agentic-qe 3.13.0 → 3.13.2

package/.agents/skills/aqe-plan-quality/SKILL.md

@@ -0,0 +1,25 @@
+---
+name: aqe-plan-quality
+description: Create risk-based quality and test plans for Agentic QE changes. Use when scoping a feature, translating requirements into verification work, selecting QE specialties, assessing testability, or deciding which `.claude/agents/v3/qe-*.md` capabilities should inform a task. Do not use merely to execute an already-defined test command.
+---
+
+# Plan AQE Quality
+
+1. Read `AGENTS.md`, the requested change, and the affected public boundaries.
+2. Inspect the closest implementation and tests before proposing work.
+3. Classify the risk domains using [references/qe-routing.md](references/qe-routing.md).
+4. Select only the QE specialties needed. Treat Claude agent definitions as
+   domain references; do not copy their Claude-only `Task`, `SendMessage`, or
+   hook syntax into Codex actions.
+5. Produce a plan that covers:
+   - observable requirements and acceptance criteria;
+   - affected boundaries, dependencies, and failure modes;
+   - durable tests first, then implementation-coupled tests;
+   - security, performance, accessibility, contract, or resilience checks when
+     triggered by the change;
+   - the narrowest repository verification commands.
+6. Flag assumptions, unavailable environments, destructive operations, and
+   decisions requiring human approval.
+
+Prefer evidence from repository code and configuration over generic test
+checklists. Keep the selected fleet small and explain each specialty’s purpose.

package/.agents/skills/aqe-plan-quality/agents/openai.yaml

@@ -0,0 +1,4 @@
+interface:
+  display_name: "AQE Plan Quality"
+  short_description: "Build risk-based QE plans with the AQE fleet"
+  default_prompt: "Use $aqe-plan-quality to create a risk-based quality plan for this change."

package/.agents/skills/aqe-plan-quality/references/qe-routing.md

@@ -0,0 +1,22 @@
+# QE Routing Reference
+
+Use the shipped definitions in `.claude/agents/v3/qe-*.md` as detailed domain
+references. The principal routing map is:
+
+| Need | QE specialties |
+| --- | --- |
+| Requirements and testability | `qe-requirements-validator`, `qe-bdd-generator`, `qe-product-factors-assessor`, `qe-quality-criteria-recommender` |
+| Test design and implementation | `qe-test-architect`, `qe-tdd-specialist`, `qe-property-tester`, `qe-integration-tester`, `qe-test-idea-rewriter` |
+| Execution and reliability | `qe-parallel-executor`, `qe-flaky-hunter`, `qe-retry-handler` |
+| Coverage and effectiveness | `qe-coverage-specialist`, `qe-gap-detector`, `qe-mutation-tester` |
+| Code impact | `qe-code-intelligence`, `qe-dependency-mapper`, `qe-impact-analyzer`, `qe-kg-builder`, `qe-code-complexity` |
+| Defect analysis | `qe-regression-analyzer`, `qe-root-cause-analyzer`, `qe-defect-predictor` |
+| Quality decision | `qe-risk-assessor`, `qe-quality-gate`, `qe-deployment-advisor`, `qe-devils-advocate` |
+| Security | `qe-security-scanner`, `qe-security-auditor`, `qe-pentest-validator`, `qe-sod-analyzer` |
+| Contract and integration | `qe-contract-validator`, `qe-graphql-tester`, `qe-soap-tester`, `qe-odata-contract-tester`, `qe-message-broker-tester`, `qe-middleware-validator`, `qe-sap-rfc-tester`, `qe-sap-idoc-tester` |
+| Performance and resilience | `qe-performance-tester`, `qe-load-tester`, `qe-chaos-engineer` |
+| UX and accessibility | `qe-visual-tester`, `qe-responsive-tester`, `qe-accessibility-auditor`, `qe-qx-partner` |
+| Learning | `qe-pattern-learner`, `qe-learning-coordinator`, `qe-metrics-optimizer`, `qe-transfer-specialist` |
+| Cross-domain coordination | `qe-fleet-commander`, `qe-queen-coordinator` |
+
+Select the smallest set that covers the actual change.

