@claude-flow/cli 3.42.2 → 3.42.3

package/plugins/ruflo-metaharness/scripts/test-mcp-tools.mjs

@@ -1,498 +1,498 @@
-#!/usr/bin/env node
-// test-mcp-tools.mjs — runtime test for the iter-20/21 MCP tool registry.
-//
-// tsc proves the metaharness-tools.ts module COMPILES; structural smoke
-// proves the source DECLARES the right tool names. Neither proves the
-// HANDLERS actually run without throwing. This test imports the compiled
-// module and invokes every tool's handler with minimal inputs.
-//
-// CONTRACT EACH TOOL MUST SATISFY
-//   - handler is callable as `await tool.handler({ ... })`
-//   - returns an object with keys: success, data, degraded, exitCode
-//   - never throws (even with bad/missing optional dep — graceful)
-//   - handler honors the 120s subprocess timeout (no hang)
-//
-// USAGE
-//   node scripts/test-mcp-tools.mjs                          # default
-//   node scripts/test-mcp-tools.mjs --format json
-//
-// EXIT CODES
-//   0  all tools satisfy the contract
-//   1  at least one tool failed
-//   2  setup error (compiled dist not present)
-
-import { existsSync } from 'node:fs';
-import { dirname, join, resolve } from 'node:path';
-import { fileURLToPath } from 'node:url';
-
-const SCRIPTS_DIR = dirname(fileURLToPath(import.meta.url));
-const ARGS = (() => {
-  const a = { format: 'table' };
-  for (let i = 2; i < process.argv.length; i++) {
-    if (process.argv[i] === '--format') a.format = process.argv[++i];
-  }
-  return a;
-})();
-
-let passed = 0, failed = 0;
-const failures = [];
-
-function assert(cond, label) {
-  if (cond) { console.log(`  ✓ ${label}`); passed++; }
-  else { console.log(`  ✗ ${label}`); failures.push(label); failed++; }
-}
-
-async function main() {
-  // Locate the compiled dist of metaharness-tools.
-  const distPath = resolve(SCRIPTS_DIR, '..', '..', '..',
-    'v3', '@claude-flow', 'cli', 'dist', 'src', 'mcp-tools', 'metaharness-tools.js');
-
-  if (!existsSync(distPath)) {
-    console.log(`# test-mcp-tools — SKIPPED`);
-    console.log('');
-    console.log(`Compiled dist not present: ${distPath}`);
-    console.log(`Build the CLI first:`);
-    console.log(`  cd v3/@claude-flow/cli && npm run build`);
-    console.log('');
-    console.log(`Exit 0 — this script is meaningfully runnable only post-build.`);
-    process.exit(0);
-  }
-
-  let mod;
-  try {
-    mod = await import(distPath);
-  } catch (e) {
-    console.error(`test-mcp-tools: failed to import ${distPath}: ${e.message}`);
-    process.exit(2);
-  }
-
-  const tools = mod.metaharnessTools;
-  console.log(`# test-mcp-tools — runtime contract\n`);
-
-  // ──────────────────────────────────────────────────────────────────
-  // PHASE 1 — module exports the right shape
-  // ──────────────────────────────────────────────────────────────────
-  console.log('Phase 1 — module shape');
-  assert(Array.isArray(tools), 'metaharnessTools is an array');
-  assert(tools.length === 16, `16 tools registered (got ${tools.length})`);
-
-  const expectedNames = new Set([
-    'metaharness_score',
-    'metaharness_genome',
-    'metaharness_mcp_scan',
-    'metaharness_threat_model',
-    'metaharness_oia_audit',
-    'metaharness_audit_list',
-    'metaharness_audit_trend',
-    // iter 36 — ADR-152 §3.1 production
-    'metaharness_similarity',
-    // iter 54 — one-command drift detection (composes audit-list + oia-audit + audit-trend)
-    'metaharness_drift_from_history',
-    // ADR-153 — bench suites + evolve driver + security-focused bench
-    'metaharness_bench',
-    'metaharness_evolve',
-    'metaharness_security_bench',
-    // @metaharness/redblue@~0.1.4 — adversarial red/blue LLM testing
-    'metaharness_redblue',
-    // metaharness@0.3.0 — upstream ADR-235 GEPA learning run
-    'metaharness_learn',
-    // @metaharness/darwin@0.8.0 — GEPA library surface (genome ops)
-    'metaharness_gepa',
-    // ADR-322 — in-process governed evaluation and atomic promotion
-    'metaharness_flywheel',
-  ]);
-  const actualNames = new Set(tools.map((t) => t.name));
-  for (const name of expectedNames) {
-    assert(actualNames.has(name), `${name} registered`);
-  }
-
-  // ──────────────────────────────────────────────────────────────────
-  // PHASE 2 — every tool has the required MCP shape
-  // ──────────────────────────────────────────────────────────────────
-  console.log('\nPhase 2 — per-tool shape');
-  for (const tool of tools) {
-    const ok = typeof tool.name === 'string'
-      && typeof tool.description === 'string'
-      && typeof tool.category === 'string'
-      && typeof tool.handler === 'function'
-      && typeof tool.inputSchema === 'object';
-    assert(ok, `${tool.name} has {name, description, category, handler, inputSchema}`);
-    assert(tool.category === 'metaharness', `${tool.name} category === 'metaharness'`);
-  }
-
-  // ──────────────────────────────────────────────────────────────────
-  // PHASE 3 — handlers callable + return contract shape
-  //
-  // We invoke each handler with minimal valid input. The handlers may
-  // succeed (if metaharness is installed) or report degraded (if not).
-  // EITHER way, they must return { success, data, degraded, exitCode }
-  // without throwing.
-  // ──────────────────────────────────────────────────────────────────
-  console.log('\nPhase 3 — handler invocations (allow up to 30s each)');
-  for (const tool of tools) {
-    // Construct minimal valid input per tool.
-    let input = {};
-    if (tool.name === 'metaharness_audit_trend') {
-      // Requires baselineKey + currentKey — use fake keys that won't
-      // resolve so we exercise the not-found path.
-      input = { baselineKey: 'audit-fake-base', currentKey: 'audit-fake-curr' };
-    }
-    if (tool.name === 'metaharness_similarity') {
-      // Needs --a/--b OR --a-key/--b-key. Use fake mem keys to exercise
-      // the graceful not-found path (matches audit_trend convention).
-      input = { aKey: 'harness-fake-a', bKey: 'harness-fake-b' };
-    }
-    if (tool.name === 'metaharness_drift_from_history') {
-      // iter 54 — composes 3 subprocesses, needs more time than the default.
