Add generic control kernel for evidence-backed agent workflows

docs/SERVICE_READINESS.md

@@ -6,8 +6,8 @@ Status: ready
 
 - service-loop: pass - docs/SERVICE_LOOP.md
 - onboarding: pass - docs/ONBOARDING.md
-- official-docs-coverage: pass - fixtures/replay/gajae-code/official-docs.json, fixtures/replay/awesome-codex-subagents/official-docs.json, fixtures/replay/kimi-agent-swarm-skill/official-docs.json
-- external-replay: pass - fixtures/replay/gajae-code/replay.json, fixtures/replay/awesome-codex-subagents/replay.json, fixtures/replay/kimi-agent-swarm-skill/replay.json
+- official-docs-coverage: pass - fixtures/replay/awesome-codex-subagents/official-docs.json, fixtures/replay/gajae-code/official-docs.json, fixtures/replay/kimi-agent-swarm-skill/official-docs.json
+- external-replay: pass - fixtures/replay/awesome-codex-subagents/replay.json, fixtures/replay/gajae-code/replay.json, fixtures/replay/kimi-agent-swarm-skill/replay.json
 - handoff-validation: pass - fixtures/handoffs/low.json, fixtures/handoffs/medium.json, fixtures/handoffs/high.json
 - service-acceptance-gates: pass - fixtures/service-readiness/gates.json
 - field-evidence: pass - evidence/field-readiness/oss-run-1