package/.agents/skills/aqe-review-quality/SKILL.md

@@ -0,0 +1,24 @@
+---
+name: aqe-review-quality
+description: Review Agentic QE changes and issue an evidence-backed quality verdict. Use for code review, regression-risk assessment, release readiness, quality-gate evaluation, security/performance/testability review, or checking whether a change has sufficient verification. Do not use when the user primarily asks to implement a feature.
+---
+
+# Review AQE Quality
+
+1. Read `AGENTS.md`, the diff or requested scope, affected implementation,
+   tests, and configuration.
+2. Review for correctness first: broken behavior, data loss, security,
+   concurrency, compatibility, and missing error paths.
+3. Use [references/quality-gates.md](references/quality-gates.md) to select
+   relevant QE review lenses. Do not run unrelated broad scans.
+4. Verify claims with focused commands where safe and available. Distinguish
+   observed failures from inferred risks.
+5. Report findings by severity with file and line references, the triggering
+   scenario, and the smallest useful remediation.
+6. End with one verdict:
+   - `PASS`: no material finding and adequate evidence;
+   - `CONDITIONAL`: bounded risks or missing environment-dependent evidence;
+   - `FAIL`: correctness, safety, compatibility, or verification blocker.
+7. List the exact checks run and any checks not run.
+
+Do not edit code during a review unless the user also asks for fixes.

package/.agents/skills/aqe-review-quality/agents/openai.yaml

@@ -0,0 +1,4 @@
+interface:
+  display_name: "AQE Review Quality"
+  short_description: "Assess change risk and enforce AQE quality gates"
+  default_prompt: "Use $aqe-review-quality to review this change and produce a quality verdict."

package/.agents/skills/aqe-review-quality/references/quality-gates.md

@@ -0,0 +1,20 @@
+# AQE Quality Gates
+
+Apply only relevant gates:
+
+| Gate | Blocking signal |
+| --- | --- |
+| Correctness | A realistic input violates the stated behavior or invariant |
+| Data safety | A path can corrupt, overwrite, or silently lose persisted data |
+| Security | Authorization bypass, injection, secret exposure, unsafe path, or unvalidated boundary |
+| Compatibility | Public API, CLI, MCP, schema, or shipped asset changes without migration/compatibility handling |
+| Test adequacy | Changed behavior lacks a durable assertion or the test cannot detect a plausible defect |
+| CLI/MCP parity | Equivalent surfaces behave differently or only one was verified |
+| Reliability | Race, leak, retry masking, nondeterministic result, or unbounded resource use |
+| Performance | Measured regression beyond an established threshold |
+| Distribution | `.claude/agents/v3`, `assets/agents/v3`, manifests, or generated bundles are out of sync |
+| Release evidence | Required build, invariant, parity, or smoke check failed or was not run |
+
+Use `.claude/agents/v3/qe-quality-gate.md`,
+`qe-risk-assessor.md`, `qe-deployment-advisor.md`, and the domain-specific QE
+definitions as deeper references when a gate needs specialized criteria.

package/.agents/skills/aqe-test-change/SKILL.md

@@ -0,0 +1,26 @@
+---
+name: aqe-test-change
+description: Design, implement, and run durable tests for Agentic QE code changes. Use when adding tests, reproducing a defect, filling a coverage gap, selecting affected Vitest suites, checking CLI/MCP parity, or verifying a fix in this repository. Do not use for a read-only quality review with no requested edits.
+---
+
+# Test an AQE Change
+
+1. Read `AGENTS.md`, the changed code, nearby tests, and relevant public
+   interfaces.
+2. Reproduce a reported defect with the exact scenario before changing code.
+3. Identify the affected QE specialty from
+   [references/test-specialties.md](references/test-specialties.md).
+4. Add tests in this order:
+   - durable boundary contracts, invariants, or properties;
+   - integration behavior across affected components;
+   - focused unit examples needed for the red-green loop.
+5. Keep tests deterministic. Do not hide failures with retries, skips, weakened
+   assertions, fabricated fixtures, or mocks of the behavior under test.
+6. Run the narrowest matching Vitest target first. Expand to typecheck, build,
+   agent/skill parity, MCP integration, or full CI checks in proportion to the
+   changed surface.
+7. For capabilities exposed through both CLI and MCP, verify both paths.
+8. Report commands, pass/fail counts, limitations, and untested external
+   dependencies.
+
+Never modify or reset `.agentic-qe/memory.db`. Use a copy for any database test.