-      input = { dryRun: true, threshold: 0.5 };
-    }
-    if (tool.name === 'metaharness_oia_audit') {
-      // iter 128 — composite audit runs 5 sub-audits (oia-manifest +
-      // threat-model + mcp-scan + score + genome) in parallel. Each
-      // shells out via npx. --dry-run skips memory persistence so the
-      // test doesn't pollute namespaces.
-      input = { dryRun: true };
-    }
-    if (tool.name === 'metaharness_redblue') {
-      // `attack` preview is the fastest path that exercises the upstream
-      // binary without needing OPENROUTER_API_KEY or running any model
-      // calls. Count=1 keeps cold-cache npx fetch the dominant cost.
-      input = { subcommand: 'attack', family: 'prompt', count: 1 };
-    }
-    if (tool.name === 'metaharness_learn') {
-      // No repo checkout in CI → structured {status:"checkout-required"}
-      // exit-0 path. $0: without run=true upstream never spends anyway.
-      input = {};
-    }
-    if (tool.name === 'metaharness_gepa') {
-      // op is required; `genome` loads + validates the SHIPPED cand-6
-      // genome — pure-local library call once darwin is cached.
-      input = { op: 'genome' };
-    }
-
-    // iter 124 → 130 — timeouts have crept up as CI cold-cache npx
-    // warmup costs got measured. Final budgets:
-    //   default          : 60s
-    //   chain-tools      : 180s  (drift_from_history + oia_audit + audit_list)
-    // iter 131 — bumped chain-tool budget 90s → 180s. audit_list still
-    // timed out at 90s in CI; locally it runs in ~4s, but CI's
-    // `npx @claude-flow/cli@latest memory list` invocation pays both
-    // the npx fetch AND a full CLI startup (which loads agentic-flow +
-    // ONNX). 180s gives 30x headroom over the local cost.
-    const isChainTool = tool.name === 'metaharness_drift_from_history'
-      || tool.name === 'metaharness_oia_audit'
-      || tool.name === 'metaharness_audit_list'
-      // redblue: `attack prompt --count 1` is preview-only (no model
-      // calls) but the cold-cache `npx -y @metaharness/redblue@~0.1.4`
-      // fetch can take 30-60s. 180s gives 3x headroom.
-      || tool.name === 'metaharness_redblue'
-      // learn: cold-cache `npx -y metaharness@latest` fetch dominates.
-      // gepa: one-time `npm install --prefix ~/.ruflo/darwin-cache-*`
-      // fallback install can take 30-60s on cold cache.
-      || tool.name === 'metaharness_learn'
-      || tool.name === 'metaharness_gepa';
-    const timeoutMs = isChainTool ? 180_000 : 60_000;
-    const handlerPromise = tool.handler(input);
-    const timeoutPromise = new Promise((_, reject) =>
-      setTimeout(() => reject(new Error(`${timeoutMs / 1000}s handler timeout`)), timeoutMs));
-
-    let result;
-    let threw = false;
-    try {
-      result = await Promise.race([handlerPromise, timeoutPromise]);
-    } catch (e) {
-      threw = true;
-      console.log(`    [${tool.name}] handler threw: ${e.message.slice(0, 80)}`);
-    }
-
-    assert(!threw, `${tool.name} handler did not throw`);
-    if (!threw && result) {
-      assert(typeof result === 'object', `${tool.name} returns object`);
-      assert('success' in result, `${tool.name} result has 'success'`);
-      assert('data' in result, `${tool.name} result has 'data'`);
-      assert('degraded' in result, `${tool.name} result has 'degraded'`);
-      assert('exitCode' in result, `${tool.name} result has 'exitCode'`);
-    }
-  }
-
-  // ──────────────────────────────────────────────────────────────────
-  // PHASE 4 — POSITIVE-CASE data-shape validation (iter 43)
-  //
-  // Iter 37 verified the {success, data, degraded, exitCode} envelope.
-  // It did NOT verify that data.X contains the right keys when success
-  // is genuinely true — leaving room for iter 42-style bugs where a
-  // handler returns valid-looking degraded JSON while silently
-  // misrouting input. This phase invokes each handler with VALID
-  // inputs and asserts the expected output shape.
-  //
-  // Tools that depend on `npx metaharness` (score/genome/mcp-scan/
-  // threat-model/oia-audit/audit-list/audit-trend) are SKIPPED in this
-  // phase when the optional dep isn't installed — they're covered by
-  // the no-metaharness-smoke workflow's drill. The similarity tool
-  // has no @metaharness/* dep, so its positive case ALWAYS runs.
-  // ──────────────────────────────────────────────────────────────────
-  console.log('\nPhase 4 — positive-case data shape (iter 43)');
-
-  const { writeFileSync, mkdtempSync, mkdirSync } = await import('node:fs');
-  const { tmpdir } = await import('node:os');
-  const { join: pjoin } = await import('node:path');
-  const tmp = mkdtempSync(pjoin(tmpdir(), 'mcp-positive-'));
-
-  // #2626 — upstream deliberately exits 2 for a valid `blocked` genome.
-  // The wrapper must preserve that report as data instead of converting it
-  // into an input/system error at the MCP boundary.
