@hmharness/agent 0.10.0 → 0.11.0

package/dist/pipeline.d.ts

@@ -1,5 +1,6 @@
 import { runLoop, type LoopResult, type ProviderConfig, type Registry, type ToolContext } from '@hmharness/kernel';
-export type PipelineStage = 'plan' | 'code' | 'test' | 'review' | 'judge';
+import type { DeviceTestOptions, DeviceTestStep } from '@hmharness/domain-harmony';
+export type PipelineStage = 'plan' | 'code' | 'test' | 'review' | 'judge' | 'device';
 /** stage labels used in runStage (repairer = the repair-loop role) */
 export type StageRole = PipelineStage | 'repairer';
 export interface StageRecord {
@@ -38,6 +39,20 @@ export interface PipelineOptions {
     signal?: AbortSignal;
     /** injectable loop (tests); default kernel runLoop */
     runLoopImpl?: typeof runLoop;
+    /** V3 device gate (ADR-0007): when set, run an on-device install/launch/
+     *  log-marker/uninstall pass after the test stage and feed the four steps
+     *  to the judge as mechanical evidence. Pure command execution - no model
+     *  turns. */
+    deviceGate?: {
+        hdc: string;
+        target?: string;
+        hap: string;
+        bundle: string;
+        ability: string;
+        expectLog: string;
+        /** injectable runner (tests); default = domain-harmony runDeviceTest */
+        runDeviceTestImpl?: (o: DeviceTestOptions) => Promise<DeviceTestStep[]>;
+    };
 }
 /** Pull the judge's verdict line out of the final text; absence = FAIL
  *  (an judge that did not follow the contract did not render a verdict). */

package/dist/pipeline.js

@@ -107,6 +107,35 @@ export async function runPipeline(opts) {
         catch { /* best-effort persistence */ }
         return rec;
     };
+    /** Device gate: run the four-step on-device pass, record it as a 'device'
+     *  stage (no model turns spent), and mirror it into the repair context.
+     *  Defined BEFORE the try block - the try body calls it (TDZ otherwise). */
+    const runDeviceGate = async () => {
+        const g = opts.deviceGate;
+        const runner = g.runDeviceTestImpl ?? (await import('@hmharness/domain-harmony')).runDeviceTest;
+        let steps;
+        try {
+            steps = await runner({ hdc: g.hdc, ...(g.target ? { target: g.target } : {}), hap: g.hap, bundle: g.bundle, ability: g.ability, expectLog: g.expectLog });
+        }
+        catch (err) {
+            steps = [{ step: 'device-gate', pass: false, detail: String(err).slice(0, 300) }];
+        }
+        const allPass = steps.every((s) => s.pass);
+        const rec = {
+            stage: 'device',
+            attempt: repairs + 1,
+            verdict: allPass ? 'PASS' : 'FAIL',
+            text: steps.map((s) => `${s.pass ? 'PASS' : 'FAIL'} ${s.step}: ${s.detail}`).join('\n').slice(0, 4000),
+            turns: 0,
+            toolUses: 0,
+            reason: 'final',
+        };
+        stages.push(rec);
+        try {
+            await writeFile(join(dir, `stage-${String(stages.length).padStart(2, '0')}-device.json`), JSON.stringify(rec, null, 2) + '\n', 'utf8');
+        }
+        catch { /* best-effort */ }
+    };
     try {
         // 1. plan
         const plan = await runStage('plan', 1, DIRECTIVES.plan(opts.task, ''));
@@ -126,6 +155,10 @@ export async function runPipeline(opts) {
             status = 'budget';
             return await finish();
         }
+        // 3b. device gate (V3 slice, ADR-0007): mechanical on-device evidence,
+        // zero model turns; judge sees it in the evidence summary
+        if (opts.deviceGate)
+            await runDeviceGate();
         let review = await runStage('review', 1, DIRECTIVES.review(opts.task, ''));
         if (spent >= totalBudget) {
             status = 'budget';
@@ -141,6 +174,8 @@ export async function runPipeline(opts) {
                 break;
             }
             testOut = await runStage('test', repairs + 1, DIRECTIVES.test(opts.task, `Plan:\n${plan.text.slice(0, 1500)}\nRepair ${repairs} applied - re-verify.`));
+            if (opts.deviceGate)
+                await runDeviceGate();
             review = await runStage('review', repairs + 1, DIRECTIVES.review(opts.task, `Repair ${repairs} was applied; focus on it.`));
             if (spent >= totalBudget) {
                 status = 'budget';
@@ -174,6 +209,6 @@ export async function runPipeline(opts) {
     }
 }
 function evidenceSummary(stages) {
-    const relevant = stages.filter((s) => s.stage === 'test' || s.stage === 'review');
+    const relevant = stages.filter((s) => s.stage === 'test' || s.stage === 'review' || s.stage === 'device');
     return relevant.map((s) => `[${s.stage} #${s.attempt}] ${s.text.slice(0, 600)}`).join('\n\n').slice(0, 3000);
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@hmharness/agent",
-  "version": "0.10.0",
+  "version": "0.11.0",
   "description": "hmharness agent execution layer: base tools, system prompt, sub-agent spawn, and the shared task runner that frontends (cli, web) drive.",
   "type": "module",
   "main": "dist/index.js",
@@ -15,7 +15,7 @@
     "build": "tsc -p tsconfig.build.json"
   },
   "dependencies": {
-    "@hmharness/domain-harmony": "0.8.0",
+    "@hmharness/domain-harmony": "0.11.0",
     "@hmharness/domain-ops": "0.8.0",
     "@hmharness/evolution": "0.9.0",
     "@hmharness/kernel": "0.9.0",