package/.agents/skills/aqe-test-change/agents/openai.yaml

@@ -0,0 +1,4 @@
+interface:
+  display_name: "AQE Test Change"
+  short_description: "Design and verify durable tests for code changes"
+  default_prompt: "Use $aqe-test-change to test and verify this code change."

package/.agents/skills/aqe-test-change/references/test-specialties.md

@@ -0,0 +1,23 @@
+# Test Specialty Selection
+
+- Public algorithm or data structure: invariants and property tests; consult
+  `qe-property-tester` and `qe-test-architect`.
+- CLI behavior: command-level test plus the equivalent MCP check when exposed;
+  consult `qe-integration-tester`.
+- MCP handler or protocol: handler tests are insufficient by themselves; add a
+  protocol-level integration check.
+- Coverage gap: inspect risk and branch behavior before adding examples; consult
+  `qe-coverage-specialist`, `qe-gap-detector`, and `qe-mutation-tester`.
+- Intermittent failure: preserve evidence and classify nondeterminism; consult
+  `qe-flaky-hunter` and `qe-root-cause-analyzer`.
+- API/schema change: test backward compatibility and negative cases; consult
+  `qe-contract-validator` or the protocol-specific QE definition.
+- Security boundary: test authorization, input validation, path safety, secret
+  handling, and failure disclosure; consult `qe-security-auditor`.
+- Performance-sensitive path: measure a baseline and regression threshold;
+  consult `qe-performance-tester`.
+- Browser or UI change: cover keyboard/accessibility, responsive layout, and
+  stable visual evidence where relevant.
+
+Repository test commands are defined in `package.json`; follow `AGENTS.md` for
+the recommended narrow-to-broad sequence.

package/.claude/helpers/ruflo-hook.cjs

@@ -52,7 +52,11 @@ function main() {
 
   if (commandExists('ruflo')) { invokeHook('ruflo', [], hookArgs, stdinData); done(); }
   if (commandExists('claude-flow')) { invokeHook('claude-flow', [], hookArgs, stdinData); done(); }
-  invokeHook('npx', ['--prefer-offline', '--yes', 'ruflo@latest'], hookArgs, stdinData);
+  // Repository-local Codex hooks disable the network-capable fallback. They
+  // should use an installed stack binary and quietly no-op when it is absent.
+  if (process.env.RUFLO_HOOK_NPX !== '0') {
+    invokeHook('npx', ['--prefer-offline', '--yes', 'ruflo@latest'], hookArgs, stdinData);
+  }
   done();
 }
 

package/.claude/skills/skills-manifest.json

@@ -940,7 +940,7 @@
   },
   "metadata": {
     "generatedBy": "Agentic QE Fleet",
-    "fleetVersion": "3.13.0",
+    "fleetVersion": "3.13.2",
     "manifestVersion": "1.4.0",
     "lastUpdated": "2026-04-13T00:00:00.000Z",
     "contributors": [

package/.codex/hooks/aqe-codex-hook.cjs

@@ -0,0 +1,69 @@
+#!/usr/bin/env node
+'use strict';
+
+const { spawnSync } = require('node:child_process');
+const path = require('node:path');
+
+let input = {};
+try {
+  input = JSON.parse(require('node:fs').readFileSync(0, 'utf8') || '{}');
+} catch {
+  process.exit(0);
+}
+
+const root = input.cwd
+  ? spawnSync('git', ['-C', input.cwd, 'rev-parse', '--show-toplevel'], {
+      encoding: 'utf8',
+      timeout: 1000,
+    }).stdout.trim()
+  : '';
+if (!root) process.exit(0);
+
+const subcommand = process.argv[2];
+if (!subcommand) process.exit(0);
+
+const codexShim = path.join(root, '.codex', 'hooks', 'aqe-runtime.cjs');
+const shim = require('node:fs').existsSync(codexShim)
+  ? codexShim
+  : path.join(root, '.claude', 'hooks', 'aqe-hook.cjs');
+const result = spawnSync(process.execPath, [shim, subcommand, '--json'], {
+  cwd: root,
+  env: { ...process.env, CLAUDE_PROJECT_DIR: root },
+  input: JSON.stringify(input),
+  encoding: 'utf8',
+  timeout: Number(process.env.AQE_CODEX_HOOK_TIMEOUT_MS) || 20000,
+  maxBuffer: 4 * 1024 * 1024,
+});
+
+if (!result.stdout) process.exit(0);
+
+let output;
+try {
+  output = JSON.parse(result.stdout);
+} catch {
+  process.exit(0);
+}
+
+// AQE route results are useful context but are not themselves a Codex hook
+// response. Wrap them in the supported additional-context envelope.
+if (subcommand === 'route' && !output.hookSpecificOutput) {
+  const agent = output.recommendedAgent || 'unspecified';
+  const confidence = Number.isFinite(output.confidence)
+    ? ` (${Math.round(output.confidence * 100)}% confidence)`
+    : '';
+  const domains = Array.isArray(output.domains) && output.domains.length
+    ? ` Domains: ${output.domains.join(', ')}.`
+    : '';
+  const guidance = Array.isArray(output.guidance) && output.guidance.length
+    ? ` Guidance: ${output.guidance.join('; ')}.`
+    : '';
+  output = {
+    hookSpecificOutput: {
+      hookEventName: 'UserPromptSubmit',
+      additionalContext:
+        `AQE routing recommends ${agent}${confidence}.${domains}${guidance}`,
+    },
+  };
+}
+
+process.stdout.write(`${JSON.stringify(output)}\n`);