-  const genomeTool = tools.find((t) => t.name === 'metaharness_genome');
-  if (genomeTool) {
-    const blockedRepo = pjoin(tmp, 'blocked-repo');
-    mkdirSync(blockedRepo);
-    const r = await genomeTool.handler({ path: blockedRepo });
-    if (!r.degraded) {
-      assert(r.success === true,
-        'genome blocked verdict: wrapper succeeds with a valid report (#2626)');
-      assert(r.exitCode === 0,
-        'genome blocked verdict: wrapper exitCode === 0 (#2626)');
-      assert(r.data?.verdict === 'blocked',
-        `genome blocked verdict: data.verdict === blocked (got ${r.data?.verdict})`);
-      assert(r.data?.verdictExitCode === 2,
-        `genome blocked verdict: preserves upstream exit 2 (got ${r.data?.verdictExitCode})`);
-      assert(typeof r.data?.risk_score === 'number' && r.data.risk_score >= 0.7,
-        `genome blocked verdict: risk_score >= 0.7 (got ${r.data?.risk_score})`);
-    } else {
-      console.log(`    ⊘ genome: metaharness absent — graceful skip`);
-    }
-  }
-
-  // metaharness_similarity — full positive case (no @metaharness/* needed)
-  const simTool = tools.find((t) => t.name === 'metaharness_similarity');
-  if (simTool) {
-    const aPath = pjoin(tmp, 'a.json');
-    const bPath = pjoin(tmp, 'b.json');
-    writeFileSync(aPath, JSON.stringify({
-      score: { harnessFit: 78, compileConfidence: 92, taskCoverage: 65, toolSafety: 88, memoryUsefulness: 70, estCostPerRunUsd: 0.04, recommendedMode: 'CLI + MCP', archetype: 'compliance-harness', template: 'vertical:legal' },
-      genome: { repo_type: 'node_mcp_ci', agent_topology: ['x','y','z','w'], risk_score: 0.45, test_confidence: 0.7, publish_readiness: 0.6 },
-    }));
-    writeFileSync(bPath, JSON.stringify({
-      score: { harnessFit: 75, compileConfidence: 90, taskCoverage: 70, toolSafety: 90, memoryUsefulness: 72, estCostPerRunUsd: 0.05, recommendedMode: 'CLI + MCP', archetype: 'compliance-harness', template: 'vertical:support' },
-      genome: { repo_type: 'node_mcp_ci', agent_topology: ['x','y','z','q','r'], risk_score: 0.40, test_confidence: 0.75, publish_readiness: 0.65 },
-    }));
-    const r = await simTool.handler({ aFile: aPath, bFile: bPath });
-    assert(r.degraded === false, 'similarity positive case: degraded === false');
-    assert(r.success === true, 'similarity positive case: success === true');
-    assert(r.exitCode === 0, 'similarity positive case: exitCode === 0');
-    const d = r.data ?? {};
-    assert(typeof d.overall === 'number', 'similarity data has numeric `overall`');
-    assert(typeof d.components === 'object' && d.components !== null,
-      'similarity data has `components` object');
-    assert(typeof d.components?.cosine === 'number',
-      'similarity components.cosine numeric');
-    assert(typeof d.components?.categorical === 'number',
-      'similarity components.categorical numeric');
-    assert(typeof d.components?.jaccard === 'number',
-      'similarity components.jaccard numeric');
-    assert(typeof d.weights === 'object' && d.weights !== null,
-      'similarity data has `weights` object');
-    assert(d.adr === 'ADR-152', 'similarity data tagged adr=ADR-152');
-    // Regression anchor — same fixtures as iter-35 spike with non-matching topologies
-    assert(d.overall > 0 && d.overall < 1,
-      `similarity overall in (0, 1) — got ${d.overall}`);
-
-    // Per-dimension variant
-    const rPD = await simTool.handler({ aFile: aPath, bFile: bPath, perDimension: true });
-    assert(typeof rPD.data?.perDimension === 'object',
-      'similarity perDimension=true populates breakdown');
-
-    // Alert-below variant exercises non-zero exit
-    const rAlert = await simTool.handler({ aFile: aPath, bFile: bPath, alertBelow: 0.99 });
-    assert(rAlert.data?.alert?.triggered === true,
-      'similarity alertBelow=0.99 triggers alert');
-    assert(rAlert.exitCode === 1, 'similarity alertBelow=0.99 → exitCode 1');
-    // iter 44 — success semantic anchor (was true under the pre-iter-44
-    // `!degraded` rule; now false because exitCode !== 0 dominates).
-    assert(rAlert.success === false,
-      'similarity alertBelow=0.99 → success === false (iter 44 fix)');
-  }
-
-  // metaharness_mcp_scan — positive case post iter-50 parser landing.
-  // Until iter 50, mcp_scan's data field was an alert-only object with
-  // no structured findings. After iter 50, findings[] is always present
-  // (parsed from upstream text) and summary{overallSeverity, totalCount}
-  // accompanies it.
-  const scanTool = tools.find((t) => t.name === 'metaharness_mcp_scan');
-  if (scanTool) {
-    // Run against ruflo itself — guaranteed to produce at least the
-    // INFO finding the iter-50 parser test verified manually.
-    const r = await scanTool.handler({ path: '.', failOn: 'high' });
-    // Either succeeds with structured findings, or gracefully degrades
-    // if metaharness isn't installed in this environment.
-    if (!r.degraded) {
-      assert(r.success === true, 'mcp_scan positive: success === true');
-      assert(r.exitCode === 0, 'mcp_scan positive: exitCode === 0');
-      assert(Array.isArray(r.data?.findings),
-        'mcp_scan positive: data.findings is an array (iter 50 fix)');
-      // Cwd-dependent: when scanning a dir without .mcp/servers.json the
-      // upstream emits no findings. Only verify shape contract when array
-      // is populated — the array-presence assertion above is the
-      // load-bearing one for iter 50.
-      if (r.data?.findings.length > 0) {
-        const first = r.data.findings[0];
-        assert(typeof first?.severity === 'string',
-          'mcp_scan positive: first finding has string severity');
-        assert(typeof first?.message === 'string',
-          'mcp_scan positive: first finding has string message');
-      }
-      // summary may be null if the upstream produced no Result: line —
-      // verify the field's presence (null OR object) but only deep-check
-      // when populated.
-      if (r.data?.summary) {
-        assert(typeof r.data.summary.totalCount === 'number',
-          'mcp_scan positive: data.summary.totalCount is numeric (when summary present)');
-      }
-    } else {
-      console.log(`    ⊘ mcp_scan: metaharness absent — graceful skip`);
-    }
-  }
-
-  // metaharness_audit_trend — positive case via file inputs
-  const trendTool = tools.find((t) => t.name === 'metaharness_audit_trend');
-  if (trendTool) {
-    const basePath = pjoin(tmp, 'base.json');
-    const currPath = pjoin(tmp, 'curr.json');
-    const fingerprint = {
-      score: { harnessFit: 80, recommendedMode: 'CLI + MCP', archetype: 'typescript-sdk-harness', template: 'vertical:coding' },
-      genome: { repo_type: 'node_mcp_ci', agent_topology: ['a','b','c'], risk_score: 0.3, test_confidence: 0.85, publish_readiness: 0.9 },
-    };
-    writeFileSync(basePath, JSON.stringify({
-      startedAt: '2026-06-15T00:00:00Z',
-      composite: { worst: 'clean' },
-      components: { oiaManifest: {}, threatModel: {}, mcpScan: { json: { findings: [] } } },
-      fingerprint,
-    }));
-    writeFileSync(currPath, JSON.stringify({
-      startedAt: '2026-06-16T00:00:00Z',
-      composite: { worst: 'clean' },
-      components: { oiaManifest: {}, threatModel: {}, mcpScan: { json: { findings: [] } } },
-      fingerprint,
-    }));
-    // audit_trend tool only supports keys, not files at the MCP layer.
