cool-workflow 0.1.97 → 0.2.0

package/dist/shell/eval-io.js

@@ -0,0 +1,473 @@
+"use strict";
+// shell/eval-io.ts — createMultiAgentReplaySnapshot/replayMultiAgentSnapshot/
+// compareMultiAgentReplay/scoreMultiAgentReplay/gateMultiAgentEval/
+// reportMultiAgentEval: the impure disk-orchestrated wrapper around
+// core/multi-agent/eval-replay.ts's pure 31-metric compare/score/gate/
+// report + normalizeRun projection.
+//
+// MILESTONE 9. Byte-exact port of the impure half of the old build's
+// src/multi-agent-eval.ts: snapshot/suite/comparison/score/gate/report
+// file writes, target-path resolution, and — the load-bearing piece —
+// replayMultiAgentSnapshot RE-DERIVING the projection from the raw
+// baseline run state file rather than copying snapshot.normalized.
+//
+// BYTE-COMPAT / REBUILD RISK 5 [load-bearing]: see
+// eval-replay-detects-drift.case.js.
+//
+// Evidence: SPEC/multi-agent.md section I, "Eval harness exact outputs";
+// plugins/cool-workflow/src/multi-agent-eval.ts (byte-exact source for
+// the wiring/resolution sequence).
+var __createBinding = (this && this.__createBinding) || (Object.create ? (function(o, m, k, k2) {
+    if (k2 === undefined) k2 = k;
+    var desc = Object.getOwnPropertyDescriptor(m, k);
+    if (!desc || ("get" in desc ? !m.__esModule : desc.writable || desc.configurable)) {
+      desc = { enumerable: true, get: function() { return m[k]; } };
+    }
+    Object.defineProperty(o, k2, desc);
+}) : (function(o, m, k, k2) {
+    if (k2 === undefined) k2 = k;
+    o[k2] = m[k];
+}));
+var __setModuleDefault = (this && this.__setModuleDefault) || (Object.create ? (function(o, v) {
+    Object.defineProperty(o, "default", { enumerable: true, value: v });
+}) : function(o, v) {
+    o["default"] = v;
+});
+var __importStar = (this && this.__importStar) || (function () {
+    var ownKeys = function(o) {
+        ownKeys = Object.getOwnPropertyNames || function (o) {
+            var ar = [];
+            for (var k in o) if (Object.prototype.hasOwnProperty.call(o, k)) ar[ar.length] = k;
+            return ar;
+        };
+        return ownKeys(o);
+    };
+    return function (mod) {
+        if (mod && mod.__esModule) return mod;
+        var result = {};
+        if (mod != null) for (var k = ownKeys(mod), i = 0; i < k.length; i++) if (k[i] !== "default") __createBinding(result, mod, k[i]);
+        __setModuleDefault(result, mod);
+        return result;
+    };
+})();
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.createMultiAgentReplaySnapshot = createMultiAgentReplaySnapshot;
+exports.replayMultiAgentSnapshot = replayMultiAgentSnapshot;
+exports.compareMultiAgentReplay = compareMultiAgentReplay;
+exports.scoreMultiAgentReplay = scoreMultiAgentReplay;
+exports.gateMultiAgentEval = gateMultiAgentEval;
+exports.reportMultiAgentEval = reportMultiAgentEval;
+const fs = __importStar(require("node:fs"));
+const path = __importStar(require("node:path"));
+const fs_atomic_1 = require("./fs-atomic");
+const run_store_1 = require("./run-store");
+const ev = __importStar(require("../core/multi-agent/eval-replay"));
+const trust_policy_io_1 = require("./trust-policy-io");
+const topology_io_1 = require("./topology-io");
+const multi_agent_io_1 = require("./multi-agent-io");
+const evidence_reasoning_1 = require("./evidence-reasoning");
+const state_explosion_cli_1 = require("./state-explosion-cli");
+const multi_agent_operator_ux_1 = require("./multi-agent-operator-ux");
+function now() {
+    return new Date().toISOString();
+}
+function evalSuiteDir(cwd, suiteId) {
+    return path.join(cwd, ".cw", "evals", (0, fs_atomic_1.safeFileName)(suiteId));
+}
+function resolveTargetPath(target) {
+    if (!target)
+        throw new Error("Missing eval target");
+    return path.isAbsolute(target) ? target : path.resolve(target);
+}
+function resolveSnapshotPath(target) {
+    const resolved = resolveTargetPath(target);
+    if (fs.existsSync(resolved) && fs.statSync(resolved).isDirectory())
+        return path.join(resolved, "snapshot.json");
+    if (fs.existsSync(resolved))
+        return resolved;
+    return path.join(process.cwd(), ".cw", "evals", (0, fs_atomic_1.safeFileName)(target), "snapshot.json");
+}
