deadbyte-mcp 0.13.0 → 0.15.0

package/test/agent-eval-runner-r42.test.mjs

@@ -0,0 +1,61 @@
+import test from 'node:test';
+import assert from 'node:assert/strict';
+import { mkdtemp, readFile, rm, writeFile } from 'node:fs/promises';
+import os from 'node:os';
+import path from 'node:path';
+import { runAgentEvalSuiteR42, readAndVerifyR42EvalEvidence } from '../src/agent-eval-runner-r42.mjs';
+
+const H=value=>String(value).repeat(64).slice(0,64);
+const suite={schema:'deadbyte.agent-eval-suite.v1',suite_id:'suite',
+  cases:[{schema:'deadbyte.agent-eval-case.v1',case_id:'case-1',task:'Safe action',
+    allowed_context:{visible:'YES'},hidden_oracle:{secret:'ORACLE_ONLY',expected:'bounded'},acceptance:['bounded']}]
+};
+
+test('R42 model eval keeps hidden oracle out of generator and binds passed evidence to manifest',async()=>{
+  const root=await mkdtemp(path.join(os.tmpdir(),'deadbyte-r42-eval-'));
+  const calls=[];
+  try{
+    const run=await runAgentEvalSuiteR42({suite,subjectManifestSha256:H('a'),evidenceRoot:root,
+      clock:()=>new Date('2026-09-20T02:00:00.000Z'),
+      requestModel:async call=>{
+        calls.push(call);
+        if(call.role==='generator')return {backend:'fake',model:'fake',response_id:'g1',raw_text:JSON.stringify({artifact:{plan:['inspect','verify']}})};
+        return {backend:'fake',model:'fake',response_id:'j1',raw_text:JSON.stringify({schema:'deadbyte.agent-eval-judge-result.v1',
+          verdict:'pass',advisory:true,findings:[],evidence_refs:[H('1')]})};
+      }});
+    assert.equal(calls.length,2);
+    assert.equal(calls[0].prompt.includes('ORACLE_ONLY'),false);
+    assert.equal(calls[1].prompt.includes('ORACLE_ONLY'),true);
+    assert.equal(run.payload.status,'passed');
+    assert.equal(run.payload.subject_manifest_sha256,H('a'));
+    assert.equal(run.publication.status,'created');
+    const verified=await readAndVerifyR42EvalEvidence(run.publication.path,{subjectManifestSha256:H('a'),suiteSha256:run.payload.suite_sha256});
+    assert.equal(verified.status,'passed');
+  }finally{await rm(root,{recursive:true,force:true});}
+});
+
+test('R42 model eval rejects failed judge and publishes no PASS evidence',async()=>{
+  const root=await mkdtemp(path.join(os.tmpdir(),'deadbyte-r42-eval-fail-'));
+  try{
+    await assert.rejects(runAgentEvalSuiteR42({suite,subjectManifestSha256:H('b'),evidenceRoot:root,
+      requestModel:async call=>call.role==='generator'
+        ?{raw_text:JSON.stringify({artifact:{plan:['unsafe']}})}
+        :{raw_text:JSON.stringify({schema:'deadbyte.agent-eval-judge-result.v1',verdict:'fail',advisory:true,
+          findings:[{code:'AUTH',severity:'critical',evidence:'authority widening'}],evidence_refs:[H('2')]})}}),/suite failed/i);
+    await assert.rejects(readFile(path.join(root,'agent_eval',`r42-${H('b').slice(0,16)}-`),'utf8'));
+  }finally{await rm(root,{recursive:true,force:true});}
+});
+
+test('R42 eval evidence tampering fails independent verification',async()=>{
+  const root=await mkdtemp(path.join(os.tmpdir(),'deadbyte-r42-eval-tamper-'));
+  try{
+    const run=await runAgentEvalSuiteR42({suite,subjectManifestSha256:H('c'),evidenceRoot:root,
+      requestModel:async call=>call.role==='generator'
+        ?{raw_text:JSON.stringify({artifact:{plan:['safe']}})}
+        :{raw_text:JSON.stringify({schema:'deadbyte.agent-eval-judge-result.v1',verdict:'pass',advisory:true,findings:[],evidence_refs:[H('3')]})}});
+    const raw=JSON.parse(await readFile(run.publication.path,'utf8'));
+    raw.payload.cases[0].result.verdict='fail';
+    await writeFile(run.publication.path,JSON.stringify(raw));
+    await assert.rejects(readAndVerifyR42EvalEvidence(run.publication.path,{subjectManifestSha256:H('c')}),/hash mismatch/i);
+  }finally{await rm(root,{recursive:true,force:true});}
+});