-    // Document its actual wrapper semantics so future-us doesn't get
-    // surprised:
-    //   - bad keys → script exits 2 with stderr (no JSON payload)
-    //   - runScript() can't parse a {degraded:true} marker, so it
-    //     returns degraded:false / success:true / exitCode:2
-    // This is a real wrapper bug (success should not be true when
-    // exit!=0 AND no JSON came back), tracked separately. Asserting
-    // current behavior here protects against silent semantic shifts.
-    // iter 46 — file-input path. audit_trend now accepts baselineFile/currentFile.
-    const rFiles = await trendTool.handler({ baselineFile: basePath, currentFile: currPath });
-    assert(rFiles.success === true,
-      'audit_trend file-input path: success === true (iter 46)');
-    assert(rFiles.exitCode === 0, 'audit_trend file-input path: exitCode === 0');
-    assert(typeof rFiles.data?.delta === 'object',
-      'audit_trend file-input path: data.delta object present');
-    assert(rFiles.data?.delta?.structuralDistance?.verdict === 'near-identical',
-      `audit_trend file-input path: identical fingerprints → near-identical (got ${rFiles.data?.delta?.structuralDistance?.verdict})`);
-
-    // iter 54 — metaharness_drift_from_history positive case
-    const driftTool = tools.find((t) => t.name === 'metaharness_drift_from_history');
-    if (driftTool) {
-      // iter 71 — verify iter-66/67 fast-path flags are now MCP-callable
-      // Synthesize a baseline file on disk; pass via the new baselineFile input.
-      const baselinePath = pjoin(tmp, 'drift-baseline.json');
-      writeFileSync(baselinePath, JSON.stringify({
-        startedAt: '2026-06-16T00:00:00Z',
-        composite: { worst: 'clean' },
-        components: { oiaManifest: {}, threatModel: {}, mcpScan: { json: { findings: [] } } },
-        fingerprint: {
-          score: { harnessFit: 82, recommendedMode: 'CLI + MCP', archetype: 'typescript-sdk-harness', template: 'vertical:coding' },
-          genome: { repo_type: 'node_mcp_ci', agent_topology: ['m', 't'], risk_score: 0.3 },
-        },
-      }));
-      const rFastFast = await driftTool.handler({
-        path: '.', dryRun: true, threshold: 0.5, baselineFile: baselinePath,
-      });
-      if (!rFastFast.degraded) {
-        assert(rFastFast.data?.timing?.usedBaselineFile === true,
-          'drift_from_history MCP-layer: baselineFile fastpath fires (iter 71)');
-        assert(rFastFast.data?.timing?.skippedAuditList === true,
-          'drift_from_history MCP-layer: skippedAuditList=true via baselineFile (iter 71)');
-      }
-
-      // iter 85 — verify iter-78's alertOnNewSeverity MCP input plumbs
-      // through. baselineFile has no findings; current ruflo audit has
-      // 1 INFO finding. With alertOnNewSeverity='info' the gate fires
-      // and surfaces in the response.
-      const baselineNoFindings = pjoin(tmp, 'drift-baseline-no-findings.json');
-      writeFileSync(baselineNoFindings, JSON.stringify({
-        startedAt: '2026-06-16T00:00:00Z',
-        composite: { worst: 'clean' },
-        components: { oiaManifest: {}, threatModel: {}, mcpScan: { json: { findings: [] } } },
-        fingerprint: {
-          score: { harnessFit: 82, recommendedMode: 'CLI + MCP', archetype: 'typescript-sdk-harness', template: 'vertical:coding' },
-          genome: { repo_type: 'node_mcp_ci', agent_topology: ['m', 't'], risk_score: 0.3 },
-        },
-      }));
-      const rSevAlert = await driftTool.handler({
-        path: '.', dryRun: true, threshold: 0.5,
-        baselineFile: baselineNoFindings,
-        alertOnNewSeverity: 'info',
-      });
-      if (!rSevAlert.degraded) {
-        assert(rSevAlert.data?.alert?.newSeverityThreshold === 'info',
-          'drift_from_history MCP-layer: alertOnNewSeverity echoed in payload (iter 85)');
-        // Triggered AND exit code reflects (only if the audit actually had findings)
-        if (rSevAlert.data?.alert?.triggered === true) {
-          assert(rSevAlert.exitCode === 1,
-            `drift_from_history MCP-layer: alertOnNewSeverity exitCode=1 when triggered (got ${rSevAlert.exitCode})`);
-          assert(rSevAlert.success === false,
-            'drift_from_history MCP-layer: success===false when alert fires (iter 44 fix)');
-        }
-      }
-
-      const r54 = await driftTool.handler({ path: '.', dryRun: true, threshold: 0.5 });
-      if (!r54.degraded) {
-        assert(typeof r54.data === 'object' && r54.data !== null,
-          'drift_from_history positive: data is an object');
-        // Either it produced the structured drift report OR the no-history error
-        const isOk = r54.data?.command === 'drift-from-history';
-        const isNoHistory = typeof r54.data?.error === 'string' && r54.data.error.includes('no audit records');
-        assert(isOk || isNoHistory,
-          `drift_from_history positive: structured report OR no-history error (got ${JSON.stringify(r54.data).slice(0,80)})`);
-        if (isOk) {
-          assert(typeof r54.data.baseline?.key === 'string',
-            'drift_from_history: baseline.key is a string');
-          assert(typeof r54.data.alert?.threshold === 'number',
-            'drift_from_history: alert.threshold echoed numerically');
-        }
-      } else {
-        console.log(`    ⊘ drift_from_history: degraded (metaharness or memory absent)`);
-      }
-    }
-
-    const r = await trendTool.handler({ baselineKey: 'missing-X', currentKey: 'missing-Y' });
-    assert(r.exitCode === 2,
-      'audit_trend bad-keys path exits 2 (script-level guard fires)');
-    assert(r.data === null || r.data === undefined,
-      'audit_trend bad-keys path: data null (no JSON emitted on stderr exit)');
-    // iter 44 — success semantic anchor. Pre-iter-44 wrapper returned
-    // success:true for this case (because no degraded marker). Now
-    // returns false because exitCode !== 0.
-    assert(r.success === false,
-      'audit_trend bad-keys path: success === false (iter 44 fix)');
-  }
-
-  // Cleanup
-  try { (await import('node:fs')).rmSync(tmp, { recursive: true, force: true }); } catch { /* ignore */ }
-
-  console.log(`\n${passed} passed, ${failed} failed`);
-  if (failed > 0) {
-    console.log('\nFailures:');
-    for (const f of failures) console.log(`  - ${f}`);
-    process.exit(1);
-  }
-  console.log('\n✓ All 15 MCP tools satisfy the runtime contract.');
-}
-
-main().catch((e) => {
-  console.error('test-mcp-tools crashed:', e.message || e);
-  process.exit(2);
-});
+#!/usr/bin/env node
+// test-mcp-tools.mjs — runtime test for the iter-20/21 MCP tool registry.