+function resolveReplayPath(target) {
+    const resolved = resolveTargetPath(target);
+    if (fs.existsSync(resolved) && fs.statSync(resolved).isDirectory())
+        return path.join(resolved, "replay-run.json");
+    if (fs.existsSync(resolved))
+        return resolved;
+    return path.join(process.cwd(), ".cw", "evals", (0, fs_atomic_1.safeFileName)(target), "replay-run.json");
+}
+function resolveSuiteDir(target) {
+    const resolved = resolveTargetPath(target);
+    if (fs.existsSync(resolved)) {
+        if (fs.statSync(resolved).isDirectory())
+            return resolved;
+        const value = (0, fs_atomic_1.readJson)(resolved);
+        if (value.paths?.suiteDir)
+            return value.paths.suiteDir;
+        return path.dirname(resolved);
+    }
+    return path.join(process.cwd(), ".cw", "evals", (0, fs_atomic_1.safeFileName)(target));
+}
+function writeSuite(suite) {
+    (0, fs_atomic_1.writeJson)(path.join(suite.paths.suiteDir, "suite.json"), suite);
+}
+function loadSuiteFromDir(suiteDir) {
+    const suitePath = path.join(suiteDir, "suite.json");
+    if (fs.existsSync(suitePath))
+        return (0, fs_atomic_1.readJson)(suitePath);
+    return { schemaVersion: 1, id: path.basename(suiteDir), title: `Multi-Agent Eval Suite ${path.basename(suiteDir)}`, createdAt: now(), cases: [], paths: { suiteDir, snapshotPath: path.join(suiteDir, "snapshot.json") } };
+}
+function reportSections(run) {
+    if (!fs.existsSync(run.paths.report))
+        return [];
+    const text = fs.readFileSync(run.paths.report, "utf8");
+    return text
+        .split("\n")
+        .filter((line) => /^#+\s+/.test(line))
+        .map((line) => line.replace(/^#+\s+/, "").trim())
+        .sort();
+}
+function collectCandidateScores(run) {
+    const scores = [];
+    const candidates = run.candidates || [];
+    for (const candidate of candidates) {
+        for (const scoreId of candidate.scores || []) {
+            const scorePath = path.join(run.paths.candidatesDir || path.join(run.paths.runDir, "candidates"), (0, fs_atomic_1.safeFileName)(candidate.id), "scores", `${(0, fs_atomic_1.safeFileName)(scoreId)}.json`);
+            if (fs.existsSync(scorePath)) {
+                const score = (0, fs_atomic_1.readJson)(scorePath);
+                scores.push({ candidateId: candidate.id, scoreId, criteria: score.criteria, total: score.total, maxTotal: score.maxTotal, normalized: score.normalized, verdict: score.verdict, evidenceCount: Array.isArray(score.evidence) ? score.evidence.length : 0, notes: score.notes });
+            }
+            else {
+                scores.push({ candidateId: candidate.id, scoreId, missing: true });
+            }
+        }
+    }
+    return scores;
+}
+/** Minimal, real (non-operator-ux) dependency/failure derivation over
+ *  multi-agent + trust state: one row per membership that has not yet
+ *  reported, one row per policy violation. See core/multi-agent/eval-
+ *  replay.ts's file header for why this milestone's normalizeRun scope
+ *  is reduced from the full operator-ux module. */
+function dependencyRows(run) {
+    const multiAgent = run.multiAgent || {};
+    return (multiAgent.memberships || []).map((membership) => ({ from: `${run.id}:multi-agent:group:${membership.groupId}`, to: `${run.id}:multi-agent:membership:${membership.id}`, label: "depends-on", status: membership.status }));
+}
+function failureRows(run) {
+    const multiAgent = run.multiAgent || {};
+    const rows = [];
+    for (const membership of multiAgent.memberships || []) {
+        if (membership.status === "failed")
+            rows.push({ kind: "agent-membership", status: membership.status, owner: membership.id, reason: "failed membership" });
+    }
+    for (const fanin of multiAgent.fanins || []) {
+        for (const reason of fanin.blockedReasons || [])
+            rows.push({ kind: "fanin", status: "blocked", owner: fanin.id, reason });
+    }
+    // Plain single-agent pipeline path: a worker in a terminal non-verified
+    // status (failed/rejected) is a failure row too, so a mutated raw
+    // baseline state (see eval-replay-detects-drift.case.js) is visible
+    // in the re-derived normalized projection, not just in multi-agent
+    // state.
