldDeprecate:!1,reason:null};if(t.deprecated_at)return{shouldDeprecate:!1,reason:null};let n=t.consecutive_failures||0;if(n>=this.config.deprecationFailureThreshold)return{shouldDeprecate:!0,reason:"failures",consecutiveFailures:n};let r=t.lastUsedAt instanceof Date?t.lastUsedAt.getTime():new Date(t.lastUsedAt).getTime(),s=(Date.now()-r)/(1e3*60*60*24),a=v(t.successRate,t.usageCount),c=this.config.staleDaysThreshold*(1+a*R);if(s>=c)return{shouldDeprecate:!0,reason:"stale",daysSinceLastUse:s};if(t.confidence<this.config.minActiveConfidence)return{shouldDeprecate:!0,reason:"low_confidence",currentConfidence:t.confidence};let i=t.createdAt instanceof Date?t.createdAt.getTime():new Date(t.createdAt).getTime(),o=(Date.now()-i)/(1e3*60*60*24);return o>=this.config.maxAgeForActivePatterns&&t.usageCount<5?{shouldDeprecate:!0,reason:"age",ageInDays:o}:{shouldDeprecate:!1,reason:null}}deprecatePattern(e,t){let n=this.checkDeprecation(e);if(!n.shouldDeprecate&&!t)return!1;let r=t||n.reason||"manual";this.db.prepare(`
-      UPDATE qe_patterns
-      SET deprecated_at = datetime('now'),
-          updated_at = datetime('now')
-      WHERE id = ?
-    `).run(e);let s=this.getPattern(e);if(s){let a={deprecation_reason:r,deprecated_at:new Date().toISOString(),final_confidence:s.confidence,final_success_rate:s.successRate,total_usage:s.usageCount};this.db.prepare(`
-        UPDATE qe_patterns
-        SET context_json = json_patch(COALESCE(context_json, '{}'), ?)
-        WHERE id = ?
-      `).run(JSON.stringify({deprecation_metadata:a}),e)}return console.log(`[PatternLifecycle] Deprecated pattern ${e}: ${r}`),!0}deprecateStalePatterns(){let e=this.db.prepare(`
-      SELECT id FROM qe_patterns
-      WHERE deprecated_at IS NULL
-    `).all(),t=0;for(let{id:n}of e){let r=this.checkDeprecation(n);r.shouldDeprecate&&(this.deprecatePattern(n,r.reason||void 0),t++)}return{deprecated:t,checked:e.length}}applyConfidenceDecay(e=1){let t=this.config.confidenceDecayRate*e,n=Date.now()-e*24*60*60*1e3,r=this.db.prepare(`
-      UPDATE qe_patterns
-      SET confidence = MAX(?, confidence * (1 - ? * (1 - ? *
-            (CASE WHEN usage_count >= ?
-                  THEN MIN(MAX(COALESCE(success_rate, 0.0), 0.0), 1.0)
-                  ELSE 0.0 END)))),
-          updated_at = datetime('now')
-      WHERE deprecated_at IS NULL
-        AND (last_used_at IS NULL OR datetime(last_used_at) < datetime(?, 'unixepoch'))
-    `).run(this.config.minActiveConfidence,t,L,y,n/1e3);return{updated:r.changes,decayed:r.changes}}recordUsage(e,t){t?this.db.prepare(`
-        UPDATE qe_patterns
-        SET usage_count = usage_count + 1,
-            successful_uses = successful_uses + 1,
-            consecutive_failures = 0,
-            confidence = MIN(1.0, confidence + 0.02),
-            success_rate = CAST(successful_uses + 1 AS REAL) / CAST(usage_count + 1 AS REAL),
-            last_used_at = datetime('now'),
-            updated_at = datetime('now')
-        WHERE id = ?
-      `).run(e):this.db.prepare(`
-        UPDATE qe_patterns
-        SET usage_count = usage_count + 1,
-            consecutive_failures = consecutive_failures + 1,
-            confidence = MAX(0.1, confidence - 0.05),
-            success_rate = CAST(successful_uses AS REAL) / CAST(usage_count + 1 AS REAL),
-            last_used_at = datetime('now'),
-            updated_at = datetime('now')
-        WHERE id = ?
