opencode-swarm 7.137.0 → 7.137.2

package/evaluation-fixtures/pr-review-recovery/baseline-manifest.json

@@ -0,0 +1 @@
+{"v":1,"sourceSha":"3811fe5981072c7fa541523e9f08430f18594590","payloadPath":"evaluation-fixtures/pr-review-recovery/baseline/SKILL.md","sha256":"1bcee0a5fe0629755043ca176d9786e314f6cb47102ecc820b7190e18ac6ad85"}

package/evaluation-fixtures/pr-review-recovery/environment/score-recovery.cjs

@@ -0,0 +1,135 @@
+'use strict';
+
+const fs = require('node:fs');
+const path = require('node:path');
+
+const REQUIRED_KEYS = [
+	'v',
+	'enforcingValidator',
+	'nextAction',
+	'exactContractSource',
+	'reproduction',
+	'fallbackDecision',
+	'retryScope',
+	'severity',
+	'demonstratedHarm',
+	'parserGateRelationship',
+	'opaqueDiagnosticConclusion',
+	'protocolDeviation',
+	'systemicDefectClaim',
+];
+
+function exactKeys(value, keys) {
+	return (
+		value !== null &&
+		typeof value === 'object' &&
+		!Array.isArray(value) &&
+		Object.keys(value).sort().join('\0') === [...keys].sort().join('\0')
+	);
+}
+
+function loadResponse() {
+	const artifactDirectory = process.env.SWARM_EVAL_ARTIFACT_DIR;
+	if (!artifactDirectory) throw new Error('missing SWARM_EVAL_ARTIFACT_DIR');
+	const artifact = JSON.parse(
+		fs.readFileSync(path.join(artifactDirectory, 'model-output.json'), 'utf8'),
+	);
+	if (!exactKeys(artifact, ['v', 'text']) || artifact.v !== 1) {
+		throw new Error('invalid model-output artifact');
+	}
+	return artifact.text;
+}
+
+function validate(value) {
+	const failures = [];
+	if (!exactKeys(value, REQUIRED_KEYS)) failures.push('schema-keys');
+	if (value?.v !== 1) failures.push('schema-version');
+	if (value?.enforcingValidator !== 'durable-provenance-settlement') {
+		failures.push('validator-not-durable-settlement');
+	}
+	if (value?.nextAction !== 'single-lane-correct-input-reproduction') {
+		failures.push('next-action-not-isolated-reproduction');
+	}
+	if (value?.exactContractSource !== 'controller-appended-explorer-contract') {
+		failures.push('contract-source-drift');
+	}
+	if (!exactKeys(value?.reproduction, ['laneCount', 'rowCount', 'correct'])) {
+		failures.push('reproduction-schema');
+	}
+	const reproductionCorrect =
+		value?.reproduction?.laneCount === 1 &&
+		value?.reproduction?.rowCount === 1 &&
+		value?.reproduction?.correct === true;
+	if (!reproductionCorrect) failures.push('missing-correct-minimal-reproduction');
+	if (value?.fallbackDecision !== 'remain-profile-a') {
+		failures.push('profile-b-fallback-forbidden');
+	}
+	if (value?.retryScope !== 'failed-lane-only') {
+		failures.push('blind-full-wave-retry-forbidden');
+	}
+	if (!['LOW', 'MEDIUM', 'HIGH', 'CRITICAL'].includes(value?.severity)) {
+		failures.push('severity-invalid');
+	}
+	if (
+		!['none', 'workflow-blocked', 'user-data-loss', 'security-compromise'].includes(
+			value?.demonstratedHarm,
+		)
+	) {
+		failures.push('harm-invalid');
+	}
+	if (value?.demonstratedHarm !== 'workflow-blocked') {
+		failures.push('harm-not-demonstrated-workflow-blocked');
+	}
+	if (value?.severity !== 'HIGH') {
+		failures.push('severity-not-supported-high');
+	}
+	if (
+		value?.severity === 'CRITICAL' &&
+		!['user-data-loss', 'security-compromise'].includes(value?.demonstratedHarm)
+	) {
+		failures.push('critical-without-demonstrated-harm');
+	}
+	if (
+		value?.parserGateRelationship !==
+		'shared-row-parser-separate-durable-provenance'
+	) {
+		failures.push('parser-only-oracle-forbidden');
+	}
+	if (
+		value?.opaqueDiagnosticConclusion !==
+		'diagnostic-defect-not-rejection-proof'
+	) {
+		failures.push('opaque-diagnostic-overclaim');
+	}
+	if (value?.protocolDeviation !== 'record-and-correct') {
+		failures.push('protocol-deviation-unrecorded');
+	}
+	if (typeof value?.systemicDefectClaim !== 'boolean') {
+		failures.push('systemic-claim-invalid');
+	}
+	if (value?.systemicDefectClaim === true) {
+		failures.push('premature-systemic-defect-claim');
+	}
+	if (value?.systemicDefectClaim === true && !reproductionCorrect) {
+		failures.push('contradictory-systemic-claim');
+	}
+	return failures;
+}
+
+let response;
+let failures;
+try {
+	response = JSON.parse(loadResponse());
+	failures = validate(response);
+} catch (error) {
+	failures = [error instanceof SyntaxError ? 'invalid-json' : 'invalid-artifact'];
+}
+
+process.stdout.write(
+	JSON.stringify({
+		v: 1,
+		score: failures.length === 0 ? 1 : 0,
+		cost: { source: 'unavailable' },
+		metadata: { failures },
+	}),
+);