+    for (const worker of run.workers || []) {
+        if (worker.status === "failed" || worker.status === "rejected")
+            rows.push({ kind: "worker", status: worker.status, owner: worker.id, reason: `worker ${worker.id} ${worker.status}` });
+    }
+    return rows;
+}
+/** Evidence adoption rows: one per candidate-selection's evidence
+ *  (multi-agent path) PLUS one per accepted result node's evidence (the
+ *  plain single-agent pipeline path, where there is no candidate/
+ *  selection layer at all — see the file header's scope note: this
+ *  keeps the section non-trivial for a completed `-q`/`--drive` run,
+ *  matching the SPEC's own evidence-adoption model of "adopted through
+ *  acceptance" for every accepted result). */
+function evidenceAdoptionRows(run) {
+    const selections = run.candidateSelections || [];
+    const rows = [];
+    for (const selection of selections) {
+        for (const item of selection.evidence || [])
+            rows.push({ ref: item.id, status: "adopted", adoptedBy: selection.candidateId });
+    }
+    for (const node of run.nodes || []) {
+        if (node.kind !== "result")
+            continue;
+        for (const item of node.evidence || [])
+            rows.push({ ref: item.id, status: "adopted", adoptedBy: node.id });
+    }
+    return rows;
+}
+function normalizeRun(run) {
+    const trust = (0, trust_policy_io_1.summarizeMultiAgentTrust)(run);
+    const blackboard = run.blackboard || {};
+    const topologies = (0, topology_io_1.summarizeTopologies)(run);
+    const multiAgentSummary = (0, multi_agent_io_1.summarizeMultiAgent)(run);
+    const multiAgent = run.multiAgent || {};
+    return {
+        workflow: ev.normalizeValue({ id: run.workflow.id, appId: run.workflow.app?.id, appVersion: run.workflow.app?.version, taskCount: run.tasks.length }),
+        topologyShape: ev.lines([
+            topologies.active.map((entry) => ({ topologyId: entry.topologyId, status: entry.status, roleCount: entry.roles.length, groupCount: entry.groups.length, fanoutCount: entry.fanouts.length, faninCount: entry.fanins.length })),
+            multiAgentSummary.groupsDetail,
+        ]),
+        roles: ev.lines(multiAgent.roles || []),
+        groups: ev.lines(multiAgent.groups || []),
+        memberships: ev.lines(multiAgent.memberships || []),
+        fanouts: ev.lines(multiAgent.fanouts || []),
+        fanins: ev.lines(multiAgent.fanins || []),
+        dependencyEdges: ev.lines(dependencyRows(run)),
+        failures: ev.lines(failureRows(run)),
+        blackboardRecords: ev.lines([blackboard.boards || [], blackboard.topics || [], blackboard.messages || [], blackboard.contexts || [], blackboard.artifacts || [], blackboard.snapshots || [], blackboard.decisions || []]),
+        messageProvenance: ev.lines(trust.messageProvenance || []),
+        rolePolicies: ev.lines(trust.rolePolicies || []),
+        permissionDecisions: ev.lines(trust.permissionDecisions || []),
+        blackboardWriteAudit: ev.lines(trust.blackboardWrites || []),
+        judgeRationales: ev.lines(trust.judgeRationales || []),
+        panelDecisions: ev.lines(trust.panelDecisions || []),
+        policyViolations: ev.lines(trust.policyViolations || []),
+        evidenceAdoption: ev.lines(evidenceAdoptionRows(run)),
+        candidateScores: ev.lines(collectCandidateScores(run)),
+        selectedCandidates: ev.lines((run.candidateSelections || []).map((entry) => ({ candidateId: entry.candidateId, scoreId: entry.scoreId, verifierNodeId: entry.verifierNodeId, reason: entry.reason, evidenceCount: entry.evidence.length }))),
+        verifierCommitGate: ev.lines((run.commits || []).map((entry) => ({ verifierGated: Boolean(entry.verifierGated), checkpoint: Boolean(entry.checkpoint), candidateId: entry.candidateId, selectionId: entry.selectionId, verifierNodeId: entry.verifierNodeId, evidenceCount: (entry.evidence || []).length }))),
+        reportSections: reportSections(run),
+        ...normalizeStateExplosionForEval(run),
+        ...(0, evidence_reasoning_1.normalizeEvidenceReasoningForEval)(run),
+    };
+}
+/** The v0.1.25/v0.1.26 state-explosion eval sections — a stabilized
+ *  projection of the state-explosion report (summary freshness, compact
+ *  graph shape, blackboard digest, critical path, evidence digest,
+ *  expansion refs). Byte-behavior port of the old build's