package/test/agent-governance-r42.test.mjs

@@ -0,0 +1,77 @@
+import test from 'node:test';
+import assert from 'node:assert/strict';
+import {
+  bindApprovalSubject, validateBoundApproval, createTaskDag, invalidateTaskDag,
+  createEvidenceClaim, evidenceSatisfies, deriveEffectiveRigor, taskDagFromPlan,
+  createDecisionRecord, evaluateDecisionFreshness
+} from '../src/agent-governance-r42.mjs';
+
+const H=value=>String(value).repeat(64).slice(0,64);
+
+test('R42 approval binds exact artifact, event head, scope and expiry',()=>{
+  const subject=bindApprovalSubject({goal_id:H('1'),reviewed_event_seq:9,reviewed_event_sha256:H('2'),
+    journal_head_sha256:H('3'),scope:'resume_goal',subject:{diff_id:H('4'),operation_count:2}});
+  const approval={schema:'deadbyte.bound-approval.v1',approval_subject_sha256:subject.approval_subject_sha256,
+    reviewed_event_seq:9,reviewed_event_sha256:H('2'),journal_head_sha256:H('3'),scope:'resume_goal',
+    issued_at_utc:'2026-09-20T01:00:00.000Z',expires_at_utc:'2026-09-20T01:05:00.000Z',provenance:'operator_mcp'};
+  assert.equal(validateBoundApproval({approval,subject,now:new Date('2026-09-20T01:01:00Z')}),true);
+  assert.throws(()=>validateBoundApproval({approval:{...approval,journal_head_sha256:H('9')},subject,
+    now:new Date('2026-09-20T01:01:00Z')}),/journal|subject/i);
+  assert.throws(()=>validateBoundApproval({approval,subject:{...subject,subject:{diff_id:H('5'),operation_count:2}},
+    now:new Date('2026-09-20T01:01:00Z')}),/subject/i);
+  assert.throws(()=>validateBoundApproval({approval,subject,now:new Date('2026-09-20T01:06:00Z')}),/expired/i);
+});
+
+test('R42 canonical DAG propagates STALE only through affected descendants',()=>{
+  const dag=createTaskDag({run_id:'run-1',goal_id:H('a'),tasks:[
+    {task_id:'inspect',depends_on:[],input_artifacts:['source'],output_artifacts:['analysis']},
+    {task_id:'patch',depends_on:['inspect'],input_artifacts:['analysis'],output_artifacts:['diff']},
+    {task_id:'test',depends_on:['patch'],input_artifacts:['diff'],output_artifacts:['test-result']},
+    {task_id:'docs',depends_on:[],input_artifacts:['readme'],output_artifacts:['docs-result']}
+  ]});
+  const next=invalidateTaskDag(dag,{changed_artifact_ids:['analysis']});
+  assert.deepEqual(next.tasks.filter(task=>task.state==='STALE').map(task=>task.task_id),['patch','test']);
+  assert.equal(next.tasks.find(task=>task.task_id==='inspect').state,'PENDING');
+  assert.equal(next.tasks.find(task=>task.task_id==='docs').state,'PENDING');
+  assert.throws(()=>createTaskDag({run_id:'x',goal_id:H('a'),tasks:[
+    {task_id:'a',depends_on:['b'],input_artifacts:[],output_artifacts:[]},
+    {task_id:'b',depends_on:['a'],input_artifacts:[],output_artifacts:[]}
+  ]}),/cycle/i);
+});
+
+test('R42 one canonical DAG is deterministically derived from the authoritative plan',()=>{
+  const plan={steps:[{step_id:'inspect',status:'completed'},{step_id:'patch',status:'active'},{step_id:'verify',status:'pending'}]};
+  const one=taskDagFromPlan({run_id:H('7'),goal_id:H('7'),plan});
+  const two=taskDagFromPlan({run_id:H('7'),goal_id:H('7'),plan});
+  assert.equal(one.dag_sha256,two.dag_sha256);
+  assert.deepEqual(one.tasks.map(task=>[task.task_id,task.depends_on,task.state]),[
+    ['inspect',[],'COMPLETE'],['patch',['inspect'],'RUNNING'],['verify',['patch'],'PENDING']
+  ]);
+});
+
+test('R42 evidence ladder never lets weak evidence masquerade as runtime proof',()=>{
+  const source=createEvidenceClaim({claim_id:'claim-1',statement:'provider shape matches source',level:'source',refs:[H('b')]});
+  assert.equal(evidenceSatisfies(source,'source'),true);
+  assert.equal(evidenceSatisfies(source,'automated'),false);
+  assert.equal(evidenceSatisfies(source,'runtime'),false);
+  assert.throws(()=>createEvidenceClaim({claim_id:'claim-2',statement:'bad',level:'runtime',refs:[]}),/reference/i);
+});
+
+test('R42 rigor is derived once and cannot downgrade below risk minimum',()=>{
+  assert.deepEqual(deriveEffectiveRigor({project_default:'lightweight',risk_level:'high',explicit_override:'structured'}),{
+    project_default:'lightweight',risk_minimum:'agentic',explicit_override:'structured',effective_rigor:'agentic'
+  });
+  assert.equal(deriveEffectiveRigor({project_default:'structured',risk_level:'low',explicit_override:'agentic'}).effective_rigor,'agentic');
+});
+
+test('R42 decisions become stale deterministically without deleting history',()=>{
+  const record=createDecisionRecord({decision_id:'provider-choice',decision:'Use Responses adapter',reason:'strict bounded JSON',
+    evidence_refs:[H('c')],invalidates_if:[
+      {kind:'artifact_sha_changed',artifact_id:'provider-contract',expected_sha256:H('d')},
+      {kind:'policy_sha_changed',expected_sha256:H('e')}
+    ]});
+  assert.equal(evaluateDecisionFreshness(record,{artifact_sha256:{'provider-contract':H('d')},policy_sha256:H('e')}).state,'CURRENT');
+  const stale=evaluateDecisionFreshness(record,{artifact_sha256:{'provider-contract':H('f')},policy_sha256:H('e')});
+  assert.equal(stale.state,'STALE');
+  assert.deepEqual(stale.invalidated_by,[{kind:'artifact_sha_changed',artifact_id:'provider-contract',expected_sha256:H('d'),observed_sha256:H('f')}]);
+});