fixtures/control-kernel/gp-screening-shadow-v0.json

@@ -0,0 +1,218 @@
+{
+  "schemaVersion": "boulder.control.fixture.v1",
+  "policy": {
+    "schemaVersion": "boulder.control.policy.v1",
+    "id": "gp-screening-shadow",
+    "version": "0.1.0",
+    "blockingSeverities": [
+      "critical",
+      "major"
+    ],
+    "hardFailures": [
+      {
+        "id": "mandate-exclusion",
+        "signal": "mandate.exclusion.missed",
+        "severity": "critical",
+        "description": "A hard mandate exclusion was missed or suppressed."
+      },
+      {
+        "id": "conflict-escalation",
+        "signal": "conflict.escalation.missed",
+        "severity": "critical",
+        "description": "A required conflict or related-party escalation was missed."
+      },
+      {
+        "id": "unsupported-material-claim",
+        "signal": "evidence.material-claim.unsupported",
+        "severity": "major",
+        "description": "A material recommendation claim lacks admissible evidence."
+      }
+    ],
+    "metricRules": [
+      {
+        "metricId": "decision-accuracy",
+        "operator": "gte",
+        "threshold": 0.75,
+        "description": "Decision agreement or adjudicated quality must meet the pilot floor."
+      },
+      {
+        "metricId": "evidence-coverage",
+        "operator": "gte",
+        "threshold": 0.9,
+        "description": "Material claims must be covered by admissible evidence."
+      },
+      {
+        "metricId": "risk-recall",
+        "operator": "gte",
+        "threshold": 0.8,
+        "description": "Known material risks must be surfaced."
+      }
+    ]
+  },
+  "evidenceManifest": {
+    "schemaVersion": "boulder.control.evidence-manifest.v1",
+    "manifestId": "synthetic-screening-001-manifest",
+    "caseId": "synthetic-screening-001",
+    "evidenceCutoffAt": "2025-01-15T00:00:00.000Z",
+    "generatedAt": "2026-07-18T14:00:00.000Z",
+    "entries": [
+      {
+        "evidenceId": "cim-v1",
+        "sha256": "1111111111111111111111111111111111111111111111111111111111111111",
+        "sourceType": "document",
+        "classification": "confidential",
+        "sourceVersion": "v1",
+        "observedAt": "2025-01-10T00:00:00.000Z",
+        "sourceUriHash": "2222222222222222222222222222222222222222222222222222222222222222"
+      },
+      {
+        "evidenceId": "mandate-v3",
+        "sha256": "3333333333333333333333333333333333333333333333333333333333333333",
+        "sourceType": "database",
+        "classification": "restricted",
+        "sourceVersion": "v3",
+        "observedAt": "2025-01-01T00:00:00.000Z",
+        "sourceUriHash": "4444444444444444444444444444444444444444444444444444444444444444"
+      }
+    ]
+  },
+  "runs": {
+    "pass": {
+      "schemaVersion": "boulder.control.run-event.v1",
+      "caseId": "synthetic-screening-001",
+      "taskId": "screening-decision-package",
+      "agentId": "underwriting-agent",
+      "agentVersion": "0.1.0",
+      "profileId": "gp-screening-shadow",
+      "profileVersion": "0.1.0",
+      "parentRunId": null,
+      "startedAt": "2026-07-18T14:01:00.000Z",
+      "completedAt": "2026-07-18T14:02:00.000Z",
+      "evidenceCutoffAt": "2025-01-15T00:00:00.000Z",
+      "evidenceManifestHash": "39c90fcde10b441df6cdf1c81999a0eda124a72a8946f812b817ca58f22b6ac4",
+      "policyId": "gp-screening-shadow",
+      "policyVersion": "0.1.0",
+      "policyHash": "9d82eaf6062bc15d597a374f791b64405be8ccb06379dcc9c4722dbb80b451ea",
+      "promptVersion": "screening-v0.1",
+      "model": {
+        "provider": "test",
+        "name": "synthetic-screening-model",
+        "version": "1.0.0"
+      },
+      "toolCalls": [
+        {
+          "toolId": "evidence-index",
+          "toolVersion": "0.1.0",
+          "action": "read",
+          "status": "completed",
+          "inputHash": "5555555555555555555555555555555555555555555555555555555555555555",
+          "outputHash": "6666666666666666666666666666666666666666666666666666666666666666"
+        }
+      ],
+      "artifactHashes": [
+        "7777777777777777777777777777777777777777777777777777777777777777"
+      ],
+      "status": "completed",
+      "runId": "synthetic-pass-run",
+      "idempotencyKey": "synthetic-pass-run-v1",
+      "metrics": {
+        "decision-accuracy": 0.83,
+        "evidence-coverage": 0.96,
+        "risk-recall": 0.88
+      },
+      "hardFailureSignals": []
+    },
+    "hardFailure": {
+      "schemaVersion": "boulder.control.run-event.v1",
+      "caseId": "synthetic-screening-001",
+      "taskId": "screening-decision-package",
+      "agentId": "underwriting-agent",
+      "agentVersion": "0.1.0",
+      "profileId": "gp-screening-shadow",
+      "profileVersion": "0.1.0",
+      "parentRunId": null,
+      "startedAt": "2026-07-18T14:01:00.000Z",
+      "completedAt": "2026-07-18T14:02:00.000Z",
+      "evidenceCutoffAt": "2025-01-15T00:00:00.000Z",
+      "evidenceManifestHash": "39c90fcde10b441df6cdf1c81999a0eda124a72a8946f812b817ca58f22b6ac4",
+      "policyId": "gp-screening-shadow",
+      "policyVersion": "0.1.0",
+      "policyHash": "9d82eaf6062bc15d597a374f791b64405be8ccb06379dcc9c4722dbb80b451ea",
+      "promptVersion": "screening-v0.1",
+      "model": {
+        "provider": "test",
+        "name": "synthetic-screening-model",
+        "version": "1.0.0"
+      },
+      "toolCalls": [
+        {
+          "toolId": "evidence-index",
+          "toolVersion": "0.1.0",
+          "action": "read",
+          "status": "completed",
+          "inputHash": "5555555555555555555555555555555555555555555555555555555555555555",
+          "outputHash": "6666666666666666666666666666666666666666666666666666666666666666"
+        }
+      ],
+      "artifactHashes": [
+        "7777777777777777777777777777777777777777777777777777777777777777"
+      ],
+      "status": "completed",
+      "runId": "synthetic-hard-failure-run",
+      "idempotencyKey": "synthetic-hard-failure-run-v1",
+      "metrics": {
+        "decision-accuracy": 1.0,
+        "evidence-coverage": 1.0,
+        "risk-recall": 1.0
+      },
+      "hardFailureSignals": [
+        "mandate.exclusion.missed"
+      ]
+    },
+    "metricFailure": {
+      "schemaVersion": "boulder.control.run-event.v1",
+      "caseId": "synthetic-screening-001",
+      "taskId": "screening-decision-package",
+      "agentId": "underwriting-agent",
+      "agentVersion": "0.1.0",
+      "profileId": "gp-screening-shadow",
+      "profileVersion": "0.1.0",
+      "parentRunId": null,
+      "startedAt": "2026-07-18T14:01:00.000Z",
+      "completedAt": "2026-07-18T14:02:00.000Z",
+      "evidenceCutoffAt": "2025-01-15T00:00:00.000Z",
+      "evidenceManifestHash": "39c90fcde10b441df6cdf1c81999a0eda124a72a8946f812b817ca58f22b6ac4",
+      "policyId": "gp-screening-shadow",
+      "policyVersion": "0.1.0",
+      "policyHash": "9d82eaf6062bc15d597a374f791b64405be8ccb06379dcc9c4722dbb80b451ea",
+      "promptVersion": "screening-v0.1",
+      "model": {
+        "provider": "test",
+        "name": "synthetic-screening-model",
+        "version": "1.0.0"
+      },
+      "toolCalls": [
+        {
+          "toolId": "evidence-index",
+          "toolVersion": "0.1.0",
+          "action": "read",
+          "status": "completed",
+          "inputHash": "5555555555555555555555555555555555555555555555555555555555555555",
+          "outputHash": "6666666666666666666666666666666666666666666666666666666666666666"
+        }
+      ],
+      "artifactHashes": [
+        "7777777777777777777777777777777777777777777777777777777777777777"
+      ],
+      "status": "completed",
+      "runId": "synthetic-metric-failure-run",
+      "idempotencyKey": "synthetic-metric-failure-run-v1",
+      "metrics": {
+        "decision-accuracy": 0.83,
+        "evidence-coverage": 0.96,
+        "risk-recall": 0.6
+      },
+      "hardFailureSignals": []
+    }
+  }
+}