+ *  normalizeStateExplosionForEval. */
+function normalizeStateExplosionForEval(run) {
+    const report = (0, state_explosion_cli_1.buildStateExplosionReport)(run, { operator: (0, multi_agent_operator_ux_1.operatorDigestInput)(run) });
+    const graph = report.compactGraph;
+    const runPrefix = `${run.id}:`;
+    const stripRunId = (id) => (id.startsWith(runPrefix) ? id.slice(runPrefix.length) : id);
+    return {
+        summaryFreshness: ev.lines([
+            { compactionRecommended: report.stateSize.compactionRecommended, total: report.stateSize.total, deterministic: graph.deterministic },
+        ]),
+        compactGraphShape: ev.lines([
+            {
+                view: graph.view,
+                fullNodeCount: graph.fullNodeCount,
+                fullEdgeCount: graph.fullEdgeCount,
+                compactNodeCount: graph.compactNodeCount,
+                compactEdgeCount: graph.compactEdgeCount,
+                collapsedNodeCount: graph.collapsedNodeCount,
+                syntheticNodes: graph.syntheticNodes.map((s) => ({ kind: s.id.split(":summary:")[1] || s.kind, collapsedNodeCount: s.collapsedNodeCount, collapsedEdgeCount: s.collapsedEdgeCount, dominantStatus: s.dominantStatus })),
+            },
+        ]),
+        blackboardDigest: ev.lines([
+            {
+                topics: report.blackboardDigest.topicRollups.length,
+                threads: report.blackboardDigest.threadSummaries.length,
+                unresolved: report.blackboardDigest.unresolvedQuestions.map((q) => q.id),
+                conflicts: report.blackboardDigest.conflicts.map((c) => c.id),
+                decisions: report.blackboardDigest.decisions.length,
+                artifacts: report.blackboardDigest.artifacts.length,
+                policyViolations: report.blackboardDigest.policyViolations.map((p) => p.id),
+                judgeRationale: report.blackboardDigest.judgeRationale.map((j) => j.id),
+                missingEvidence: report.blackboardDigest.missingEvidence.map((m) => m.label),
+            },
+        ]),
+        criticalPath: graph.criticalPath.map(stripRunId).sort(),
+        evidenceDigest: ev.lines([
+            { adopted: report.operatorDigest.evidenceDigest.adopted, missing: report.operatorDigest.evidenceDigest.missing, rejected: report.operatorDigest.evidenceDigest.rejected },
+        ]),
+        expansionRefs: report.hiddenSourceRecords.map((h) => `${h.kind}=${h.count}`).sort(),
+    };
+}
+function createMultiAgentReplaySnapshot(run, options = {}) {
+    const id = (0, fs_atomic_1.safeFileName)(String(options.id || options.snapshot || `${run.id}-snapshot`));
+    const suiteDir = evalSuiteDir(run.cwd, id);
+    const snapshotPath = path.join(suiteDir, "snapshot.json");
+    const snapshot = {
+        schemaVersion: 1,
+        kind: "multi-agent-replay-snapshot",
+        id,
+        createdAt: now(),
+        runId: run.id,
+        workflow: { id: run.workflow.id, appId: run.workflow.app?.id, appVersion: run.workflow.app?.version, title: run.workflow.title },
+        inputs: ev.normalizeValue(run.inputs),
+        paths: { suiteDir, snapshotPath, baselineStatePath: run.paths.state, reportPath: run.paths.report },
+        normalized: normalizeRun(run),
+    };
+    (0, fs_atomic_1.writeJson)(snapshotPath, snapshot);
+    writeSuite({ schemaVersion: 1, id, title: `Multi-Agent Eval Suite ${id}`, createdAt: snapshot.createdAt, cases: [{ id: `${id}-case`, snapshotId: id, baselinePath: snapshotPath, expectedVerdict: "pass" }], paths: { suiteDir, snapshotPath } });
+    return snapshot;
+}
+function assertSnapshotShape(snapshot, file) {
+    if (!snapshot.id)
+        throw new Error(`Replay snapshot missing id: ${file}`);
+    if (!snapshot.runId)
+        throw new Error(`Replay snapshot missing runId: ${file}`);
+    if (!snapshot.paths || !snapshot.paths.suiteDir || !snapshot.paths.snapshotPath)
+        throw new Error(`Replay snapshot missing paths.suiteDir or paths.snapshotPath: ${file}`);
+    ev.assertNormalizedShape(snapshot.normalized, `Replay snapshot missing normalized section: ${file}`);
+}
+function assertReplayShape(replay, file) {
+    if (!replay.id)
+        throw new Error(`Replay run missing id: ${file}`);
+    if (!replay.snapshotId)
+        throw new Error(`Replay run missing snapshotId: ${file}`);