package/.codex/hooks/ruflo-codex-hook.cjs

@@ -0,0 +1,94 @@
+#!/usr/bin/env node
+'use strict';
+
+const fs = require('node:fs');
+const path = require('node:path');
+const { spawnSync } = require('node:child_process');
+
+let input = {};
+let rawInput = '';
+try {
+  rawInput = fs.readFileSync(0, 'utf8');
+  input = JSON.parse(rawInput || '{}');
+} catch {
+  process.exit(0);
+}
+
+const git = input.cwd
+  ? spawnSync('git', ['-C', input.cwd, 'rev-parse', '--show-toplevel'], {
+      encoding: 'utf8',
+      timeout: 1000,
+    })
+  : null;
+const root = git && git.status === 0 ? git.stdout.trim() : '';
+if (!root) process.exit(0);
+
+const requested = process.argv[2];
+if (!requested) process.exit(0);
+
+const toolInput = input.tool_input || {};
+const toolResponse = input.tool_response || input.tool_result || {};
+const args = [requested];
+
+function add(flag, value) {
+  if (value !== undefined && value !== null && String(value).length) {
+    args.push(flag, String(value));
+  }
+}
+
+switch (requested) {
+  case 'route':
+  case 'pre-task':
+    add('--task', input.prompt || input.user_prompt || toolInput.prompt);
+    break;
+  case 'pre-command':
+  case 'post-command':
+    add('--command', toolInput.command);
+    if (requested === 'post-command') {
+      add('--success', toolResponse.success !== false);
+    }
+    break;
+  case 'pre-edit':
+  case 'post-edit':
+    add(
+      '--file',
+      toolInput.file_path || toolInput.path || toolInput.target_file,
+    );
+    if (requested === 'post-edit') {
+      add('--success', toolResponse.success !== false);
+    }
+    break;
+  case 'session-restore':
+  case 'session-end':
+    add('--session-id', input.session_id);
+    break;
+  case 'post-task':
+    add('--task-id', input.agent_id || input.subagent_id || input.turn_id);
+    add('--success', true);
+    break;
+  default:
+    break;
+}
+
+const codexHelper = path.join(root, '.codex', 'hooks', 'ruflo-runtime.cjs');
+const helper = fs.existsSync(codexHelper)
+  ? codexHelper
+  : path.join(root, '.claude', 'helpers', 'ruflo-hook.cjs');
+if (!fs.existsSync(helper)) process.exit(0);
+
+// Ruflo is development-time coordination tooling. Keep this hook fail-open,
+// avoid network installation during a lifecycle event, and suppress its
+// provider-specific output rather than exposing unsupported Codex hook fields.
+spawnSync(process.execPath, [helper, ...args], {
+  cwd: root,
+  env: {
+    ...process.env,
+    CLAUDE_PROJECT_DIR: root,
+    RUFLO_HOOK_NPX: '0',
+  },
+  input: rawInput,
+  stdio: ['pipe', 'ignore', 'ignore'],
+  timeout: Number(process.env.RUFLO_CODEX_HOOK_TIMEOUT_MS) || 15000,
+});
+
+process.exit(0);

