@qvac/harness 0.4.0 → 0.5.0

package/dist-lib/spec/hyperschema/index.js

@@ -1,12 +1,12 @@
 // This file is autogenerated by the hyperschema compiler
-// Schema Version: 14
+// Schema Version: 15
 /* eslint-disable camelcase */
 /* eslint-disable quotes */
 /* eslint-disable space-before-function-paren */
 
 import { c } from 'hyperschema/runtime'
 
-const VERSION = 14
+const VERSION = 15
 
 // eslint-disable-next-line no-unused-vars
 let version = VERSION
@@ -1610,17 +1610,19 @@ const encoding31 = {
     c.string.preencode(state, m.type)
     c.string.preencode(state, m.registryPath)
     c.string.preencode(state, m.registrySource)
-    state.end++ // max flag is 8 so always one byte
+    state.end++ // max flag is 16 so always one byte
     c.uint.preencode(state, m.expectedSize)
 
     if (m.actualSize) c.uint.preencode(state, m.actualSize)
+    if (version >= 15 && m.label) c.string.preencode(state, m.label)
   },
   encode(state, m) {
     const flags =
       (m.isCached ? 1 : 0) |
       (m.isLoaded ? 2 : 0) |
       (m.supportsImages ? 4 : 0) |
-      (m.actualSize ? 8 : 0)
+      (m.actualSize ? 8 : 0) |
+      ((version >= 15 && m.label) ? 16 : 0)
 
     c.string.encode(state, m.name)
     c.string.encode(state, m.type)
@@ -1630,6 +1632,7 @@ const encoding31 = {
     c.uint.encode(state, m.expectedSize)
 
     if (m.actualSize) c.uint.encode(state, m.actualSize)
+    if (version >= 15 && m.label) c.string.encode(state, m.label)
   },
   decode(state) {
     const r0 = c.string.decode(state)
@@ -1647,7 +1650,8 @@ const encoding31 = {
       isLoaded: (flags & 2) !== 0,
       supportsImages: (flags & 4) !== 0,
       expectedSize: c.uint.decode(state),
-      actualSize: (flags & 8) !== 0 ? c.uint.decode(state) : 0
+      actualSize: (flags & 8) !== 0 ? c.uint.decode(state) : 0,
+      label: (version >= 15 && (flags & 16) !== 0) ? c.string.decode(state) : null
     }
   }
 }
@@ -2635,17 +2639,21 @@ const encoding65 = {
   preencode(state, m) {
     c.string.preencode(state, m.name)
     c.string.preencode(state, m.src)
-    state.end++ // max flag is 4 so always one byte
+    state.end++ // max flag is 16 so always one byte
 
     if (version >= 5 && m.projectionSrc) c.string.preencode(state, m.projectionSrc)
     if (version >= 14 && m.taskType) c.string.preencode(state, m.taskType)
     if (version >= 14 && m.expectedSize) c.uint.preencode(state, m.expectedSize)
+    if (version >= 15 && m.label) c.string.preencode(state, m.label)
+    if (version >= 15 && m.contextLength) c.uint.preencode(state, m.contextLength)
   },
   encode(state, m) {
     const flags =
       ((version >= 5 && m.projectionSrc) ? 1 : 0) |
       ((version >= 14 && m.taskType) ? 2 : 0) |
-      ((version >= 14 && m.expectedSize) ? 4 : 0)
+      ((version >= 14 && m.expectedSize) ? 4 : 0) |
+      ((version >= 15 && m.label) ? 8 : 0) |
+      ((version >= 15 && m.contextLength) ? 16 : 0)
 
     c.string.encode(state, m.name)
     c.string.encode(state, m.src)
@@ -2654,6 +2662,8 @@ const encoding65 = {
     if (version >= 5 && m.projectionSrc) c.string.encode(state, m.projectionSrc)
     if (version >= 14 && m.taskType) c.string.encode(state, m.taskType)
     if (version >= 14 && m.expectedSize) c.uint.encode(state, m.expectedSize)
+    if (version >= 15 && m.label) c.string.encode(state, m.label)
+    if (version >= 15 && m.contextLength) c.uint.encode(state, m.contextLength)
   },
   decode(state) {
     const r0 = c.string.decode(state)
@@ -2665,7 +2675,9 @@ const encoding65 = {
       src: r1,
       projectionSrc: (version >= 5 && (flags & 1) !== 0) ? c.string.decode(state) : null,
       taskType: (version >= 14 && (flags & 2) !== 0) ? c.string.decode(state) : null,
-      expectedSize: (version >= 14 && (flags & 4) !== 0) ? c.uint.decode(state) : 0
+      expectedSize: (version >= 14 && (flags & 4) !== 0) ? c.uint.decode(state) : 0,
+      label: (version >= 15 && (flags & 8) !== 0) ? c.string.decode(state) : null,
+      contextLength: (version >= 15 && (flags & 16) !== 0) ? c.uint.decode(state) : 0
     }
   }
 }
@@ -3336,6 +3348,23 @@ const encoding87 = {
   }
 }
 