package/test/agent-runtime.test.mjs

@@ -1,4 +1,4 @@
-import test from 'node:test';
+import test from 'node:test';
 import assert from 'node:assert/strict';
 import { mkdtemp, mkdir, rm, writeFile } from 'node:fs/promises';
 import os from 'node:os';
@@ -20,7 +20,7 @@ const AGENT_TOOLS = [
   'agent_task_submit',
   'agent_task_verify'
 ];
-const BASE_TOOLS = ['contained_transform', 'evidence_verify', 'file_read', 'proof_execute', 'workspace_open'];
+const BASE_TOOLS = ['contained_transform', 'evidence_verify', 'file_read', 'proof_execute', 'skill_catalog', 'workspace_open'];
 
 function structured(result, label) {
   if (result.isError) throw new Error(`${label}: ${JSON.stringify(result.structuredContent ?? result.content)}`);
@@ -68,7 +68,7 @@ async function cleanup(f, client) {
   await rm(f.trust, { recursive: true, force: true }).catch(() => {});
 }
 
-test('agent runtime without an explicit host policy exposes only base plus durable task tools', async () => {
+test('R44 agent runtime exposes base skill catalog plus durable task tools without host authority', async () => {
   const f = await fixture();
   let client;
   try {

package/test/autonomous-loop.test.mjs

@@ -32,7 +32,7 @@ async function startPlanner(decisions) {
   await new Promise(resolve => server.listen(0,'127.0.0.1',resolve));
   return { server, prompts, endpoint:`http://127.0.0.1:${server.address().port}/v1/chat/completions` };
 }
-async function fixture(decisionPrefix = []) {
+async function fixture(decisionPrefix = [], options = {}) {
   const root = await mkdtemp(path.join(os.tmpdir(),'deadbyte-autonomous-loop-'));
   const project = path.join(root,'project'); const trust = path.join(root,'trust');
   await mkdir(project); await mkdir(trust);
@@ -101,14 +101,17 @@ async function fixture(decisionPrefix = []) {
   const policy = await loadAutonomousPolicy(autonomousPolicyPath,codingPolicy);
   await writeFile(policy.lease_file,JSON.stringify({ schema:'deadbyte.autonomous-lease.v1',
     policy_sha256:policy.policy_sha256, instance_nonce:randomBytes(32).toString('hex'), root_ids:['project'],
-    armed_at_utc:new Date().toISOString(), expires_at_utc:new Date(Date.now()+1800000).toISOString() }));
+    armed_at_utc:new Date().toISOString(), expires_at_utc:new Date(Date.now()+(options.autonomousLeaseMs??1800000)).toISOString() }));
   const codingRuntime = createCodingRuntime({ policy:codingPolicy,
     signingKeyPath:keys.privateKeyPath, verifyKeyPath:keys.publicKeyPath });
+  const autonomousCodingRuntime = options.codingRuntimeFactory
+    ? options.codingRuntimeFactory(codingRuntime,{policy})
+    : codingRuntime;
   const nativeDiff = createNativeDiffRuntime({ policy,codingRuntime,
     signingKeyPath:keys.privateKeyPath, verifyKeyPath:keys.publicKeyPath });
-  const runtime = createAutonomousRuntime({ policy,codingRuntime,nativeDiffRuntime:nativeDiff,
+  const runtime = createAutonomousRuntime({ policy,codingRuntime:autonomousCodingRuntime,nativeDiffRuntime:nativeDiff,
     signingKeyPath:keys.privateKeyPath, verifyKeyPath:keys.publicKeyPath });
-  return { root,project,program,keys,policy,codingRuntime,nativeDiff,runtime,planner,
+  return { root,project,program,keys,policy,codingRuntime,autonomousCodingRuntime,nativeDiff,runtime,planner,
     initialHash,almostHash,fixedHash };
 }
 
@@ -144,6 +147,83 @@ test('R24 autonomous goal diagnoses a failing test, replans with native diff, be
   }
 });
 