package/.codex/hooks.json

@@ -0,0 +1,169 @@
+{
+  "description": "AQE routing, lifecycle learning, and edit/command guardrails for Codex.",
+  "hooks": {
+    "SessionStart": [
+      {
+        "matcher": "startup|resume|clear|compact",
+        "hooks": [
+          {
+            "type": "command",
+            "command": "node \"$(git rev-parse --show-toplevel)/.codex/hooks/aqe-codex-hook.cjs\" session-start",
+            "timeout": 25,
+            "statusMessage": "Loading AQE learning context"
+          },
+          {
+            "type": "command",
+            "command": "node \"$(git rev-parse --show-toplevel)/.codex/hooks/ruflo-codex-hook.cjs\" session-restore",
+            "timeout": 15,
+            "statusMessage": "Restoring ruflo coordination context"
+          }
+        ]
+      }
+    ],
+    "UserPromptSubmit": [
+      {
+        "hooks": [
+          {
+            "type": "command",
+            "command": "node \"$(git rev-parse --show-toplevel)/.codex/hooks/aqe-codex-hook.cjs\" route",
+            "timeout": 20,
+            "statusMessage": "Routing through the AQE fleet"
+          },
+          {
+            "type": "command",
+            "command": "node \"$(git rev-parse --show-toplevel)/.codex/hooks/ruflo-codex-hook.cjs\" route",
+            "timeout": 15,
+            "statusMessage": "Recording ruflo task routing"
+          }
+        ]
+      }
+    ],
+    "PreToolUse": [
+      {
+        "matcher": "^Bash$",
+        "hooks": [
+          {
+            "type": "command",
+            "command": "node \"$(git rev-parse --show-toplevel)/.codex/hooks/aqe-codex-hook.cjs\" pre-command",
+            "timeout": 20,
+            "statusMessage": "Applying AQE command guardrails"
+          },
+          {
+            "type": "command",
+            "command": "node \"$(git rev-parse --show-toplevel)/.codex/hooks/ruflo-codex-hook.cjs\" pre-command",
+            "timeout": 15,
+            "statusMessage": "Assessing command risk with ruflo"
+          }
+        ]
+      },
+      {
+        "matcher": "^(apply_patch|Edit|Write)$",
+        "hooks": [
+          {
+            "type": "command",
+            "command": "node \"$(git rev-parse --show-toplevel)/.codex/hooks/aqe-codex-hook.cjs\" pre-edit",
+            "timeout": 20,
+            "statusMessage": "Applying AQE edit guardrails"
+          },
+          {
+            "type": "command",
+            "command": "node \"$(git rev-parse --show-toplevel)/.codex/hooks/ruflo-codex-hook.cjs\" pre-edit",
+            "timeout": 15,
+            "statusMessage": "Loading ruflo edit context"
+          }
+        ]
+      }
+    ],
+    "PostToolUse": [
+      {
+        "matcher": "^Bash$",
+        "hooks": [
+          {
+            "type": "command",
+            "command": "node \"$(git rev-parse --show-toplevel)/.codex/hooks/aqe-codex-hook.cjs\" post-command",
+            "timeout": 20,
+            "statusMessage": "Recording AQE command outcome"
+          },
+          {
+            "type": "command",
+            "command": "node \"$(git rev-parse --show-toplevel)/.codex/hooks/ruflo-codex-hook.cjs\" post-command",
+            "timeout": 15,
+            "statusMessage": "Recording command outcome in ruflo"
+          }
+        ]
+      },
+      {
+        "matcher": "^(apply_patch|Edit|Write)$",
+        "hooks": [
+          {
+            "type": "command",
+            "command": "node \"$(git rev-parse --show-toplevel)/.codex/hooks/aqe-codex-hook.cjs\" post-edit",
+            "timeout": 20,
+            "statusMessage": "Recording AQE edit outcome"
+          },
+          {
+            "type": "command",
+            "command": "node \"$(git rev-parse --show-toplevel)/.codex/hooks/ruflo-codex-hook.cjs\" post-edit",
+            "timeout": 15,
+            "statusMessage": "Recording edit outcome in ruflo"
+          }
+        ]
+      }
+    ],
+    "PreCompact": [
+      {
+        "matcher": "manual|auto",
+        "hooks": [
+          {
+            "type": "command",
+            "command": "node \"$(git rev-parse --show-toplevel)/.codex/hooks/ruflo-codex-hook.cjs\" session-end",
+            "timeout": 15,
+            "statusMessage": "Checkpointing ruflo state before compaction"
+          }
+        ]
+      }
+    ],
+    "SubagentStart": [
+      {
+        "hooks": [
+          {
+            "type": "command",
+            "command": "node \"$(git rev-parse --show-toplevel)/.codex/hooks/ruflo-codex-hook.cjs\" pre-task",
+            "timeout": 15,
+            "statusMessage": "Registering subagent task with ruflo"
+          }
+        ]
+      }
+    ],
+    "SubagentStop": [
+      {
+        "hooks": [
+          {
+            "type": "command",
+            "command": "node \"$(git rev-parse --show-toplevel)/.codex/hooks/ruflo-codex-hook.cjs\" post-task",
+            "timeout": 15,
+            "statusMessage": "Recording subagent outcome in ruflo"
+          }
+        ]
+      }
+    ],
+    "Stop": [
+      {
+        "hooks": [
+          {
+            "type": "command",
+            "command": "node \"$(git rev-parse --show-toplevel)/.codex/hooks/aqe-codex-hook.cjs\" session-end",
+            "timeout": 20,
+            "statusMessage": "Saving AQE session state"
+          },
+          {
+            "type": "command",
+            "command": "node \"$(git rev-parse --show-toplevel)/.codex/hooks/ruflo-codex-hook.cjs\" session-end",
+            "timeout": 15,
+            "statusMessage": "Saving ruflo session state"
+          }
+        ]
+      }
+    ]
+  }
+}