fixtures/package-inventory/packaged-files.v0.json

@@ -1,11 +1,11 @@
 {
   "schemaVersion": "packaged-files.v0",
-  "totalUniqueFiles": 186,
-  "totalPackedFiles": 187,
+  "totalUniqueFiles": 191,
+  "totalPackedFiles": 192,
   "classes": [
     {
       "class": "runtime",
-      "count": 68,
+      "count": 72,
       "files": [
         "bin/boulder.js",
         "bin/boulder.ts",
@@ -22,6 +22,10 @@
         "src/cli-options.ts",
         "src/cli-run-recording.ts",
         "src/cli.ts",
+        "src/control-kernel-command.ts",
+        "src/control-kernel-types.ts",
+        "src/control-kernel-validation.ts",
+        "src/control-kernel.ts",
         "src/executor-adapters.ts",
         "src/executors.ts",
         "src/export.ts",
@@ -176,12 +180,13 @@
     },
     {
       "class": "fixture",
-      "count": 24,
+      "count": 25,
       "files": [
         "fixtures/benchmarks/mcp-server.json",
         "fixtures/benchmarks/python-package.json",
         "fixtures/benchmarks/typescript-library.json",
         "fixtures/capabilities/codex-installed.json",
+        "fixtures/control-kernel/gp-screening-shadow-v0.json",
         "fixtures/docs/doc-registry.v0.json",
         "fixtures/handoffs/high.json",
         "fixtures/handoffs/low.json",

src/cli-format.ts

@@ -55,6 +55,10 @@ export function printHelp(): void {
     "  boulder replay-check [--cwd path] [--json]",
     "  boulder replay-run [--cwd path] --dry-run [--json]",
     "  boulder doctor [--cwd path] [--json]",
+    "  boulder control record --event path [--cwd path] [--json]",
+    "  boulder control evaluate --event path --manifest path --policy path [--cwd path] [--json]",
+    "  boulder control seal --event path --manifest path --policy path [--cwd path] [--json]",
+    "  boulder control verify-seal --seal path --event path --manifest path --policy path [--cwd path] [--json]",
     "  boulder record field-readiness --run-id id --evidence path [--cwd path] [--json]",
     "  boulder export [--cwd path] [--force]",
     "",

src/cli.ts

@@ -4,6 +4,7 @@ import { runCapabilityCommand } from "./capability-command";
 import { formatLines, prettyJson, printHelp } from "./cli-format";
 import { runOperationalCommand } from "./cli-ops-command";
 import { optionValue, parseOptions, valueAfter } from "./cli-options";
+import { runControlKernelCommand } from "./control-kernel-command";
 import { UnsafeGeneratedWritePathError, writeGeneratedText } from "./fs";
 import { exportHarness } from "./export";
 import { runHandoffCommand } from "./handoff-command";
@@ -92,6 +93,10 @@ async function runMain(args: string[]): Promise<void> {
     await runRunsCommand(parsed.commandArgs, args, options.cwd, options.json);
     return;
   }
+  if (command === "control") {
+    await runControlKernelCommand(args, { cwd: options.cwd, json: options.json });
+    return;
+  }
   if (command === "bootstrap" && args.includes("interview")) {
     const report = buildBootstrapInterview(optionValue(args, "--task"));
     if (options.json) {