+test('autonomous profile execution is capped by the absolute autonomous lease without changing coding request identity', async () => {
+  const calls=[];
+  const f=await fixture([],{
+    autonomousLeaseMs:30000,
+    codingRuntimeFactory:base=>Object.freeze({
+      ...base,
+      commandRun:async args=>{
+        const result=await base.commandRun(args);
+        calls.push({kind:'command',deadline:args.executionDeadlineUtc,has_precheck:typeof args.preExecutionCheck==='function',
+          timeout_ms:result.timeout_ms,request_sha256:result.request_sha256});
+        return result;
+      },
+      benchmarkRun:async args=>{
+        const result=await base.benchmarkRun(args);
+        calls.push({kind:'benchmark',deadline:args.executionDeadlineUtc,has_precheck:typeof args.preExecutionCheck==='function',
+          request_sha256:result.request_sha256});
+        return result;
+      }
+    })
+  });
+  try {
+    const submitted=await f.runtime.submit({submissionId:'profile-deadline',rootId:'project',
+      title:'deadline-bound profiles',goal:'make exported value equal fixed',
+      acceptance:'build, test, and benchmark must pass',maxIterations:10});
+    const result=await f.runtime.run({goalId:submitted.goal_id,maxCycles:16});
+    assert.equal(result.status,'completed');
+    assert.ok(calls.length>=3);
+    assert.ok(calls.every(call=>typeof call.deadline==='string'&&Number.isFinite(Date.parse(call.deadline))));
+    assert.ok(calls.every(call=>call.has_precheck===true));
+    const command=calls.find(call=>call.kind==='command');
+    assert.ok(command.timeout_ms>=100&&command.timeout_ms<30000);
+    const build=f.policy.profiles.build;
+    assert.equal(command.request_sha256,f.codingRuntime.requestSha('coding_command_run',{
+      root_id:build.root_id,command_id:build.command_id,script_path:build.script_path,script_sha256:build.script_sha256,
+      args:build.args,cwd:build.cwd,timeout_ms:build.timeout_ms
+    }));
+  } finally {
+    await new Promise(resolve=>f.planner.server.close(resolve));
+    await rm(f.root,{recursive:true,force:true});
+  }
+});
+
+test('autonomous profile nonce rotation immediately before spawn fails closed without a coding receipt', async () => {
+  let rotated=false;
+  const f=await fixture([],{
+    codingRuntimeFactory:(base,{policy})=>Object.freeze({
+      ...base,
+      commandRun:async args=>{
+        if(!rotated){
+          rotated=true;
+          const lease=JSON.parse(await readFile(policy.lease_file,'utf8'));
+          await writeFile(policy.lease_file,JSON.stringify({...lease,instance_nonce:'f'.repeat(64)}));
+        }
+        return base.commandRun(args);
+      }
+    })
+  });
+  try {
+    const submitted=await f.runtime.submit({submissionId:'profile-revoke-before-spawn',rootId:'project',
+      title:'profile revoke',goal:'make exported value equal fixed',
+      acceptance:'profile dispatch must fail closed after authority rotation',maxIterations:10});
+    const result=await f.runtime.run({goalId:submitted.goal_id,maxCycles:16});
+    assert.equal(rotated,true);
+    assert.equal(result.status,'paused');
+    const journal=await f.runtime.events({goalId:submitted.goal_id,fromSeq:1,limit:200});
+    const started=journal.events.find(event=>event.type==='profile_started'&&event.payload?.profile_id==='build');
+    assert.ok(started);
+    assert.match(started.payload.authority_instance_nonce,/^[0-9a-f]{64}$/);
+    assert.ok(journal.events.some(event=>event.type==='goal_paused'&&event.payload?.reason==='profile_execution_without_receipt'));
+    assert.equal(journal.events.some(event=>event.type==='profile_completed'&&event.payload?.profile_id==='build'),false);
+    assert.equal((await f.codingRuntime.verifyReceiptDetailed(started.payload.receipt_id)).ok,false);
+  } finally {
+    await new Promise(resolve=>f.planner.server.close(resolve));
+    await rm(f.root,{recursive:true,force:true});
+  }
+});
+
 test('provider planner repairs an invalid structured decision inside the same run', async () => {
   const f = await fixture(['not-json']);
   try {
@@ -166,7 +246,7 @@ test('provider planner repairs an invalid structured decision inside the same ru
   }
 });
 
-test('provider circuit survives runtime restart, suppresses OPEN calls, then permits one HALF_OPEN probe', async () => {
+test('provider circuit survives runtime restart and permits exactly one probe at the HALF_OPEN boundary', async () => {
   const failures=[{__http_status:503},{__http_status:503},{__http_status:503}];
   const f=await fixture(failures);
   try {
@@ -178,17 +258,23 @@ test('provider circuit survives runtime restart, suppresses OPEN calls, then per
     const opened=await f.runtime.run({goalId:submitted.goal_id,maxCycles:2});
     assert.equal(opened.status,'paused');
     assert.equal(opened.reason,'provider_circuit_open');
+    const openedJournal=await f.runtime.events({goalId:submitted.goal_id,fromSeq:1,limit:200});
+    const openPause=[...openedJournal.events].reverse().find(event=>event.type==='goal_paused'&&event.payload?.reason==='provider_circuit_open');
+    assert.equal(openPause?.payload?.circuit_state,'OPEN',JSON.stringify(openPause));
     assert.equal(f.planner.prompts.length,3);
 