package/CHANGELOG.md

@@ -5,6 +5,129 @@ All notable changes to the Agentic QE project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [3.13.2] - 2026-07-24
+
+Stops short-lived learning hooks from repeatedly producing
+`brain.rvf.corrupt-*` and `patterns.rvf.corrupt-*` files. A busy project could
+accumulate thousands of these derived-cache artifacts while vector search
+repeatedly fell back to SQLite.
+
+### Fixed
+
+- **Hook processes now close every RVF store before exiting** ([#574]).
+  Every `aqe hooks` invocation opened the shared pattern index and brain
+  dual-writer, but the CLI had no teardown step. The native stores therefore
+  exited without a clean close; the next hook found an unusable 162-byte store,
+  quarantined it, rebuilt it, and repeated the cycle. Hook commands now dispose
+  the reasoning bank and reset both RVF singletons after every action.
+- **A duplicate opener can no longer break its own process's live RVF lock**
+  ([#574]). Recovery protected locks owned by other live processes but treated
+  a lock containing the current PID as stale. Another initialization path in
+  that process could then remove the lock and replace a store still in use.
+  Every live owner is now protected; duplicate paths degrade safely to SQLite
+  instead of moving a live store.
+
+### Changed
+
+- **RVF hook cleanup is idempotent and fail-open.** Partial initialization,
+  optional native bindings, or an individual close failure cannot block the
+  host editor or command. True stale locks remain recoverable after their owner
+  exits.
+
+## [3.13.1] - 2026-07-23
+
+Fixes four reported issues, three of which come down to the same thing: a tool
+told you something it had not actually established. Coverage analysis reported
+numbers it never measured, test generation claimed AI enhancement it never
+performed, and a security advisory reached users through a dependency they never
+asked for.
+
+### Fixed
+
+- **Coverage analysis no longer reports numbers it did not measure** ([#569]).
+  On a Rust crate, `coverage_analyze_sublinear` returned `78.3%` lines,
+  `100%` branches and `0%` functions — none of it measured. It flagged
+  `#[cfg(test)]` blocks as untested production code, cited line numbers past the
+  end of the file, and analysed test files as coverage targets, all with a
+  confidence score attached. Rust crates are now measured with `cargo llvm-cov`
+  when it is available. When nothing can be measured, the result says so
+  (`estimated: true`, `measured: false`) instead of presenting a guess as data.
+- **`test_generate_enhanced` actually calls your LLM now** ([#567]). The tool
+  advertised AI-powered generation but the LLM router never reached the code path
+  it runs, so it always returned the same generic scaffold regardless of how you
+  configured your provider. It now routes through your configured model, and the
+  result reports `llmEnhanced` so a silent fall back to templates is visible
+  rather than something you infer from a suspiciously fast response. A broken
+  provider (bad key, timeout) reports the fallback honestly instead of claiming
+  enhancement it did not perform.
+- **A HIGH-severity advisory no longer reaches consumers** ([#565]).
+  `@huggingface/transformers` was declared an *optional* peer dependency, but npm
+  installs optional peers — so every install pulled in
+  `onnxruntime-node -> adm-zip` and its advisory. It is now an explicit opt-in:
+  install it yourself only if you want local in-process embeddings. Everyone
+  using an embedding endpoint (the default) no longer carries the advisory.
+- **`codex` is selectable through the LLM router.** The provider existed and