+//
+// tsc proves the metaharness-tools.ts module COMPILES; structural smoke
+// proves the source DECLARES the right tool names. Neither proves the
+// HANDLERS actually run without throwing. This test imports the compiled
+// module and invokes every tool's handler with minimal inputs.
+//
+// CONTRACT EACH TOOL MUST SATISFY
+//   - handler is callable as `await tool.handler({ ... })`
+//   - returns an object with keys: success, data, degraded, exitCode
+//   - never throws (even with bad/missing optional dep — graceful)
+//   - handler honors the 120s subprocess timeout (no hang)
+//
+// USAGE
+//   node scripts/test-mcp-tools.mjs                          # default
+//   node scripts/test-mcp-tools.mjs --format json
+//
+// EXIT CODES
+//   0  all tools satisfy the contract
+//   1  at least one tool failed
+//   2  setup error (compiled dist not present)
+
+import { existsSync } from 'node:fs';
+import { dirname, join, resolve } from 'node:path';
+import { fileURLToPath } from 'node:url';
+
+const SCRIPTS_DIR = dirname(fileURLToPath(import.meta.url));
+const ARGS = (() => {
+  const a = { format: 'table' };
+  for (let i = 2; i < process.argv.length; i++) {
+    if (process.argv[i] === '--format') a.format = process.argv[++i];
+  }
+  return a;
+})();
+
+let passed = 0, failed = 0;
+const failures = [];
+
+function assert(cond, label) {
+  if (cond) { console.log(`  ✓ ${label}`); passed++; }
+  else { console.log(`  ✗ ${label}`); failures.push(label); failed++; }
+}
+
+async function main() {
+  // Locate the compiled dist of metaharness-tools.
+  const distPath = resolve(SCRIPTS_DIR, '..', '..', '..',
+    'v3', '@claude-flow', 'cli', 'dist', 'src', 'mcp-tools', 'metaharness-tools.js');
+
+  if (!existsSync(distPath)) {
+    console.log(`# test-mcp-tools — SKIPPED`);
+    console.log('');
+    console.log(`Compiled dist not present: ${distPath}`);
+    console.log(`Build the CLI first:`);
+    console.log(`  cd v3/@claude-flow/cli && npm run build`);
+    console.log('');
+    console.log(`Exit 0 — this script is meaningfully runnable only post-build.`);
+    process.exit(0);
+  }
+
+  let mod;
+  try {
+    mod = await import(distPath);
+  } catch (e) {
+    console.error(`test-mcp-tools: failed to import ${distPath}: ${e.message}`);
+    process.exit(2);
+  }
+
+  const tools = mod.metaharnessTools;
+  console.log(`# test-mcp-tools — runtime contract\n`);
+
+  // ──────────────────────────────────────────────────────────────────
+  // PHASE 1 — module exports the right shape
+  // ──────────────────────────────────────────────────────────────────
+  console.log('Phase 1 — module shape');
+  assert(Array.isArray(tools), 'metaharnessTools is an array');
+  assert(tools.length === 16, `16 tools registered (got ${tools.length})`);
+
+  const expectedNames = new Set([
+    'metaharness_score',
+    'metaharness_genome',
+    'metaharness_mcp_scan',
+    'metaharness_threat_model',
+    'metaharness_oia_audit',
+    'metaharness_audit_list',
+    'metaharness_audit_trend',
+    // iter 36 — ADR-152 §3.1 production
+    'metaharness_similarity',
+    // iter 54 — one-command drift detection (composes audit-list + oia-audit + audit-trend)
+    'metaharness_drift_from_history',
+    // ADR-153 — bench suites + evolve driver + security-focused bench
+    'metaharness_bench',
+    'metaharness_evolve',
+    'metaharness_security_bench',
+    // @metaharness/redblue@~0.1.4 — adversarial red/blue LLM testing
+    'metaharness_redblue',
+    // metaharness@0.3.0 — upstream ADR-235 GEPA learning run
+    'metaharness_learn',
+    // @metaharness/darwin@0.8.0 — GEPA library surface (genome ops)
+    'metaharness_gepa',
+    // ADR-322 — in-process governed evaluation and atomic promotion
+    'metaharness_flywheel',
+  ]);
+  const actualNames = new Set(tools.map((t) => t.name));
+  for (const name of expectedNames) {
+    assert(actualNames.has(name), `${name} registered`);
+  }
+
+  // ──────────────────────────────────────────────────────────────────
+  // PHASE 2 — every tool has the required MCP shape
+  // ──────────────────────────────────────────────────────────────────
+  console.log('\nPhase 2 — per-tool shape');
+  for (const tool of tools) {
+    const ok = typeof tool.name === 'string'
+      && typeof tool.description === 'string'
+      && typeof tool.category === 'string'
+      && typeof tool.handler === 'function'
+      && typeof tool.inputSchema === 'object';
+    assert(ok, `${tool.name} has {name, description, category, handler, inputSchema}`);
+    assert(tool.category === 'metaharness', `${tool.name} category === 'metaharness'`);
+  }
+
+  // ──────────────────────────────────────────────────────────────────
+  // PHASE 3 — handlers callable + return contract shape
+  //
+  // We invoke each handler with minimal valid input. The handlers may
+  // succeed (if metaharness is installed) or report degraded (if not).
+  // EITHER way, they must return { success, data, degraded, exitCode }
+  // without throwing.
+  // ──────────────────────────────────────────────────────────────────
+  console.log('\nPhase 3 — handler invocations (allow up to 30s each)');
+  for (const tool of tools) {
+    // Construct minimal valid input per tool.
+    let input = {};
+    if (tool.name === 'metaharness_audit_trend') {
+      // Requires baselineKey + currentKey — use fake keys that won't
+      // resolve so we exercise the not-found path.
+      input = { baselineKey: 'audit-fake-base', currentKey: 'audit-fake-curr' };
+    }
+    if (tool.name === 'metaharness_similarity') {
+      // Needs --a/--b OR --a-key/--b-key. Use fake mem keys to exercise
+      // the graceful not-found path (matches audit_trend convention).
+      input = { aKey: 'harness-fake-a', bKey: 'harness-fake-b' };
+    }
+    if (tool.name === 'metaharness_drift_from_history') {
+      // iter 54 — composes 3 subprocesses, needs more time than the default.
