feat: integrate pinned original Purge with host execution gates

This commit is contained in:
Agent Workbench Dev
2026-11-02 22:32:31 +08:00
parent 1408e7ac76
commit 9106655d46
26 changed files with 3217 additions and 38 deletions
+24 -11
View File
@@ -9,7 +9,7 @@ import { Runtime } from '../src/server/runtime.mjs';
import { EngineRunner } from '../src/core/engines/engine-runner.mjs';
import { resolveControlledInstallation } from '../src/core/engines/controlled-installation.mjs';
import { originalEngineRun, verifyEngineModelInput } from '../src/shared/contracts/engine-execution.mjs';
import { resolvePurgeSource, assertPurgeModelInput, nativeSystemTexts, PURGE_DEFAULT_PROMPT } from '../src/core/engines/purge-strategy.mjs';
import { resolvePurgeSource, resolveOriginalPurgeSource, ORIGINAL_PURGE_KIND, assertPurgeModelInput, nativeSystemTexts, PURGE_DEFAULT_PROMPT } from '../src/core/engines/purge-strategy.mjs';
import { deliveryGit as git, writeDeliveryFile as write } from '../tests/fixtures/delivery-project.mjs';
import { ownerBytes } from '../tests/fixtures/workers-project.mjs';
import { desktopProjectQa } from './desktop-project-qa.mjs';
@@ -18,12 +18,15 @@ import { responsesEngineeringHttp } from '../tests/fixtures/responses-engineerin
import { nativeContextSpecification } from '../tests/fixtures/native-context-project.mjs';
const args = process.argv.slice(2);
assert.ok(args.every(value => ['--help', '--purge-comparison', '--desktop', '--responses', '--context-16k'].includes(value)), 'Unknown option; use --help.');
assert.ok(args.every(value => ['--help', '--purge-comparison', '--desktop', '--responses', '--context-16k','--context-32k','--original-purge'].includes(value)), 'Unknown option; use --help.');
if (args.includes('--help')) {
console.log('node scripts/verify-controlled-projects.mjs [--purge-comparison | --desktop | --context-16k] [--responses] — actual pinned Harness and owned HTTP model fixture; default verifies Node/React/Python with approvals, Edge and Git. --context-16k uses a new copy of the sealed three-module/nine-test pilot, original failure/log paging and a complete reviewed batch at 16K. --responses uses actual Responses SSE and ModelHub routing. Set AGENT_WORKBENCH_DEEPSEEK_ROOT and AGENT_WORKBENCH_TEST_PYTHON; context mode needs no Python. No configured-model requests.'); process.exit(0);
console.log('node scripts/verify-controlled-projects.mjs [--purge-comparison | --desktop | --context-16k | --context-32k] [--responses] [--original-purge] — actual pinned Harness and owned HTTP model fixture; default verifies Node/React/Python with approvals, Edge and Git. Context modes use a new copy of the sealed three-module/nine-test pilot, original failure/log paging and a complete reviewed batch at the explicit limit. Original Purge is separate from the independent prompt and its fixed instructions need the 32K profile. --responses uses actual Responses SSE and ModelHub routing. Set AGENT_WORKBENCH_DEEPSEEK_ROOT and AGENT_WORKBENCH_TEST_PYTHON; context mode needs no Python. No configured-model requests.'); process.exit(0);
}
const comparisonMode = args.includes('--purge-comparison'), desktopMode = args.includes('--desktop');
const contextMode = args.includes('--context-16k');assert.ok(!contextMode || !comparisonMode && !desktopMode, '16K bridge acceptance is a separate run.');
assert.ok(!(args.includes('--context-16k')&&args.includes('--context-32k')),'Choose one explicit native context limit.');
const contextMode = args.includes('--context-16k')||args.includes('--context-32k');assert.ok(!contextMode || !comparisonMode && !desktopMode, 'Native context acceptance is a separate run.');
const contextLimit=args.includes('--context-32k')?32000:16000;
const originalPurgeMode=args.includes('--original-purge');assert.ok(!originalPurgeMode||contextMode,'Original Purge engineering acceptance uses a separate owned context case.');
const responsesMode = args.includes('--responses'); assert.ok(!responsesMode || !comparisonMode, 'Protocol acceptance and Purge comparison remain separate.');