-    const immediate=await f.runtime.run({goalId:submitted.goal_id,maxCycles:2});
-    assert.equal(immediate.status,'paused');
-    assert.equal(f.planner.prompts.length,3,'OPEN circuit must suppress provider invocation');
-
     const restarted=createAutonomousRuntime({policy:f.policy,codingRuntime:f.codingRuntime,nativeDiffRuntime:f.nativeDiff,
       signingKeyPath:f.keys.privateKeyPath,verifyKeyPath:f.keys.publicKeyPath});
-    await new Promise(resolve=>setTimeout(resolve,130));
-    const circuitBeforeResume=await restarted.providerCircuitStatus({goalId:submitted.goal_id});
+    const deadline=Date.now()+2_000;
+    let circuitBeforeResume;
+    do {
+      circuitBeforeResume=await restarted.providerCircuitStatus({goalId:submitted.goal_id});
+      if(circuitBeforeResume.state==='HALF_OPEN') break;
+      assert.equal(circuitBeforeResume.state,'OPEN',JSON.stringify(circuitBeforeResume));
+      await new Promise(resolve=>setTimeout(resolve,10));
+    } while(Date.now()<deadline);
     assert.equal(circuitBeforeResume.state,'HALF_OPEN',JSON.stringify(circuitBeforeResume));
+
     const resumed=await restarted.run({goalId:submitted.goal_id,maxCycles:1});
     assert.notEqual(resumed.status,'paused');
     assert.equal(f.planner.prompts.length,4,'HALF_OPEN must permit exactly one provider probe');

package/test/autonomous-r34-e2e.test.mjs

@@ -75,7 +75,8 @@ test('R34 missing authority waits durably and approval resumes the same phase af
     assert.equal(journal.events.filter(event=>event.type==='action_started').length,0);
 
     await f.arm();
-    const approved = await f.runtime.approve({goalId:goal.goal_id,decision:'approve'});
+    const approved = await f.runtime.approve({goalId:goal.goal_id,decision:'approve',
+      approvalSubjectSha256:result.loop.approval_subject.approval_subject_sha256});
     assert.equal(approved.status,'approved');
     assert.equal(approved.loop.phase,'ACTING');
     const resumed = await f.runtime.run({goalId:goal.goal_id,maxCycles:2});

package/test/autonomous-runtime.test.mjs

@@ -20,7 +20,7 @@ const r24Tools = [
   'native_diff_apply', 'native_diff_preview', 'native_diff_verify'
 ];
 const r23Tools = [
-  'contained_transform','evidence_verify','file_read','proof_execute','workspace_open',
+  'contained_transform','evidence_verify','file_read','proof_execute','skill_catalog','workspace_open',
   'agent_task_approve','agent_task_cancel','agent_task_events','agent_task_run',
   'agent_task_status','agent_task_submit','agent_task_verify',
   'coding_benchmark_run','coding_command_run','coding_file_read','coding_file_write','coding_files_read',
@@ -75,7 +75,7 @@ async function fixture() {
   return { root, client, transport };
 }
 
-test('R36 hermetic autonomous fixture exposes the exact 49-tool no-host catalog', async () => {
+test('R44 hermetic autonomous fixture exposes the exact 50-tool no-host catalog', async () => {
   const f = await fixture();
   try {
     await f.client.connect(f.transport);

package/test/capability-ledger-r36.test.mjs

@@ -92,12 +92,12 @@ test('R36 closure fails when a production route is lost or advertised tool is ab
   }), /compact tool/i);
 });
 
-test('R36 compact surface hard-fails above the observed 42-tool ceiling', () => {
+test('R44 compact surface hard-fails above the observed 43-tool ceiling', () => {
   const ledger=createCapabilityLedger([entry({})]);
-  const tooMany=Array.from({length:43},(_,i)=>i===0?'coding_inspect':`extra_${i}`);
+  const tooMany=Array.from({length:44},(_,i)=>i===0?'coding_inspect':`extra_${i}`);
   assert.throws(() => assertCapabilityClosure({
     ledger,
     fullToolSnapshot:snapshot(['coding_search']),
     compactToolSnapshot:snapshot(tooMany)
-  }), /42/i);
+  }), /43/i);
 });

package/test/cli-entrypoint.test.mjs

@@ -21,7 +21,7 @@ test('npx CLI reports the canonical package version through --version, -v and ve
   for(const arg of ['--version','-v','version']){
     const result=spawnSync(process.execPath,[cliPath,arg],{cwd:root,encoding:'utf8',windowsHide:true});
     assert.equal(result.status,0,result.stderr);
-    assert.equal(result.stdout.trim(),'0.13.0');
+    assert.equal(result.stdout.trim(),'0.15.0');
   }
 });
 