+      input = { dryRun: true, threshold: 0.5 };
+    }
+    if (tool.name === 'metaharness_oia_audit') {
+      // iter 128 — composite audit runs 5 sub-audits (oia-manifest +
+      // threat-model + mcp-scan + score + genome) in parallel. Each
+      // shells out via npx. --dry-run skips memory persistence so the
+      // test doesn't pollute namespaces.
+      input = { dryRun: true };
+    }
+    if (tool.name === 'metaharness_redblue') {
+      // `attack` preview is the fastest path that exercises the upstream
+      // binary without needing OPENROUTER_API_KEY or running any model
+      // calls. Count=1 keeps cold-cache npx fetch the dominant cost.
+      input = { subcommand: 'attack', family: 'prompt', count: 1 };
+    }
+    if (tool.name === 'metaharness_learn') {
+      // No repo checkout in CI → structured {status:"checkout-required"}
+      // exit-0 path. $0: without run=true upstream never spends anyway.
+      input = {};
+    }
+    if (tool.name === 'metaharness_gepa') {
+      // op is required; `genome` loads + validates the SHIPPED cand-6
+      // genome — pure-local library call once darwin is cached.
+      input = { op: 'genome' };
+    }
+
+    // iter 124 → 130 — timeouts have crept up as CI cold-cache npx
+    // warmup costs got measured. Final budgets:
+    //   default          : 60s
+    //   chain-tools      : 180s  (drift_from_history + oia_audit + audit_list)
+    // iter 131 — bumped chain-tool budget 90s → 180s. audit_list still
+    // timed out at 90s in CI; locally it runs in ~4s, but CI's
+    // `npx @claude-flow/cli@latest memory list` invocation pays both
+    // the npx fetch AND a full CLI startup (which loads agentic-flow +
+    // ONNX). 180s gives 30x headroom over the local cost.
+    const isChainTool = tool.name === 'metaharness_drift_from_history'
+      || tool.name === 'metaharness_oia_audit'
+      || tool.name === 'metaharness_audit_list'
+      // redblue: `attack prompt --count 1` is preview-only (no model
+      // calls) but the cold-cache `npx -y @metaharness/redblue@~0.1.4`
+      // fetch can take 30-60s. 180s gives 3x headroom.
+      || tool.name === 'metaharness_redblue'
+      // learn: cold-cache `npx -y metaharness@latest` fetch dominates.
+      // gepa: one-time `npm install --prefix ~/.ruflo/darwin-cache-*`
+      // fallback install can take 30-60s on cold cache.
+      || tool.name === 'metaharness_learn'
+      || tool.name === 'metaharness_gepa';
+    const timeoutMs = isChainTool ? 180_000 : 60_000;
+    const handlerPromise = tool.handler(input);
+    const timeoutPromise = new Promise((_, reject) =>
+      setTimeout(() => reject(new Error(`${timeoutMs / 1000}s handler timeout`)), timeoutMs));
+
+    let result;
+    let threw = false;
+    try {
+      result = await Promise.race([handlerPromise, timeoutPromise]);
+    } catch (e) {
+      threw = true;
+      console.log(`    [${tool.name}] handler threw: ${e.message.slice(0, 80)}`);
+    }
+
+    assert(!threw, `${tool.name} handler did not throw`);
+    if (!threw && result) {
+      assert(typeof result === 'object', `${tool.name} returns object`);
+      assert('success' in result, `${tool.name} result has 'success'`);
+      assert('data' in result, `${tool.name} result has 'data'`);
+      assert('degraded' in result, `${tool.name} result has 'degraded'`);
+      assert('exitCode' in result, `${tool.name} result has 'exitCode'`);
+    }
+  }
+
+  // ──────────────────────────────────────────────────────────────────
+  // PHASE 4 — POSITIVE-CASE data-shape validation (iter 43)
+  //
+  // Iter 37 verified the {success, data, degraded, exitCode} envelope.
+  // It did NOT verify that data.X contains the right keys when success
+  // is genuinely true — leaving room for iter 42-style bugs where a
+  // handler returns valid-looking degraded JSON while silently
+  // misrouting input. This phase invokes each handler with VALID
+  // inputs and asserts the expected output shape.
+  //
+  // Tools that depend on `npx metaharness` (score/genome/mcp-scan/
+  // threat-model/oia-audit/audit-list/audit-trend) are SKIPPED in this
+  // phase when the optional dep isn't installed — they're covered by
+  // the no-metaharness-smoke workflow's drill. The similarity tool
+  // has no @metaharness/* dep, so its positive case ALWAYS runs.
+  // ──────────────────────────────────────────────────────────────────
+  console.log('\nPhase 4 — positive-case data shape (iter 43)');
+
+  const { writeFileSync, mkdtempSync, mkdirSync } = await import('node:fs');
+  const { tmpdir } = await import('node:os');
+  const { join: pjoin } = await import('node:path');
+  const tmp = mkdtempSync(pjoin(tmpdir(), 'mcp-positive-'));
+
+  // #2626 — upstream deliberately exits 2 for a valid `blocked` genome.
+  // The wrapper must preserve that report as data instead of converting it
+  // into an input/system error at the MCP boundary.