+  worked, but was missing from the router's provider list, so it could never
+  actually be chosen.
+- **Fixes now take effect immediately after an upgrade.** Operation results were
+  cached per project and keyed without the build version, so an upgraded install
+  could keep replaying the previous version's answers for up to an hour — making
+  a fix look like it had not worked. Set `AQE_SESSION_CACHE_SALT` to force a cold
+  cache at any time.
+
+### Added
+
+- **Per-agent LLM routing you can configure on disk** ([#568]). Route individual
+  agent types through different providers via an `agentOverrides` map in
+  `.agentic-qe/llm-config.json` — a strong reasoning model for security agents, a
+  free local model for mechanical ones — without changing your global default:
+
+  ```jsonc
+  {
+    "defaultProvider": "ollama",
+    "agentOverrides": {
+      "qe-security-scanner": { "provider": "claude-code", "model": "opus" },
+      "qe-mutation-tester":  { "provider": "ollama" }
+    }
+  }
+  ```
+
+  Entries are partial (unset fields keep their defaults), invalid entries are
+  dropped with an explanation rather than failing startup, and credentials are
+  never persisted here. See `docs/guides/llm-provider-routing.md`.
+- **`AQE_COVERAGE_NO_EXEC=1` — opt out of running build tooling.** Measuring
+  coverage means running your project's tests, which executes code from the
+  analysed repository (`cargo` also runs `build.rs`; `npx` runs lifecycle
+  scripts). That is what you want for your own code, but is worth being able to
+  turn off when pointing the tool at something you do not trust. Existing
+  coverage reports are still read; everything else degrades to a labelled
+  estimate.
+- **QE skills and lifecycle hooks for the Codex CLI.** `aqe init` now installs
+  three Codex-native skills — `aqe-test-change`, `aqe-review-quality`,
+  `aqe-plan-quality` — under `.agents/skills/`, plus Codex lifecycle hooks in
+  `.codex/`, and an `AGENTS.md` describing the repository's conventions. If you
+  drive this project with Codex rather than Claude Code, the QE workflows are now
+  available there natively instead of Claude-only.
+- **Security scans state how deeply they looked.** Full SAST is implemented for
+  JS/TS only; other languages get cross-language pattern matching. Results now
+  carry `analysisDepth`, so a zero-finding scan reads as "nothing matched a
+  generic pattern" rather than "no vulnerabilities".
+
+### Changed
+
+- **BREAKING (result shape):** `coverage_analyze_sublinear` now returns
+  `null` — not `0` — for any coverage metric that was not collected.
+  `lineCoverage`, `statementCoverage`, `branchCoverage` and `functionCoverage`
+  are all `number | null`. `0` now means a measured zero; `null` means unknown.
+  Callers doing arithmetic or comparisons on these fields should handle `null`;
+  check the new `measured` / `estimated` flags to tell the cases apart. This is
+  deliberately a visible break rather than a silent wrong number.
+
+[#565]: https://github.com/proffesor-for-testing/agentic-qe/issues/565
+[#567]: https://github.com/proffesor-for-testing/agentic-qe/issues/567
+[#568]: https://github.com/proffesor-for-testing/agentic-qe/issues/568
+[#569]: https://github.com/proffesor-for-testing/agentic-qe/issues/569
+
 ## [3.13.0] - 2026-07-18
 
 Introduces **QE-Court** — an adversarial review court where a change is

package/assets/skills/skills-manifest.json

@@ -940,7 +940,7 @@
   },
   "metadata": {
     "generatedBy": "Agentic QE Fleet",
-    "fleetVersion": "3.13.0",
+    "fleetVersion": "3.13.2",
     "manifestVersion": "1.4.0",
     "lastUpdated": "2026-04-13T00:00:00.000Z",
     "contributors": [