package/test/codex-app-server-r44.test.mjs

@@ -0,0 +1,115 @@
+import test from 'node:test';
+import assert from 'node:assert/strict';
+import { EventEmitter } from 'node:events';
+import { mkdtemp, rm } from 'node:fs/promises';
+import os from 'node:os';
+import path from 'node:path';
+import { createCodexAppServerR44 } from '../src/codex-app-server-r44.mjs';
+
+async function waitFor(predicate,timeoutMs=2000){
+  const deadline=Date.now()+timeoutMs;
+  while(Date.now()<deadline){const value=predicate();if(value)return value;await new Promise(resolve=>setTimeout(resolve,5));}
+  throw new Error('waitFor timeout');
+}
+class FakeTransport extends EventEmitter{
+  constructor(){super();this.sent=[];}
+  async start(){}
+  async send(value){this.sent.push(structuredClone(value));}
+  async stop(){}
+  request(method){return this.sent.findLast?.(x=>x.method===method&&Object.hasOwn(x,'id'))??[...this.sent].reverse().find(x=>x.method===method&&Object.hasOwn(x,'id'));}
+  reply(id,result){this.emit('message',{id,result});}
+  notify(method,params){this.emit('message',{method,params});}
+  serverRequest(id,method,params){this.emit('message',{id,method,params});}
+}
+
+test('R44 Codex adapter initializes no-jsonrpc wire and starts safe persistent planner thread',async()=>{
+  const root=await mkdtemp(path.join(os.tmpdir(),'deadbyte-r44-codex-'));const transport=new FakeTransport();
+  try{
+    const codex=await createCodexAppServerR44({stateDir:root,transportFactory:async()=>transport,
+      model:'gpt-test',cwd:root,clientVersion:'0.15.0'});
+    const start=codex.start();
+    const init=await waitFor(()=>transport.request('initialize'));
+    assert.equal('jsonrpc' in init,false);
+    assert.deepEqual(init.params.clientInfo,{name:'deadbyte-mcp',version:'0.15.0'});
+    assert.equal(init.params.capabilities.experimentalApi,false);
+    transport.reply(init.id,{userAgent:'codex-test'});
+    await start;
+    const initialized=transport.sent.find(x=>x.method==='initialized'&&!Object.hasOwn(x,'id'));
+    assert.ok(initialized);assert.equal('jsonrpc' in initialized,false);
+
+    const ensure=codex.ensureThread({logicalId:'planner'});
+    const req=await waitFor(()=>transport.request('thread/start'));
+    assert.equal(req.params.model,'gpt-test');assert.equal(req.params.cwd,root);
+    assert.equal(req.params.approvalPolicy,'never');assert.equal(req.params.sandbox,'read-only');
+    assert.equal(req.params.ephemeral,false);assert.equal(req.params.approvalsReviewer,'user');
+    transport.reply(req.id,{thread:{id:'thread-1'},model:'gpt-test',modelProvider:'openai',cwd:root,
+      approvalPolicy:'never',approvalsReviewer:'user',sandbox:{type:'readOnly'}});
+    const thread=await ensure;assert.equal(thread.thread_id,'thread-1');assert.equal(thread.resumed,false);
+    await codex.close();
+  }finally{await rm(root,{recursive:true,force:true});}
+});
+
+test('R44 Codex planner routes final answer by turn id and declines provider approvals',async()=>{
+  const root=await mkdtemp(path.join(os.tmpdir(),'deadbyte-r44-codex-turn-'));const transport=new FakeTransport();
+  try{
+    const codex=await createCodexAppServerR44({stateDir:root,transportFactory:async()=>transport,model:'gpt-test',cwd:root});
+    const start=codex.start();const init=await waitFor(()=>transport.request('initialize'));transport.reply(init.id,{});await start;
+    const ensure=codex.ensureThread({logicalId:'planner'});const ts=await waitFor(()=>transport.request('thread/start'));
+    transport.reply(ts.id,{thread:{id:'thread-1'},model:'gpt-test'});await ensure;
+
+    const planned=codex.runPlanner({logicalId:'planner',prompt:'Return JSON',timeoutMs:2000});
+    const turn=await waitFor(()=>transport.request('turn/start'));
+    assert.equal(turn.params.threadId,'thread-1');assert.deepEqual(turn.params.input,[{type:'text',text:'Return JSON'}]);
+    assert.equal(turn.params.approvalPolicy,'never');
+    transport.reply(turn.id,{turn:{id:'turn-1',status:'inProgress'}});