assert.ok(!comparisonMode || !desktopMode, 'Desktop engineering and prompt comparison are separate acceptance runs.');
const version = JSON.parse(fs.readFileSync('package.json', 'utf8')).version;
@@ -31,6 +34,9 @@ const directory = path.resolve('.qa', (contextMode ? 'controlled-context-' : des
const hash = bytes => createHash('sha256').update(bytes).digest('hex');
const installation = resolveControlledInstallation(process.cwd(), process.env.AGENT_WORKBENCH_DEEPSEEK_ROOT);
assert.equal(installation.controlledStatus, 'available', installation.message);
const originalSource=originalPurgeMode?resolveOriginalPurgeSource(process.cwd(),process.env.AGENT_WORKBENCH_PURGE_178_ROOT,installation):null;
if(originalSource)assert.equal(originalSource.status,'available',originalSource.message);
const originalStrategy=originalSource?{kind:ORIGINAL_PURGE_KIND,root:originalSource.root,sourceSha256:originalSource.sourceSha256,runtimeSha256:originalSource.runtimeSha256,prompt:PURGE_DEFAULT_PROMPT}:null;
const python = process.env.AGENT_WORKBENCH_TEST_PYTHON;
if (!comparisonMode && !contextMode) assert.ok(python && path.isAbsolute(python) && fs.existsSync(python), 'Provide the existing isolated Python 3.10+ interpreter.');
const sources = comparisonMode ? ['1.1.78', '1.1.81'].map((expected, index) => {
@@ -111,12 +117,12 @@ async function runSample(kind, { sampleName = kind, strategy = null, comparison
return { choices: [{ message: { role: 'assistant', content: calls.length ? '' : 'Owned native run complete; host proof is still required.', ...(calls.length ? { tool_calls: calls } : {}) },
finish_reason: calls.length ? 'tool_calls' : 'stop' }], usage: { prompt_tokens: 123, completion_tokens: 45, total_tokens: 168 } };
});
f.store.saveConfig({ mode: 'compatible', model: 'owned-controlled-engineering', baseUrl: model.baseUrl, contextLimit: contextMode ? 16000 : 64000,
f.store.saveConfig({ mode: 'compatible', model: 'owned-controlled-engineering', baseUrl: model.baseUrl, contextLimit: contextMode ? contextLimit : 64000,
...(contextMode ? {maxOutputTokens:2048} : {}) });
const provider = new Provider(); provider.setKey('self-owned-controlled-model-fixture');
if (responsesMode) {
const hub = provider.modelHub(f.store), connection = hub.upsert({ name: 'Owned Responses engineering', protocol: 'responses', model: 'owned-controlled-engineering', baseUrl: model.baseUrl,
requiresKey: false, contextLimit: contextMode ? 16000 : 64000, ...(contextMode ? {maxOutputTokens:2048} : {}), streaming: true, capabilityDeclarations: { streaming: 'supported' } });
requiresKey: false, contextLimit: contextMode ? contextLimit : 64000, ...(contextMode ? {maxOutputTokens:2048} : {}), streaming: true, capabilityDeclarations: { streaming: 'supported' } });
hub.select({ type: 'connection', id: connection.id });
}
runtime = new Runtime(f.store, provider); f.onCleanup(() => runtime.stopAll());
@@ -166,10 +172,17 @@ async function runSample(kind, { sampleName = kind, strategy = null, comparison
const engineRuns = [...f.store.core.engine_run.values()].filter(job => job.taskId === child.taskId), childSession = f.store.session(child.sessionId);
for (const job of engineRuns) {
originalEngineRun(f.store, childSession, job); assert.equal(job.status, 'completed', job.error); assert.equal(job.native.cleanup.treeConfirmed, true);
assert.equal(job.native.stderr, ''); assert.equal(job.native.activity.endEvents[0].data.reason.kind, 'completed');
if(originalPurgeMode){
const warning='(node:'+job.native.process.pid+') [PERM0001] SecurityWarning: The flag --allow-addons must be used with extreme caution. It could invalidate the permission model.\n(Use `node --trace-warnings ...` to show where the warning was created)\n';
const expected=warning+['package.json','lib/index.js'].map(file=>'AWB_MODULE_REFUSED '+path.join(installation.root,'packages/context/agent-instructions',file)+'\n').join('');
assert.equal(job.native.stderr,expected);assert.equal(job.native.originalPurge.readiness.originalApplyCalled,true);
assert.equal(job.native.originalPurge.nativeLoads.length,2);assert.ok(job.native.originalPurge.durabilityReceipts.length>0);