+  const genomeTool = tools.find((t) => t.name === 'metaharness_genome');
+  if (genomeTool) {
+    const blockedRepo = pjoin(tmp, 'blocked-repo');
+    mkdirSync(blockedRepo);
+    const r = await genomeTool.handler({ path: blockedRepo });
+    if (!r.degraded) {
+      assert(r.success === true,
+        'genome blocked verdict: wrapper succeeds with a valid report (#2626)');
+      assert(r.exitCode === 0,
+        'genome blocked verdict: wrapper exitCode === 0 (#2626)');
+      assert(r.data?.verdict === 'blocked',
+        `genome blocked verdict: data.verdict === blocked (got ${r.data?.verdict})`);
+      assert(r.data?.verdictExitCode === 2,
+        `genome blocked verdict: preserves upstream exit 2 (got ${r.data?.verdictExitCode})`);
+      assert(typeof r.data?.risk_score === 'number' && r.data.risk_score >= 0.7,
+        `genome blocked verdict: risk_score >= 0.7 (got ${r.data?.risk_score})`);
+    } else {
+      console.log(`    ⊘ genome: metaharness absent — graceful skip`);
+    }
+  }
+
+  // metaharness_similarity — full positive case (no @metaharness/* needed)
+  const simTool = tools.find((t) => t.name === 'metaharness_similarity');
+  if (simTool) {
+    const aPath = pjoin(tmp, 'a.json');
+    const bPath = pjoin(tmp, 'b.json');
+    writeFileSync(aPath, JSON.stringify({
+      score: { harnessFit: 78, compileConfidence: 92, taskCoverage: 65, toolSafety: 88, memoryUsefulness: 70, estCostPerRunUsd: 0.04, recommendedMode: 'CLI + MCP', archetype: 'compliance-harness', template: 'vertical:legal' },
+      genome: { repo_type: 'node_mcp_ci', agent_topology: ['x','y','z','w'], risk_score: 0.45, test_confidence: 0.7, publish_readiness: 0.6 },
+    }));
+    writeFileSync(bPath, JSON.stringify({
+      score: { harnessFit: 75, compileConfidence: 90, taskCoverage: 70, toolSafety: 90, memoryUsefulness: 72, estCostPerRunUsd: 0.05, recommendedMode: 'CLI + MCP', archetype: 'compliance-harness', template: 'vertical:support' },
+      genome: { repo_type: 'node_mcp_ci', agent_topology: ['x','y','z','q','r'], risk_score: 0.40, test_confidence: 0.75, publish_readiness: 0.65 },
+    }));
+    const r = await simTool.handler({ aFile: aPath, bFile: bPath });
+    assert(r.degraded === false, 'similarity positive case: degraded === false');
+    assert(r.success === true, 'similarity positive case: success === true');
+    assert(r.exitCode === 0, 'similarity positive case: exitCode === 0');
+    const d = r.data ?? {};
+    assert(typeof d.overall === 'number', 'similarity data has numeric `overall`');
+    assert(typeof d.components === 'object' && d.components !== null,
+      'similarity data has `components` object');
+    assert(typeof d.components?.cosine === 'number',
+      'similarity components.cosine numeric');
+    assert(typeof d.components?.categorical === 'number',
+      'similarity components.categorical numeric');
+    assert(typeof d.components?.jaccard === 'number',
+      'similarity components.jaccard numeric');
+    assert(typeof d.weights === 'object' && d.weights !== null,
+      'similarity data has `weights` object');
+    assert(d.adr === 'ADR-152', 'similarity data tagged adr=ADR-152');
+    // Regression anchor — same fixtures as iter-35 spike with non-matching topologies
+    assert(d.overall > 0 && d.overall < 1,
+      `similarity overall in (0, 1) — got ${d.overall}`);
+
+    // Per-dimension variant
+    const rPD = await simTool.handler({ aFile: aPath, bFile: bPath, perDimension: true });
+    assert(typeof rPD.data?.perDimension === 'object',
+      'similarity perDimension=true populates breakdown');
+
+    // Alert-below variant exercises non-zero exit
+    const rAlert = await simTool.handler({ aFile: aPath, bFile: bPath, alertBelow: 0.99 });
+    assert(rAlert.data?.alert?.triggered === true,
+      'similarity alertBelow=0.99 triggers alert');
+    assert(rAlert.exitCode === 1, 'similarity alertBelow=0.99 → exitCode 1');
+    // iter 44 — success semantic anchor (was true under the pre-iter-44
+    // `!degraded` rule; now false because exitCode !== 0 dominates).
+    assert(rAlert.success === false,
+      'similarity alertBelow=0.99 → success === false (iter 44 fix)');
+  }
+
+  // metaharness_mcp_scan — positive case post iter-50 parser landing.
+  // Until iter 50, mcp_scan's data field was an alert-only object with
+  // no structured findings. After iter 50, findings[] is always present
+  // (parsed from upstream text) and summary{overallSeverity, totalCount}
+  // accompanies it.
+  const scanTool = tools.find((t) => t.name === 'metaharness_mcp_scan');
+  if (scanTool) {
+    // Run against ruflo itself — guaranteed to produce at least the
+    // INFO finding the iter-50 parser test verified manually.
+    const r = await scanTool.handler({ path: '.', failOn: 'high' });
+    // Either succeeds with structured findings, or gracefully degrades
+    // if metaharness isn't installed in this environment.
+    if (!r.degraded) {
+      assert(r.success === true, 'mcp_scan positive: success === true');
+      assert(r.exitCode === 0, 'mcp_scan positive: exitCode === 0');
+      assert(Array.isArray(r.data?.findings),
+        'mcp_scan positive: data.findings is an array (iter 50 fix)');
+      // Cwd-dependent: when scanning a dir without .mcp/servers.json the
+      // upstream emits no findings. Only verify shape contract when array
+      // is populated — the array-presence assertion above is the
+      // load-bearing one for iter 50.
+      if (r.data?.findings.length > 0) {
+        const first = r.data.findings[0];
+        assert(typeof first?.severity === 'string',
+          'mcp_scan positive: first finding has string severity');
+        assert(typeof first?.message === 'string',
+          'mcp_scan positive: first finding has string message');
+      }
+      // summary may be null if the upstream produced no Result: line —
+      // verify the field's presence (null OR object) but only deep-check
+      // when populated.
+      if (r.data?.summary) {
+        assert(typeof r.data.summary.totalCount === 'number',
+          'mcp_scan positive: data.summary.totalCount is numeric (when summary present)');
+      }
+    } else {
+      console.log(`    ⊘ mcp_scan: metaharness absent — graceful skip`);
+    }
+  }
+
+  // metaharness_audit_trend — positive case via file inputs
+  const trendTool = tools.find((t) => t.name === 'metaharness_audit_trend');
+  if (trendTool) {
+    const basePath = pjoin(tmp, 'base.json');
+    const currPath = pjoin(tmp, 'curr.json');
+    const fingerprint = {
+      score: { harnessFit: 80, recommendedMode: 'CLI + MCP', archetype: 'typescript-sdk-harness', template: 'vertical:coding' },
+      genome: { repo_type: 'node_mcp_ci', agent_topology: ['a','b','c'], risk_score: 0.3, test_confidence: 0.85, publish_readiness: 0.9 },
+    };
+    writeFileSync(basePath, JSON.stringify({
+      startedAt: '2026-06-15T00:00:00Z',
+      composite: { worst: 'clean' },
+      components: { oiaManifest: {}, threatModel: {}, mcpScan: { json: { findings: [] } } },
+      fingerprint,
+    }));
+    writeFileSync(currPath, JSON.stringify({
+      startedAt: '2026-06-16T00:00:00Z',
+      composite: { worst: 'clean' },
+      components: { oiaManifest: {}, threatModel: {}, mcpScan: { json: { findings: [] } } },
+      fingerprint,
+    }));
+    // audit_trend tool only supports keys, not files at the MCP layer.
+    // Document its actual wrapper semantics so future-us doesn't get
+    // surprised:
+    //   - bad keys → script exits 2 with stderr (no JSON payload)
+    //   - runScript() can't parse a {degraded:true} marker, so it
+    //     returns degraded:false / success:true / exitCode:2
+    // This is a real wrapper bug (success should not be true when
+    // exit!=0 AND no JSON came back), tracked separately. Asserting
+    // current behavior here protects against silent semantic shifts.