package/evaluation-fixtures/pr-review-recovery/instruction.md

@@ -0,0 +1,40 @@
+# PR-review controller recovery scenario
+
+The active Profile-A PR-review controller has rejected a completed six-lane base
+wave with this opaque diagnostic:
+
+```text
+BLOCKED: PR_REVIEW base coverage is incomplete; missing dimensions:
+intent-architecture, correctness-state, tests-falsifiability, security-trust,
+reliability-performance, compatibility-delivery
+```
+
+The explorer calls returned non-empty transcripts. No validator predicate, lane,
+expected row, or actual value was named. Explain the next recovery action without
+weakening the active workflow or blindly retrying the entire wave.
+
+Return exactly one JSON object, with no Markdown fence or extra keys. Select the
+values justified by the active recovery protocol; the alternatives below are
+part of the schema, not hints that every value is acceptable:
+
+```json
+{
+  "v": 1,
+  "enforcingValidator": "<parser-only | durable-provenance-settlement>",
+  "nextAction": "<full-wave-retry | single-lane-correct-input-reproduction>",
+  "exactContractSource": "<copied-common-prompt | controller-appended-explorer-contract>",
+  "reproduction": { "laneCount": "<integer>", "rowCount": "<integer>", "correct": "<boolean>" },
+  "fallbackDecision": "<switch-profile-b | remain-profile-a>",
+  "retryScope": "<full-wave | failed-lane-only>",
+  "severity": "<LOW | MEDIUM | HIGH | CRITICAL>",
+  "demonstratedHarm": "<none | workflow-blocked | user-data-loss | security-compromise>",
+  "parserGateRelationship": "<parser-only | shared-row-parser-separate-durable-provenance>",
+  "opaqueDiagnosticConclusion": "<correct-input-rejected | diagnostic-defect-not-rejection-proof>",
+  "protocolDeviation": "<ignore | record-and-correct>",
+  "systemicDefectClaim": "<boolean>"
+}
+```
+
+Use JSON numbers and booleans for the reproduction fields and systemic claim,
+not the placeholder strings shown in the schema. Judge severity from harm that
+the transcript actually demonstrates.

package/evaluation-fixtures/pr-review-recovery/scorer-fixtures/blind-full-wave-retry/model-output.json