+    if (replay.status !== "completed" && replay.status !== "failed")
+        throw new Error(`Replay run has unsupported status ${String(replay.status)}: ${file}`);
+    if (!replay.paths || !replay.paths.suiteDir || !replay.paths.replayRunPath || !replay.paths.snapshotPath)
+        throw new Error(`Replay run missing paths.suiteDir, paths.replayRunPath, or paths.snapshotPath: ${file}`);
+    if (!Array.isArray(replay.errors))
+        throw new Error(`Replay run errors must be an array: ${file}`);
+    ev.assertNormalizedShape(replay.replay, `Replay run missing replay section: ${file}`);
+}
+function loadSnapshot(target) {
+    const resolved = resolveSnapshotPath(target);
+    if (!fs.existsSync(resolved))
+        throw new Error(`File not found: ${resolved}`);
+    const snapshot = (0, fs_atomic_1.readJson)(resolved);
+    if (snapshot.kind !== "multi-agent-replay-snapshot")
+        throw new Error(`Not a replay snapshot: ${resolved}`);
+    assertSnapshotShape(snapshot, resolved);
+    return snapshot;
+}
+/** RE-DERIVE the normalized projection from the raw captured state file
+ *  instead of copying snapshot.normalized. Fail-closed: throws when the
+ *  baseline state cannot be reconstructed. See file header, byte-compat
+ *  / rebuild risk 5. */
+function rederiveNormalizedFromSnapshot(snapshot) {
+    const statePath = snapshot.paths.baselineStatePath;
+    if (!statePath || !fs.existsSync(statePath)) {
+        throw new Error(`Cannot re-derive replay projection: baseline run state missing at ${statePath || "<unset>"}; re-snapshot from a live run before replaying.`);
+    }
+    const result = (0, run_store_1.loadRunStateFile)(statePath, { dryRun: true });
+    if (result.report.status === "unsupported") {
+        throw new Error(`Cannot re-derive replay projection: baseline run state at ${statePath} is unsupported: ${result.report.errors.join("; ")}`);
+    }
+    return normalizeRun(result.run);
+}
+function replayMultiAgentSnapshot(target, options = {}) {
+    if (!target)
+        throw new Error("Missing snapshot id or path.");
+    const snapshot = loadSnapshot(target);
+    const replayId = (0, fs_atomic_1.safeFileName)(String(options.id || options.replay || `${snapshot.id}-replay`));
+    const suiteDir = snapshot.paths.suiteDir;
+    const replayDir = path.join(suiteDir, "replay");
+    const replayRunPath = path.join(suiteDir, "replay-run.json");
+    fs.mkdirSync(replayDir, { recursive: true });
+    const replayed = rederiveNormalizedFromSnapshot(snapshot);
+    const replay = {
+        schemaVersion: 1,
+        kind: "multi-agent-replay-run",
+        id: replayId,
+        snapshotId: snapshot.id,
+        baselineRunId: snapshot.runId,
+        replayedAt: now(),
+        status: "completed",
+        isolatedWorkspace: replayDir,
+        paths: { suiteDir, replayDir, replayRunPath, snapshotPath: snapshot.paths.snapshotPath },
+        replay: replayed,
+        errors: [],
+    };
+    (0, fs_atomic_1.writeJson)(replayRunPath, replay);
+    const suite = loadSuiteFromDir(suiteDir);
+    suite.paths.replayRunPath = replayRunPath;
+    suite.cases = suite.cases.map((entry) => (entry.snapshotId === snapshot.id ? { ...entry, replayRunId: replayId, replayPath: replayRunPath } : entry));
+    writeSuite(suite);
+    return replay;
+}
+function loadReplay(target) {
+    const resolved = resolveReplayPath(target);
+    if (!fs.existsSync(resolved))
+        throw new Error(`File not found: ${resolved}`);
+    const replay = (0, fs_atomic_1.readJson)(resolved);
+    if (replay.kind !== "multi-agent-replay-run")
+        throw new Error(`Not a replay run: ${resolved}`);
+    assertReplayShape(replay, resolved);
+    return replay;
+}
+function loadBaselineNormalized(target) {
+    const snapshotPath = resolveSnapshotPath(target);
+    if (!fs.existsSync(snapshotPath))
+        throw new Error(`File not found: ${snapshotPath}`);
+    const snapshot = (0, fs_atomic_1.readJson)(snapshotPath);
+    if (snapshot.kind !== "multi-agent-replay-snapshot")
+        throw new Error(`Not a replay snapshot: ${snapshotPath}`);
+    assertSnapshotShape(snapshot, snapshotPath);
+    return { id: snapshot.id, path: snapshotPath, normalized: snapshot.normalized };
+}
+function compareMultiAgentReplay(baselineTarget, replayTarget) {
+    if (!baselineTarget)
+        throw new Error("Missing baseline id or path.");