}else assert.equal(job.native.stderr, ''); assert.equal(job.native.activity.endEvents[0].data.reason.kind, 'completed');
for (const request of job.requests) {
const saved = verifyEngineModelInput(f.store, childSession, job, request);
if (strategy) { assertPurgeModelInput(saved.native, job.installation.strategy); assert.equal(job.native.readiness.promptExperimentSha256, job.installation.strategy.sha256); }
if (strategy) { assertPurgeModelInput(saved.native, job.installation.strategy);
if(originalPurgeMode)assert.equal(job.native.originalPurge.strategySha256,job.installation.strategy.sha256);
else assert.equal(job.native.readiness.promptExperimentSha256, job.installation.strategy.sha256); }
else assert.ok(nativeSystemTexts(saved.native).every(text => !text.includes('[Agent Workbench engineering prompt experiment]')));
}
}
@@ -190,7 +203,7 @@ async function runSample(kind, { sampleName = kind, strategy = null, comparison
const failed=childChecks.find(job=>job.status==='failed'),log=f.store.core.artifact.get(failed.results[0].logArtifactId);
assert.ok(f.store.artifactStore.read(log).includes('fail 9'));assert.ok(engineRuns.some(job=>job.calls.some(call=>call.name==='read_artifact')));
assert.ok(engineRuns.some(job=>job.calls.some(call=>call.name==='prepare_change_set')));assert.ok(engineRuns.every(job=>!job.inputRejections?.length));
mark('16K: original nine failures and paged log survive verified native compaction; all three proposals form one reviewed batch without replay');
mark(contextLimit/1000+'K: original nine failures and paged log survive verified native compaction; all three proposals form one reviewed batch without replay');
}
await (await manager.collect(f.session, group.id, group.fingerprint)).promise; assert.equal(group.status, 'collected', group.summary);
const collection = await manager.readCollection(f.session, group); assert.deepEqual(collection.conflicts, []); assert.deepEqual(collection.paths, writable);
@@ -272,9 +285,9 @@ try {
mark(source.version + ' repetition ' + repetition + ': identical original task/baseline/tests/response algorithm and budgets; actual prompt overhead and host approvals recorded');
}
assert.equal(results.length, 8); assert.equal(comparisons.length, 4);
} else for (const kind of contextMode ? ['native-context'] : ['node', 'react', 'python']) results.push(await runSample(kind));
} else for (const kind of contextMode ? ['native-context'] : ['node', 'react', 'python']) results.push(await runSample(kind,{strategy:originalStrategy}));
assert.deepEqual(inputs(), frozen);
report = { success: true, targetVersion: version, testedVersion: version, protocol: responsesMode ? 'responses-stream' : 'chat-completions', directory, installation, cases, results, comparisons, sources, inputs: frozen, realModelRequests: 0,
report = { success: true, targetVersion: version, testedVersion: version, protocol: responsesMode ? 'responses-stream' : 'chat-completions', contextLimit:contextMode?contextLimit:64000, originalPurge:originalPurgeMode, directory, installation, cases, results, comparisons, sources, inputs: frozen, realModelRequests: 0,
scope: 'Pinned actual Harness and workbench execution with an owned HTTP model fixture. Checks, approvals, failures, parent merges, Edge and Git are real. Model reasoning and Purge benefit remain separate.' };
} catch (error) { report = { success: false, targetVersion: version, testedVersion: version, directory, installation, cases, results, inputs: frozen, error: error.stack, realModelRequests: 0 }; process.exitCode = 1; }
finally { fs.writeFileSync(path.join(directory, 'result.json'), JSON.stringify(report, null, 2)); fs.writeFileSync('.qa/' + (contextMode ? (responsesMode ? 'responses-context' : 'controlled-context') : responsesMode ? 'responses-engineering' : desktopMode ? 'desktop-projects' : comparisonMode ? 'controlled-purge' : 'controlled-projects') + '-location.json', JSON.stringify({ directory, result: path.join(directory, 'result.json') })); console.log('EVIDENCE ' + directory); console.log(JSON.stringify({ success: report.success, cases: cases.length, directory, realModelRequests: 0, error: report.error ?? null })); }