+    transport.notify('item/completed',{threadId:'thread-1',turnId:'other',item:{id:'x',type:'agentMessage',text:'wrong'}});
+    transport.notify('item/completed',{threadId:'thread-1',turnId:'turn-1',item:{id:'m1',type:'agentMessage',text:'working',phase:'commentary'}});
+    transport.notify('item/completed',{threadId:'thread-1',turnId:'turn-1',item:{id:'m2',type:'agentMessage',text:'{"schema":"ok"}',phase:'final_answer'}});
+    transport.notify('turn/completed',{threadId:'thread-1',turn:{id:'turn-1',status:'completed'}});
+    const result=await planned;
+    assert.equal(result.backend,'codex_app_server');assert.equal(result.raw_text,'{"schema":"ok"}');assert.equal(result.response_id,'turn-1');
+
+    transport.serverRequest('approval-1','item/commandExecution/requestApproval',{threadId:'thread-1',turnId:'turn-1'});
+    const denial=await waitFor(()=>transport.sent.find(x=>x.id==='approval-1'));
+    assert.deepEqual(denial.result,{decision:'decline'});assert.equal(denial.error,undefined);
+    await codex.close();
+  }finally{await rm(root,{recursive:true,force:true});}
+});
+
+test('R44 Codex steer and interrupt bind exact active turn identity',async()=>{
+  const root=await mkdtemp(path.join(os.tmpdir(),'deadbyte-r44-codex-control-'));const transport=new FakeTransport();
+  try{
+    const codex=await createCodexAppServerR44({stateDir:root,transportFactory:async()=>transport,model:'gpt-test',cwd:root});
+    const start=codex.start();const init=await waitFor(()=>transport.request('initialize'));transport.reply(init.id,{});await start;
+    const steerP=codex.steer({threadId:'thread-1',turnId:'turn-1',text:'new direction',timeoutMs:2000});
+    const steer=await waitFor(()=>transport.request('turn/steer'));
+    assert.deepEqual(steer.params,{threadId:'thread-1',expectedTurnId:'turn-1',input:[{type:'text',text:'new direction'}]});
+    transport.reply(steer.id,{turnId:'turn-1'});await steerP;
+    const interruptP=codex.interrupt({threadId:'thread-1',turnId:'turn-1',timeoutMs:2000});
+    const interrupt=await waitFor(()=>transport.request('turn/interrupt'));
+    assert.deepEqual(interrupt.params,{threadId:'thread-1',turnId:'turn-1'});
+    transport.reply(interrupt.id,{});await interruptP;
+    await codex.close();
+  }finally{await rm(root,{recursive:true,force:true});}
+});
+
+test('R44 Codex adapter resumes durable thread after backend process recreation',async()=>{
+  const root=await mkdtemp(path.join(os.tmpdir(),'deadbyte-r44-codex-resume-'));
+  try{
+    const firstTransport=new FakeTransport();
+    const first=await createCodexAppServerR44({stateDir:root,transportFactory:async()=>firstTransport,model:'gpt-test',cwd:root});
+    const s1=first.start();const i1=await waitFor(()=>firstTransport.request('initialize'));firstTransport.reply(i1.id,{});await s1;
+    const e1=first.ensureThread({logicalId:'planner'});const t1=await waitFor(()=>firstTransport.request('thread/start'));
+    firstTransport.reply(t1.id,{thread:{id:'thread-persist'},model:'gpt-test'});await e1;await first.close();
+
+    const secondTransport=new FakeTransport();
+    const second=await createCodexAppServerR44({stateDir:root,transportFactory:async()=>secondTransport,model:'gpt-test',cwd:root});
+    const s2=second.start();const i2=await waitFor(()=>secondTransport.request('initialize'));secondTransport.reply(i2.id,{});await s2;
+    const e2=second.ensureThread({logicalId:'planner'});const resume=await waitFor(()=>secondTransport.request('thread/resume'));
+    assert.equal(resume.params.threadId,'thread-persist');assert.equal(resume.params.approvalPolicy,'never');assert.equal(resume.params.sandbox,'read-only');
+    secondTransport.reply(resume.id,{thread:{id:'thread-persist'},model:'gpt-test'});
+    const result=await e2;assert.equal(result.thread_id,'thread-persist');assert.equal(result.resumed,true);
+    assert.equal(secondTransport.sent.some(x=>x.method==='thread/start'),false);
+    await second.close();
+  }finally{await rm(root,{recursive:true,force:true});}
+});