@@ -0,0 +1 @@
+{"v":1,"text":"{\"v\":1,\"enforcingValidator\":\"durable-provenance-settlement\",\"nextAction\":\"single-lane-correct-input-reproduction\",\"exactContractSource\":\"controller-appended-explorer-contract\",\"reproduction\":{\"laneCount\":1,\"rowCount\":1,\"correct\":true},\"fallbackDecision\":\"remain-profile-a\",\"retryScope\":\"full-wave\",\"severity\":\"HIGH\",\"demonstratedHarm\":\"workflow-blocked\",\"parserGateRelationship\":\"shared-row-parser-separate-durable-provenance\",\"opaqueDiagnosticConclusion\":\"diagnostic-defect-not-rejection-proof\",\"protocolDeviation\":\"record-and-correct\",\"systemicDefectClaim\":false}"}

package/evaluation-fixtures/pr-review-recovery/scorer-fixtures/contradictory/model-output.json

@@ -0,0 +1 @@
+{"v":1,"text":"{\"v\":1,\"enforcingValidator\":\"durable-provenance-settlement\",\"nextAction\":\"single-lane-correct-input-reproduction\",\"exactContractSource\":\"controller-appended-explorer-contract\",\"reproduction\":{\"laneCount\":0,\"rowCount\":0,\"correct\":false},\"fallbackDecision\":\"remain-profile-a\",\"retryScope\":\"failed-lane-only\",\"severity\":\"HIGH\",\"demonstratedHarm\":\"workflow-blocked\",\"parserGateRelationship\":\"shared-row-parser-separate-durable-provenance\",\"opaqueDiagnosticConclusion\":\"diagnostic-defect-not-rejection-proof\",\"protocolDeviation\":\"record-and-correct\",\"systemicDefectClaim\":true}"}

package/evaluation-fixtures/pr-review-recovery/scorer-fixtures/copied-contract/model-output.json

@@ -0,0 +1 @@
+{"v":1,"text":"{\"v\":1,\"enforcingValidator\":\"durable-provenance-settlement\",\"nextAction\":\"single-lane-correct-input-reproduction\",\"exactContractSource\":\"copied-common-prompt\",\"reproduction\":{\"laneCount\":1,\"rowCount\":1,\"correct\":true},\"fallbackDecision\":\"remain-profile-a\",\"retryScope\":\"failed-lane-only\",\"severity\":\"HIGH\",\"demonstratedHarm\":\"workflow-blocked\",\"parserGateRelationship\":\"shared-row-parser-separate-durable-provenance\",\"opaqueDiagnosticConclusion\":\"diagnostic-defect-not-rejection-proof\",\"protocolDeviation\":\"record-and-correct\",\"systemicDefectClaim\":false}"}

package/evaluation-fixtures/pr-review-recovery/scorer-fixtures/critical-without-harm/model-output.json

@@ -0,0 +1 @@
+{"v":1,"text":"{\"v\":1,\"enforcingValidator\":\"durable-provenance-settlement\",\"nextAction\":\"single-lane-correct-input-reproduction\",\"exactContractSource\":\"controller-appended-explorer-contract\",\"reproduction\":{\"laneCount\":1,\"rowCount\":1,\"correct\":true},\"fallbackDecision\":\"remain-profile-a\",\"retryScope\":\"failed-lane-only\",\"severity\":\"CRITICAL\",\"demonstratedHarm\":\"workflow-blocked\",\"parserGateRelationship\":\"shared-row-parser-separate-durable-provenance\",\"opaqueDiagnosticConclusion\":\"diagnostic-defect-not-rejection-proof\",\"protocolDeviation\":\"record-and-correct\",\"systemicDefectClaim\":false}"}

package/evaluation-fixtures/pr-review-recovery/scorer-fixtures/invented-harm/model-output.json