-      `).run(e),this.db.prepare(`
-      UPDATE qe_patterns
-      SET quality_score = confidence * 0.3 + (MIN(usage_count, 100) / 100.0) * 0.2 + success_rate * 0.5
-      WHERE id = ?
-    `).run(e)}recordAsymmetricOutcome(e,t,n){let r=this.getPattern(e);if(!r)return;let s=r.confidence,a=this.asymmetricEngine.computeConfidenceUpdate(s,t?"success":"failure",n);this.db.prepare(`
-      UPDATE qe_patterns
-      SET confidence = ?,
-          usage_count = usage_count + 1,
-          successful_uses = CASE WHEN ? THEN successful_uses + 1 ELSE successful_uses END,
-          consecutive_failures = CASE WHEN ? THEN 0 ELSE consecutive_failures + 1 END,
-          success_rate = CAST(CASE WHEN ? THEN successful_uses + 1 ELSE successful_uses END AS REAL) / CAST(usage_count + 1 AS REAL),
-          last_used_at = datetime('now'),
-          updated_at = datetime('now')
-      WHERE id = ?
-    `).run(a,t?1:0,t?1:0,t?1:0,e),this.asymmetricEngine.shouldQuarantine(a,n).shouldQuarantine&&this.quarantinePattern(e)}quarantinePattern(e){if(this.db.prepare(`
-      UPDATE qe_patterns
-      SET deprecated_at = datetime('now'),
-          updated_at = datetime('now')
-      WHERE id = ? AND deprecated_at IS NULL
-    `).run(e).changes>0){console.log(`[PatternLifecycle] Quarantined pattern ${e} (asymmetric confidence drop)`);try{this._witnessChain?.append("PATTERN_QUARANTINE",{patternId:e},"pattern-lifecycle")}catch{}return!0}return!1}checkRehabilitation(e,t){return this.asymmetricEngine.checkRehabilitation(t).canRehabilitate}rehabilitatePattern(e){return this.db.prepare(`
-      UPDATE qe_patterns
-      SET deprecated_at = NULL,
-          consecutive_failures = 0,
-          confidence = 0.5,
-          updated_at = datetime('now')
-      WHERE id = ?
-    `).run(e).changes>0?(console.log(`[PatternLifecycle] Rehabilitated pattern ${e} (10+ consecutive successes)`),!0):!1}getAsymmetricEngine(){return this.asymmetricEngine}compressPatternEmbedding(e){if(!g()||!this._compressionService)return null;let t=this.getPattern(e);if(!t||!t.embedding||t.embedding.length===0)return null;let n=t.lastUsedAt??t.createdAt,r=this._compressionService.classifyTier(n),s=new Float32Array(t.embedding),a=this._compressionService.compress(s,r);return d.debug("Compressed pattern embedding",{patternId:e,tier:r,ratio:(a.originalByteSize/a.compressedByteSize).toFixed(1)}),a}decompressPatternEmbedding(e){return this._compressionService?this._compressionService.decompress(e):null}compressAllPatternEmbeddings(){if(!g()||!this._compressionService)return 0;let e=this.db.prepare(`
-      SELECT id FROM qe_patterns
-      WHERE deprecated_at IS NULL
-    `).all(),t=0;for(let{id:n}of e)this.compressPatternEmbedding(n)&&t++;return d.info("Bulk pattern embedding compression complete",{total:e.length,compressed:t}),t}getStats(){let e=this.db.prepare(`
-      SELECT
-        COUNT(*) as total,
-        SUM(CASE WHEN deprecated_at IS NULL THEN 1 ELSE 0 END) as active,
-        SUM(CASE WHEN deprecated_at IS NOT NULL THEN 1 ELSE 0 END) as deprecated,
-        SUM(CASE WHEN tier = 'long-term' AND deprecated_at IS NULL THEN 1 ELSE 0 END) as promoted,
-        SUM(CASE WHEN tier = 'short-term' AND deprecated_at IS NULL THEN 1 ELSE 0 END) as short_term,
-        SUM(CASE WHEN tier = 'long-term' THEN 1 ELSE 0 END) as long_term,
-        AVG(CASE WHEN deprecated_at IS NULL THEN confidence ELSE NULL END) as avg_confidence,
-        AVG(CASE WHEN deprecated_at IS NULL THEN success_rate ELSE NULL END) as avg_success_rate
-      FROM qe_patterns
-    `).get(),t=this.db.prepare(`
-      SELECT COUNT(*) as count
-      FROM qe_patterns
-      WHERE deprecated_at IS NULL
-        AND (
-          consecutive_failures >= ?