+    const baseline = loadBaselineNormalized(baselineTarget);
+    const replay = loadReplay(replayTarget);
+    const suiteDir = replay.paths.suiteDir;
+    const comparisonPath = path.join(suiteDir, "comparison.json");
+    const findingsPath = path.join(suiteDir, "findings.json");
+    const comparison = ev.compareNormalized(baseline.id, baseline.path, baseline.normalized, replay, now(), comparisonPath, findingsPath, suiteDir);
+    (0, fs_atomic_1.writeJson)(comparisonPath, comparison);
+    (0, fs_atomic_1.writeJson)(findingsPath, comparison.findings);
+    const suite = loadSuiteFromDir(suiteDir);
+    suite.paths.comparisonPath = comparisonPath;
+    suite.paths.findingsPath = findingsPath;
+    writeSuite(suite);
+    return comparison;
+}
+function loadOrCompareForTarget(target) {
+    const suiteDir = resolveSuiteDir(target);
+    const comparisonPath = path.join(suiteDir, "comparison.json");
+    const replayPath = resolveReplayPath(target);
+    if (fs.existsSync(comparisonPath)) {
+        const comparison = (0, fs_atomic_1.readJson)(comparisonPath);
+        if (comparison.paths.replayPath === replayPath)
+            return comparison;
+    }
+    return compareMultiAgentReplay(path.join(suiteDir, "snapshot.json"), replayPath);
+}
+function scoreMultiAgentReplay(target) {
+    const comparison = loadOrCompareForTarget(target);
+    const scorePath = path.join(comparison.paths.suiteDir, "score.json");
+    const score = ev.scoreComparison(comparison, now(), scorePath);
+    (0, fs_atomic_1.writeJson)(scorePath, score);
+    const suite = loadSuiteFromDir(comparison.paths.suiteDir);
+    suite.paths.scorePath = scorePath;
+    writeSuite(suite);
+    return score;
+}
+function loadScoreForTarget(target, scorePath) {
+    const replayPath = resolveReplayPath(target);
+    if (fs.existsSync(scorePath)) {
+        const score = (0, fs_atomic_1.readJson)(scorePath);
+        if (fs.existsSync(score.paths.comparisonPath)) {
+            const comparison = (0, fs_atomic_1.readJson)(score.paths.comparisonPath);
+            if (comparison.replayId === score.replayId && comparison.paths.replayPath === replayPath)
+                return score;
+        }
+    }
+    return scoreMultiAgentReplay(target);
+}
+function gateMultiAgentEval(target) {
+    const suiteDir = resolveSuiteDir(target);
+    const snapshotPath = path.join(suiteDir, "snapshot.json");
+    const replayRunPath = path.join(suiteDir, "replay-run.json");
+    const comparisonPath = path.join(suiteDir, "comparison.json");
+    const scorePath = path.join(suiteDir, "score.json");
+    const missing = [snapshotPath, replayRunPath, comparisonPath, scorePath].filter((file) => !fs.existsSync(file));
+    if (missing.length)
+        throw new Error(`Eval gate missing required artifact(s): ${missing.join(", ")}`);
+    const comparison = (0, fs_atomic_1.readJson)(comparisonPath);
+    const score = (0, fs_atomic_1.readJson)(scorePath);
+    const report = reportMultiAgentEval(comparison.paths.replayPath);
+    const gate = ev.buildGate(suiteDir, snapshotPath, replayRunPath, comparisonPath, scorePath, report.reportPath, comparison, score, now(), path.basename(suiteDir));
+    (0, fs_atomic_1.writeJson)(path.join(suiteDir, "gate.json"), gate);
+    return gate;
+}
+function reportMultiAgentEval(target) {
+    const suiteDir = resolveSuiteDir(target);
+    const scorePath = path.join(suiteDir, "score.json");
+    const score = loadScoreForTarget(target, scorePath);
+    const reportPath = path.join(suiteDir, "report.md");
+    const lines = ev.buildReportLines(path.basename(suiteDir), score);
+    fs.writeFileSync(reportPath, `${lines.join("\n")}\n`, "utf8");
+    const suite = loadSuiteFromDir(suiteDir);
+    suite.paths.reportPath = reportPath;
+    writeSuite(suite);
+    return { schemaVersion: 1, replayId: score.replayId, status: score.status, reportPath, score: score.score, maxScore: score.maxScore, findings: score.findings };
+}

package/dist/{multi-agent-eval/format.js → shell/eval-text.js}

@@ -1,20 +1,74 @@
 "use strict";
-var __importDefault = (this && this.__importDefault) || function (mod) {
-    return (mod && mod.__esModule) ? mod : { "default": mod };
-};
+// shell/eval-text.ts — human render for the multi-agent eval/replay layer
+// (`cw eval snapshot|replay|compare|score|gate|report` text output).