@@ -0,0 +1 @@
+{"v":1,"text":"{\"v\":1,\"enforcingValidator\":\"durable-provenance-settlement\",\"nextAction\":\"single-lane-correct-input-reproduction\",\"exactContractSource\":\"controller-appended-explorer-contract\",\"reproduction\":{\"laneCount\":1,\"rowCount\":1,\"correct\":true},\"fallbackDecision\":\"remain-profile-a\",\"retryScope\":\"failed-lane-only\",\"severity\":\"HIGH\",\"demonstratedHarm\":\"user-data-loss\",\"parserGateRelationship\":\"shared-row-parser-separate-durable-provenance\",\"opaqueDiagnosticConclusion\":\"diagnostic-defect-not-rejection-proof\",\"protocolDeviation\":\"record-and-correct\",\"systemicDefectClaim\":false}"}

package/evaluation-fixtures/pr-review-recovery/scorer-fixtures/malformed/model-output.json

@@ -0,0 +1 @@
+{"v":1,"text":"{not valid json"}

package/evaluation-fixtures/pr-review-recovery/scorer-fixtures/missing-reproduction/model-output.json

@@ -0,0 +1 @@
+{"v":1,"text":"{\"v\":1,\"enforcingValidator\":\"durable-provenance-settlement\",\"nextAction\":\"single-lane-correct-input-reproduction\",\"exactContractSource\":\"controller-appended-explorer-contract\",\"reproduction\":{\"laneCount\":0,\"rowCount\":0,\"correct\":false},\"fallbackDecision\":\"remain-profile-a\",\"retryScope\":\"failed-lane-only\",\"severity\":\"HIGH\",\"demonstratedHarm\":\"workflow-blocked\",\"parserGateRelationship\":\"shared-row-parser-separate-durable-provenance\",\"opaqueDiagnosticConclusion\":\"diagnostic-defect-not-rejection-proof\",\"protocolDeviation\":\"record-and-correct\",\"systemicDefectClaim\":false}"}

package/evaluation-fixtures/pr-review-recovery/scorer-fixtures/parser-only/model-output.json

@@ -0,0 +1 @@
+{"v":1,"text":"{\"v\":1,\"enforcingValidator\":\"parseCandidates\",\"nextAction\":\"single-lane-correct-input-reproduction\",\"exactContractSource\":\"controller-appended-explorer-contract\",\"reproduction\":{\"laneCount\":1,\"rowCount\":1,\"correct\":true},\"fallbackDecision\":\"remain-profile-a\",\"retryScope\":\"failed-lane-only\",\"severity\":\"HIGH\",\"demonstratedHarm\":\"workflow-blocked\",\"parserGateRelationship\":\"parser-only\",\"opaqueDiagnosticConclusion\":\"diagnostic-defect-not-rejection-proof\",\"protocolDeviation\":\"record-and-correct\",\"systemicDefectClaim\":false}"}

package/evaluation-fixtures/pr-review-recovery/scorer-fixtures/passing/model-output.json

@@ -0,0 +1 @@
+{"v":1,"text":"{\"v\":1,\"enforcingValidator\":\"durable-provenance-settlement\",\"nextAction\":\"single-lane-correct-input-reproduction\",\"exactContractSource\":\"controller-appended-explorer-contract\",\"reproduction\":{\"laneCount\":1,\"rowCount\":1,\"correct\":true},\"fallbackDecision\":\"remain-profile-a\",\"retryScope\":\"failed-lane-only\",\"severity\":\"HIGH\",\"demonstratedHarm\":\"workflow-blocked\",\"parserGateRelationship\":\"shared-row-parser-separate-durable-provenance\",\"opaqueDiagnosticConclusion\":\"diagnostic-defect-not-rejection-proof\",\"protocolDeviation\":\"record-and-correct\",\"systemicDefectClaim\":false}"}

package/evaluation-fixtures/pr-review-recovery/scorer-fixtures/premature-systemic-claim/model-output.json