-          OR confidence < ?
-          OR (julianday('now') - julianday(COALESCE(last_used_at, created_at))) >= ?
-        )
-    `).get(this.config.deprecationFailureThreshold-1,this.config.minActiveConfidence+.1,this.config.staleDaysThreshold-7);return{totalPatterns:e.total||0,activePatterns:e.active||0,deprecatedPatterns:e.deprecated||0,promotedPatterns:e.promoted||0,shortTermPatterns:e.short_term||0,longTermPatterns:e.long_term||0,avgConfidence:e.avg_confidence||0,avgSuccessRate:e.avg_success_rate||0,patternsNearDeprecation:t.count||0}}getPattern(e){let t=this.db.prepare(`
-      SELECT * FROM qe_patterns WHERE id = ?
-    `).get(e);return t?this.rowToPattern(t):null}rowToPattern(e){return{id:e.id,patternType:e.pattern_type,qeDomain:e.qe_domain,domain:e.domain,name:e.name,description:e.description||"",confidence:e.confidence,usageCount:e.usage_count,successRate:e.success_rate,qualityScore:e.quality_score,tier:e.tier,template:p(e.template_json||"{}"),context:p(e.context_json||"{}"),createdAt:new Date(e.created_at),lastUsedAt:e.last_used_at?new Date(e.last_used_at):new Date(e.created_at),successfulUses:e.successful_uses,reusable:e.reusable===1,reuseCount:e.reuse_count||0,averageTokenSavings:e.average_token_savings||0,deprecated_at:e.deprecated_at??null,consecutive_failures:e.consecutive_failures??0}}};function k(u,e){return new f(u,e)}import*as l from"fs";import*as m from"path";var C={"pattern-learned":"\u{1F9E0}","pattern-promoted":"\u2B06\uFE0F","pattern-quarantined":"\u{1F512}","experience-captured":"\u{1F4F8}","search-performed":"\u{1F50D}","reward-assigned":"\u{1F3AF}"},T=class{config;buffer=[];flushTimer=null;constructor(e){this.config={logDir:e?.logDir??m.join(process.cwd(),".agentic-qe","logs"),enabled:e?.enabled??!0,maxEntriesPerDay:e?.maxEntriesPerDay??500}}log(e){this.config.enabled&&(this.buffer.push(e),this.buffer.length>=10&&this.flush())}flush(){if(this.buffer.length===0)return;let e=new Date().toISOString().split("T")[0],t=m.join(this.config.logDir,`${e}.md`);try{l.mkdirSync(this.config.logDir,{recursive:!0});let n=this.buffer.splice(0,this.buffer.length),r=[];l.existsSync(t)||(r.push(`# AQE Daily Log \u2014 ${e}
-`),r.push("| Time | Event | Summary |"),r.push("|------|-------|---------|"));for(let s of n){let a=s.timestamp.toISOString().split("T")[1]?.slice(0,8)??"00:00:00",c=C[s.type]??"\u{1F4DD}",i=s.summary.replace(/\\/g,"\\\\").replace(/\|/g,"\\|").replace(/\n/g," ");r.push(`| ${a} | ${c} ${s.type} | ${i} |`)}l.appendFileSync(t,r.join(`
-`)+`
-`)}catch(n){console.debug("[DailyLog] Write failed:",n)}}getTodayLogPath(){let e=new Date().toISOString().split("T")[0];return m.join(this.config.logDir,`${e}.md`)}dispose(){this.flush(),this.flushTimer&&(clearInterval(this.flushTimer),this.flushTimer=null)}};export{k as a,T as b};