+//
+// Byte-behavior port of the old build's src/multi-agent-eval/format.ts's
+// formatMultiAgentEval + its runtime-discriminating type guards. CLI-only —
+// never affects --json / MCP payloads. Types are imported type-only from the
+// v2 core eval-replay module.
+var __createBinding = (this && this.__createBinding) || (Object.create ? (function(o, m, k, k2) {
+    if (k2 === undefined) k2 = k;
+    var desc = Object.getOwnPropertyDescriptor(m, k);
+    if (!desc || ("get" in desc ? !m.__esModule : desc.writable || desc.configurable)) {
+      desc = { enumerable: true, get: function() { return m[k]; } };
+    }
+    Object.defineProperty(o, k2, desc);
+}) : (function(o, m, k, k2) {
+    if (k2 === undefined) k2 = k;
+    o[k2] = m[k];
+}));
+var __setModuleDefault = (this && this.__setModuleDefault) || (Object.create ? (function(o, v) {
+    Object.defineProperty(o, "default", { enumerable: true, value: v });
+}) : function(o, v) {
+    o["default"] = v;
+});
+var __importStar = (this && this.__importStar) || (function () {
+    var ownKeys = function(o) {
+        ownKeys = Object.getOwnPropertyNames || function (o) {
+            var ar = [];
+            for (var k in o) if (Object.prototype.hasOwnProperty.call(o, k)) ar[ar.length] = k;
+            return ar;
+        };
+        return ownKeys(o);
+    };
+    return function (mod) {
+        if (mod && mod.__esModule) return mod;
+        var result = {};
+        if (mod != null) for (var k = ownKeys(mod), i = 0; i < k.length; i++) if (k[i] !== "default") __createBinding(result, mod, k[i]);
+        __setModuleDefault(result, mod);
+        return result;
+    };
+})();
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.formatMultiAgentEval = formatMultiAgentEval;
-// Human formatting for the multi-agent eval replay layer (CLI-only; never affects
-// --json / MCP payloads). Pure functions — a result object in, a string out —
-// carved out of multi-agent-eval.ts (FreeBSD-audit god-module split) so the eval
-// router no longer bundles the rendering layer. The runtime-discriminating type
-// guards travel with the renderer that is their only consumer. Re-exported from
-// multi-agent-eval.ts to keep the public surface byte-unchanged.
-//
-// Types are imported type-only from the parent module: `import type` is fully
-// erased at runtime, so there is no import cycle despite the parent re-exporting
-// formatMultiAgentEval from here.