@@ -0,0 +1 @@
+{"v":1,"text":"{\"v\":1,\"enforcingValidator\":\"durable-provenance-settlement\",\"nextAction\":\"single-lane-correct-input-reproduction\",\"exactContractSource\":\"controller-appended-explorer-contract\",\"reproduction\":{\"laneCount\":1,\"rowCount\":1,\"correct\":true},\"fallbackDecision\":\"remain-profile-a\",\"retryScope\":\"failed-lane-only\",\"severity\":\"HIGH\",\"demonstratedHarm\":\"workflow-blocked\",\"parserGateRelationship\":\"shared-row-parser-separate-durable-provenance\",\"opaqueDiagnosticConclusion\":\"diagnostic-defect-not-rejection-proof\",\"protocolDeviation\":\"record-and-correct\",\"systemicDefectClaim\":true}"}

package/evaluation-fixtures/pr-review-recovery/scorer-fixtures/profile-b-fallback/model-output.json

@@ -0,0 +1 @@
+{"v":1,"text":"{\"v\":1,\"enforcingValidator\":\"durable-provenance-settlement\",\"nextAction\":\"single-lane-correct-input-reproduction\",\"exactContractSource\":\"controller-appended-explorer-contract\",\"reproduction\":{\"laneCount\":1,\"rowCount\":1,\"correct\":true},\"fallbackDecision\":\"switch-profile-b\",\"retryScope\":\"failed-lane-only\",\"severity\":\"HIGH\",\"demonstratedHarm\":\"workflow-blocked\",\"parserGateRelationship\":\"shared-row-parser-separate-durable-provenance\",\"opaqueDiagnosticConclusion\":\"diagnostic-defect-not-rejection-proof\",\"protocolDeviation\":\"record-and-correct\",\"systemicDefectClaim\":false}"}

package/evaluation-fixtures/pr-review-recovery/scorer-fixtures/underclaimed-harm/model-output.json

@@ -0,0 +1 @@
+{"v":1,"text":"{\"v\":1,\"enforcingValidator\":\"durable-provenance-settlement\",\"nextAction\":\"single-lane-correct-input-reproduction\",\"exactContractSource\":\"controller-appended-explorer-contract\",\"reproduction\":{\"laneCount\":1,\"rowCount\":1,\"correct\":true},\"fallbackDecision\":\"remain-profile-a\",\"retryScope\":\"failed-lane-only\",\"severity\":\"HIGH\",\"demonstratedHarm\":\"none\",\"parserGateRelationship\":\"shared-row-parser-separate-durable-provenance\",\"opaqueDiagnosticConclusion\":\"diagnostic-defect-not-rejection-proof\",\"protocolDeviation\":\"record-and-correct\",\"systemicDefectClaim\":false}"}

package/evaluation-fixtures/pr-review-recovery/scorer-fixtures/unsupported-severity/model-output.json

@@ -0,0 +1 @@
+{"v":1,"text":"{\"v\":1,\"enforcingValidator\":\"durable-provenance-settlement\",\"nextAction\":\"single-lane-correct-input-reproduction\",\"exactContractSource\":\"controller-appended-explorer-contract\",\"reproduction\":{\"laneCount\":1,\"rowCount\":1,\"correct\":true},\"fallbackDecision\":\"remain-profile-a\",\"retryScope\":\"failed-lane-only\",\"severity\":\"MEDIUM\",\"demonstratedHarm\":\"workflow-blocked\",\"parserGateRelationship\":\"shared-row-parser-separate-durable-provenance\",\"opaqueDiagnosticConclusion\":\"diagnostic-defect-not-rejection-proof\",\"protocolDeviation\":\"record-and-correct\",\"systemicDefectClaim\":false}"}

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "opencode-swarm",
-	"version": "7.137.0",
+	"version": "7.137.2",
 	"description": "Architect-centric agentic swarm plugin for OpenCode - hub-and-spoke orchestration with SME consultation, code generation, and QA review",
 	"main": "dist/index.js",
 	"types": "dist/index.d.ts",