+    // iter 46 — file-input path. audit_trend now accepts baselineFile/currentFile.
+    const rFiles = await trendTool.handler({ baselineFile: basePath, currentFile: currPath });
+    assert(rFiles.success === true,
+      'audit_trend file-input path: success === true (iter 46)');
+    assert(rFiles.exitCode === 0, 'audit_trend file-input path: exitCode === 0');
+    assert(typeof rFiles.data?.delta === 'object',
+      'audit_trend file-input path: data.delta object present');
+    assert(rFiles.data?.delta?.structuralDistance?.verdict === 'near-identical',
+      `audit_trend file-input path: identical fingerprints → near-identical (got ${rFiles.data?.delta?.structuralDistance?.verdict})`);
+
+    // iter 54 — metaharness_drift_from_history positive case
+    const driftTool = tools.find((t) => t.name === 'metaharness_drift_from_history');
+    if (driftTool) {
+      // iter 71 — verify iter-66/67 fast-path flags are now MCP-callable
+      // Synthesize a baseline file on disk; pass via the new baselineFile input.
+      const baselinePath = pjoin(tmp, 'drift-baseline.json');
+      writeFileSync(baselinePath, JSON.stringify({
+        startedAt: '2026-06-16T00:00:00Z',
+        composite: { worst: 'clean' },
+        components: { oiaManifest: {}, threatModel: {}, mcpScan: { json: { findings: [] } } },
+        fingerprint: {
+          score: { harnessFit: 82, recommendedMode: 'CLI + MCP', archetype: 'typescript-sdk-harness', template: 'vertical:coding' },
+          genome: { repo_type: 'node_mcp_ci', agent_topology: ['m', 't'], risk_score: 0.3 },
+        },
+      }));
+      const rFastFast = await driftTool.handler({
+        path: '.', dryRun: true, threshold: 0.5, baselineFile: baselinePath,
+      });
+      if (!rFastFast.degraded) {
+        assert(rFastFast.data?.timing?.usedBaselineFile === true,
+          'drift_from_history MCP-layer: baselineFile fastpath fires (iter 71)');
+        assert(rFastFast.data?.timing?.skippedAuditList === true,
+          'drift_from_history MCP-layer: skippedAuditList=true via baselineFile (iter 71)');
+      }
+
+      // iter 85 — verify iter-78's alertOnNewSeverity MCP input plumbs
+      // through. baselineFile has no findings; current ruflo audit has
+      // 1 INFO finding. With alertOnNewSeverity='info' the gate fires
+      // and surfaces in the response.
+      const baselineNoFindings = pjoin(tmp, 'drift-baseline-no-findings.json');
+      writeFileSync(baselineNoFindings, JSON.stringify({
+        startedAt: '2026-06-16T00:00:00Z',
+        composite: { worst: 'clean' },
+        components: { oiaManifest: {}, threatModel: {}, mcpScan: { json: { findings: [] } } },
+        fingerprint: {
+          score: { harnessFit: 82, recommendedMode: 'CLI + MCP', archetype: 'typescript-sdk-harness', template: 'vertical:coding' },
+          genome: { repo_type: 'node_mcp_ci', agent_topology: ['m', 't'], risk_score: 0.3 },
+        },
+      }));
+      const rSevAlert = await driftTool.handler({
+        path: '.', dryRun: true, threshold: 0.5,
+        baselineFile: baselineNoFindings,
+        alertOnNewSeverity: 'info',
+      });
+      if (!rSevAlert.degraded) {
+        assert(rSevAlert.data?.alert?.newSeverityThreshold === 'info',
+          'drift_from_history MCP-layer: alertOnNewSeverity echoed in payload (iter 85)');
+        // Triggered AND exit code reflects (only if the audit actually had findings)
+        if (rSevAlert.data?.alert?.triggered === true) {
+          assert(rSevAlert.exitCode === 1,
+            `drift_from_history MCP-layer: alertOnNewSeverity exitCode=1 when triggered (got ${rSevAlert.exitCode})`);
+          assert(rSevAlert.success === false,
+            'drift_from_history MCP-layer: success===false when alert fires (iter 44 fix)');
+        }
+      }
+
+      const r54 = await driftTool.handler({ path: '.', dryRun: true, threshold: 0.5 });
+      if (!r54.degraded) {
+        assert(typeof r54.data === 'object' && r54.data !== null,
+          'drift_from_history positive: data is an object');
+        // Either it produced the structured drift report OR the no-history error
+        const isOk = r54.data?.command === 'drift-from-history';
+        const isNoHistory = typeof r54.data?.error === 'string' && r54.data.error.includes('no audit records');
+        assert(isOk || isNoHistory,
+          `drift_from_history positive: structured report OR no-history error (got ${JSON.stringify(r54.data).slice(0,80)})`);
+        if (isOk) {
+          assert(typeof r54.data.baseline?.key === 'string',
+            'drift_from_history: baseline.key is a string');
+          assert(typeof r54.data.alert?.threshold === 'number',
+            'drift_from_history: alert.threshold echoed numerically');
+        }
+      } else {
+        console.log(`    ⊘ drift_from_history: degraded (metaharness or memory absent)`);
+      }
+    }
+
+    const r = await trendTool.handler({ baselineKey: 'missing-X', currentKey: 'missing-Y' });
+    assert(r.exitCode === 2,
+      'audit_trend bad-keys path exits 2 (script-level guard fires)');
+    assert(r.data === null || r.data === undefined,
+      'audit_trend bad-keys path: data null (no JSON emitted on stderr exit)');
+    // iter 44 — success semantic anchor. Pre-iter-44 wrapper returned
+    // success:true for this case (because no degraded marker). Now
+    // returns false because exitCode !== 0.
+    assert(r.success === false,
+      'audit_trend bad-keys path: success === false (iter 44 fix)');
+  }
+
+  // Cleanup
+  try { (await import('node:fs')).rmSync(tmp, { recursive: true, force: true }); } catch { /* ignore */ }
+
+  console.log(`\n${passed} passed, ${failed} failed`);
+  if (failed > 0) {
+    console.log('\nFailures:');
+    for (const f of failures) console.log(`  - ${f}`);
+    process.exit(1);
+  }
+  console.log('\n✓ All 15 MCP tools satisfy the runtime contract.');
+}
+
+main().catch((e) => {
+  console.error('test-mcp-tools crashed:', e.message || e);
+  process.exit(2);
+});