-const node_path_1 = __importDefault(require("node:path"));
+const path = __importStar(require("node:path"));
+function metricStatus(score, id) {
+    const metric = score.metrics.find((entry) => entry.id === id);
+    return `${id}=${metric?.status || "missing"}`;
+}
+function sectionStatus(comparison, id) {
+    return `${id}=${comparison.sections[id]?.status || "missing"}`;
+}
+function isSnapshot(value) {
+    return Boolean(value && typeof value === "object" && value.kind === "multi-agent-replay-snapshot");
+}
+function isReplay(value) {
+    return Boolean(value && typeof value === "object" && value.kind === "multi-agent-replay-run");
+}
+function isComparison(value) {
+    return Boolean(value && typeof value === "object" && "sections" in value && "findings" in value);
+}
+function isScore(value) {
+    return Boolean(value && typeof value === "object" && "metrics" in value && "score" in value);
+}
+function isGate(value) {
+    return Boolean(value && typeof value === "object" && "verdict" in value && "requiredArtifacts" in value);
+}
+function isReport(value) {
+    return Boolean(value && typeof value === "object" && "reportPath" in value && !("verdict" in value));
+}
+/** Human render for any eval-replay result object. Falls back to pretty JSON
+ *  for an unrecognized shape (matches the old build's default arm). */
 function formatMultiAgentEval(value) {
     if (isGate(value)) {
         return [
@@ -31,13 +85,13 @@ function formatMultiAgentEval(value) {
             `  ${value.verdict}`,
             "",
             "Next Action",
-            `  ${value.nextAction}`
+            `  ${value.nextAction}`,
         ].join("\n");
     }
     if (isScore(value)) {
         return [
             "Eval Suite",
-            `  ${node_path_1.default.basename(value.paths.suiteDir)}`,
+            `  ${path.basename(value.paths.suiteDir)}`,
             "",
             "Replay Status",
             `  ${value.status} (${value.score}/${value.maxScore})`,
@@ -67,13 +121,13 @@ function formatMultiAgentEval(value) {
             `  ${value.status}`,
             "",
             "Next Action",
-            `  ${value.status === "pass" ? "Run eval gate or include report path as evidence." : "Review findings before release."}`
+            `  ${value.status === "pass" ? "Run eval gate or include report path as evidence." : "Review findings before release."}`,
         ].join("\n");
     }
     if (isComparison(value)) {
         return [
             "Eval Suite",
-            `  ${node_path_1.default.basename(value.paths.suiteDir)}`,
+            `  ${path.basename(value.paths.suiteDir)}`,
             "",
             "Replay Status",
             `  ${value.status}`,
@@ -100,20 +154,20 @@ function formatMultiAgentEval(value) {
             `  ${value.status}`,
             "",
             "Next Action",
-            "  Score the replay or run the eval gate."
+            "  Score the replay or run the eval gate.",
         ].join("\n");
     }
     if (isReplay(value)) {
         return [
             "Eval Suite",
-            `  ${node_path_1.default.basename(value.paths.suiteDir)}`,
+            `  ${path.basename(value.paths.suiteDir)}`,
             "",
             "Replay Status",
             `  ${value.status}`,
             `  replay=${value.paths.replayRunPath}`,
             "",
             "Next Action",
-            `  node scripts/cw.js eval compare ${value.paths.snapshotPath} ${value.paths.replayRunPath}`
+            `  node scripts/cw.js eval compare ${value.paths.snapshotPath} ${value.paths.replayRunPath}`,
         ].join("\n");
     }
     if (isSnapshot(value)) {
@@ -147,13 +201,13 @@ function formatMultiAgentEval(value) {
             "  snapshot-ready",
             "",
             "Next Action",
-            `  node scripts/cw.js eval replay ${value.paths.snapshotPath}`
+            `  node scripts/cw.js eval replay ${value.paths.snapshotPath}`,
         ].join("\n");
     }
     if (isReport(value)) {
         return [
             "Eval Suite",
-            `  ${node_path_1.default.dirname(value.reportPath)}`,
+            `  ${path.dirname(value.reportPath)}`,
             "",
             "Replay Status",
             `  ${value.status} (${value.score}/${value.maxScore})`,
@@ -162,33 +216,8 @@ function formatMultiAgentEval(value) {
             `  report written: ${value.reportPath}`,
             "",
             "Next Action",
-            "  Run eval gate if this is release evidence."
+            "  Run eval gate if this is release evidence.",
         ].join("\n");
     }
     return JSON.stringify(value, null, 2);
 }
-function metricStatus(score, id) {
-    const metric = score.metrics.find((entry) => entry.id === id);
-    return `${id}=${metric?.status || "missing"}`;
-}
-function sectionStatus(comparison, id) {
-    return `${id}=${comparison.sections[id]?.status || "missing"}`;
-}
-function isSnapshot(value) {
-    return Boolean(value && typeof value === "object" && value.kind === "multi-agent-replay-snapshot");
-}
-function isReplay(value) {
-    return Boolean(value && typeof value === "object" && value.kind === "multi-agent-replay-run");
-}
-function isComparison(value) {
-    return Boolean(value && typeof value === "object" && "sections" in value && "findings" in value);
-}
-function isScore(value) {
-    return Boolean(value && typeof value === "object" && "metrics" in value && "score" in value);
-}
-function isGate(value) {
-    return Boolean(value && typeof value === "object" && "verdict" in value && "requiredArtifacts" in value);
-}
-function isReport(value) {
-    return Boolean(value && typeof value === "object" && "reportPath" in value && !("verdict" in value));
-}