package/test/codex-autonomous-r44.test.mjs

@@ -0,0 +1,87 @@
+import test from 'node:test';
+import assert from 'node:assert/strict';
+import { createHash, randomBytes } from 'node:crypto';
+import { mkdtemp, mkdir, readFile, rm, writeFile } from 'node:fs/promises';
+import os from 'node:os';
+import path from 'node:path';
+import { generateTrustKeypair } from '../src/trust.mjs';
+import { createCodingRuntime, loadCodingPolicy } from '../src/coding-plane.mjs';
+import { loadAutonomousPolicy } from '../src/autonomous-policy.mjs';
+import { createNativeDiffRuntime } from '../src/native-diff.mjs';
+import { createAutonomousRuntime } from '../src/autonomous-runtime.mjs';
+import { providerCircuitId } from '../src/provider-circuit.mjs';
+
+const sha=b=>createHash('sha256').update(b).digest('hex');
+const nonce=()=>randomBytes(32).toString('hex');
+
+async function fixture(){
+  const root=await mkdtemp(path.join(os.tmpdir(),'deadbyte-r44-codex-auto-'));
+  const project=path.join(root,'project'),trust=path.join(root,'trust');
+  await mkdir(project);await mkdir(trust);
+  const keys=await generateTrustKeypair({privateKeyPath:path.join(trust,'private.pem'),publicKeyPath:path.join(trust,'public.pem')});
+  const codingPath=path.join(root,'coding.json');
+  await writeFile(codingPath,JSON.stringify({schema:'deadbyte.coding-policy.v1',enabled:true,
+    lease_file:path.join(root,'coding-grant.json'),evidence_dir:path.join(root,'coding-evidence'),
+    roots:{project:{path:project,read:true,write:true,exec:false,deny_paths:[]}},commands:{}},null,2));
+  const codingPolicy=await loadCodingPolicy(codingPath);
+  const nodeSha=sha(await readFile(process.execPath));
+  const autonomousPath=path.join(root,'autonomous.json');
+  await writeFile(autonomousPath,JSON.stringify({schema:'deadbyte.autonomous-policy.v1',enabled:true,
+    lease_file:path.join(root,'autonomous-grant.json'),evidence_dir:path.join(root,'autonomous-evidence'),
+    coding_policy_sha256:codingPolicy.policy_sha256,root_ids:['project'],
+    planner:{backend:'codex_app_server',model:'gpt-test',root_id:'project',executable:process.execPath,
+      executable_sha256:nodeSha,args:['app-server'],max_tokens:4096,timeout_ms:30000,
+      circuit:{failure_threshold:2,open_ms:1000,half_open_probe_limit:1}},
+    profiles:{},limits:{max_iterations:8,max_mutations:4,max_wall_ms:600000,max_context_bytes:262144,
+      max_diff_bytes:262144,max_files_per_diff:4,max_stalled_cycles:2,max_post_replan_stalls:2,max_replans:4}},null,2));
+  const policy=await loadAutonomousPolicy(autonomousPath,codingPolicy);
+  await writeFile(codingPolicy.lease_file,JSON.stringify({schema:'deadbyte.coding-grant.v2',policy_sha256:codingPolicy.policy_sha256,
+    instance_nonce:nonce(),root_ids:['project'],armed_at_utc:new Date().toISOString()}));
+  await writeFile(policy.lease_file,JSON.stringify({schema:'deadbyte.autonomous-grant.v2',policy_sha256:policy.policy_sha256,
+    instance_nonce:nonce(),root_ids:['project'],armed_at_utc:new Date().toISOString()}));
+  const codingRuntime=createCodingRuntime({policy:codingPolicy,signingKeyPath:keys.privateKeyPath,verifyKeyPath:keys.publicKeyPath});
+  const nativeDiffRuntime=createNativeDiffRuntime({policy,codingRuntime,signingKeyPath:keys.privateKeyPath,verifyKeyPath:keys.publicKeyPath});
+  return {root,project,keys,codingPolicy,policy,codingRuntime,nativeDiffRuntime,cleanup:()=>rm(root,{recursive:true,force:true})};
+}
+
+test('R44 policy binds Codex executable, model, root and circuit identity',async()=>{
+  const f=await fixture();
+  try{
+    assert.equal(f.policy.planner.backend,'codex_app_server');
+    assert.equal(f.policy.planner.root_id,'project');
+    assert.equal(f.policy.planner.executable,process.execPath);
+    assert.match(f.policy.planner.executable_sha256,/^[0-9a-f]{64}$/);
+    assert.deepEqual(f.policy.planner.args,['app-server']);
+    assert.match(providerCircuitId(f.policy.planner),/^codex_app_server:[0-9a-f]{64}$/);
+  }finally{await f.cleanup();}
+});
+
+test('R44 autonomous runtime consumes injected Codex planner and does not ask client sampling',async()=>{
+  const f=await fixture();const calls=[];
+  try{
+    const plannerProviderRuntime={runPlanner:async({logicalId,prompt,timeoutMs})=>{
+      calls.push({logicalId,prompt,timeoutMs});
+      return {backend:'codex_app_server',model:'gpt-test',raw_text:JSON.stringify({
+        schema:'deadbyte.autonomous-decision.v2',kind:'plan_create',title:'Provider path',
+        reason:'prove Codex planner integration',steps:[{step_id:'prove',intent:'observe provider call'}]
+      }),response_id:'turn-test-1'};
+    }};
+    const runtime=createAutonomousRuntime({policy:f.policy,codingRuntime:f.codingRuntime,nativeDiffRuntime:f.nativeDiffRuntime,
+      plannerProviderRuntime,signingKeyPath:f.keys.privateKeyPath,verifyKeyPath:f.keys.publicKeyPath});
+    const submitted=await runtime.submit({submissionId:'codex-plan',rootId:'project',title:'codex',goal:'prove provider path',acceptance:'pause'});
+    const result=await runtime.run({goalId:submitted.goal_id,maxCycles:1});
+    assert.notEqual(result.status,'input_required');
+    assert.equal(result.status,'yielded');
+    assert.equal(calls.length,1);
+    assert.equal(calls[0].logicalId,`goal-${submitted.goal_id}`);
+    assert.match(calls[0].prompt,/DEADBYTE/);
+  }finally{await f.cleanup();}
+});
+
+test('R44 autonomous Codex policy fails closed without injected provider runtime',async()=>{
+  const f=await fixture();
+  try{
+    assert.throws(()=>createAutonomousRuntime({policy:f.policy,codingRuntime:f.codingRuntime,nativeDiffRuntime:f.nativeDiffRuntime,
+      signingKeyPath:f.keys.privateKeyPath,verifyKeyPath:f.keys.publicKeyPath}),/codex.*provider/i);
+  }finally{await f.cleanup();}
+});

package/test/coding-plane.test.mjs

@@ -19,7 +19,7 @@ const codingTools = [
   'coding_task_events', 'coding_task_run', 'coding_task_status', 'coding_task_submit',
   'coding_task_verify', 'coding_text_replace', 'coding_tree'
 ];
-const baseTools = ['contained_transform', 'evidence_verify', 'file_read', 'proof_execute', 'workspace_open'];
+const baseTools = ['contained_transform', 'evidence_verify', 'file_read', 'proof_execute', 'skill_catalog', 'workspace_open'];
 const agentTools = [
   'agent_task_approve', 'agent_task_cancel', 'agent_task_events', 'agent_task_run',
   'agent_task_status', 'agent_task_submit', 'agent_task_verify'
@@ -69,7 +69,7 @@ async function fixture() {
   return { root, client, transport };
 }
 
-test('R36 hermetic coding fixture exposes the exact 37-tool no-host catalog', async () => {
+test('R44 hermetic coding fixture exposes the exact 38-tool no-host catalog', async () => {
   const f = await fixture();
   try {
     await f.client.connect(f.transport);