+// @harness/set-model-idle-timeout-request
+const encoding88 = {
+  preencode(state, m) {
+    c.uint.preencode(state, m.ms)
+  },
+  encode(state, m) {
+    c.uint.encode(state, m.ms)
+  },
+  decode(state) {
+    const r0 = c.uint.decode(state)
+
+    return {
+      ms: r0
+    }
+  }
+}
+
 // @harness/completion-frame.grantedDrives, deferred due to recusive use
 const encoding11_23 = c.array(c.frame(encoding60))
 // @harness/skill.setupV2, deferred due to recusive use
@@ -3574,6 +3603,8 @@ function getEncoding(name) {
       return encoding86
     case '@harness/preview-model-source-response':
       return encoding87
+    case '@harness/set-model-idle-timeout-request':
+      return encoding88
     default:
       throw new Error('Encoder not found ' + name)
   }

package/dist-lib/spec/hyperschema/schema.json

@@ -1,5 +1,5 @@
 {
-  "version": 14,
+  "version": 15,
   "schema": [
     {
       "name": "empty",
@@ -1336,6 +1336,11 @@
           "name": "actualSize",
           "type": "uint",
           "version": 1
+        },
+        {
+          "name": "label",
+          "type": "string",
+          "version": 15
         }
       ]
     },
@@ -2139,6 +2144,16 @@
           "name": "expectedSize",
           "type": "uint",
           "version": 14
+        },
+        {
+          "name": "label",
+          "type": "string",
+          "version": 15
+        },
+        {
+          "name": "contextLength",
+          "type": "uint",
+          "version": 15
         }
       ]
     },
@@ -2688,6 +2703,20 @@
           "version": 14
         }
       ]
+    },
+    {
+      "name": "set-model-idle-timeout-request",
+      "namespace": "harness",
+      "compact": false,
+      "flagsPosition": -1,
+      "fields": [
+        {
+          "name": "ms",
+          "required": true,
+          "type": "uint",
+          "version": 15
+        }
+      ]
     }
   ]
 }

package/dist-lib/spec/hyperschema/types.d.ts

@@ -1,5 +1,5 @@
 // This file is autogenerated by hyperschema-ts
-// Schema version: 14
+// Schema version: 15
 // Do not edit manually.
 
 export interface HarnessEmpty {
@@ -256,6 +256,7 @@ export interface HarnessModelInfo {
   supportsImages?: boolean
   expectedSize: number
   actualSize?: number
+  label?: string | null
 }
 
 export interface HarnessListModelsRequest {
@@ -427,6 +428,8 @@ export interface HarnessCustomModel {
   projectionSrc?: string | null
   taskType?: string | null
   expectedSize?: number
+  label?: string | null
+  contextLength?: number
 }
 
 export interface HarnessSetCustomModelsRequest {
@@ -561,6 +564,10 @@ export interface HarnessPreviewModelSourceResponse {
   mmprojSrc?: string | null
 }
 
+export interface HarnessSetModelIdleTimeoutRequest {
+  ms: number
+}
+
 export interface SchemaTypes {
   "@harness/empty": HarnessEmpty
   "@harness/ok-response": HarnessOkResponse
@@ -650,4 +657,5 @@ export interface SchemaTypes {
   "@harness/preview-model-source-request": HarnessPreviewModelSourceRequest
   "@harness/model-source-file": HarnessModelSourceFile
   "@harness/preview-model-source-response": HarnessPreviewModelSourceResponse
+  "@harness/set-model-idle-timeout-request": HarnessSetModelIdleTimeoutRequest
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@qvac/harness",
-  "version": "0.4.0",
+  "version": "0.5.0",
   "type": "module",
   "description": "An AI harness implementing generate(context) via @qvac/sdk. Knows nothing about workbench, hyperdb, or chunks — just inference in, events out.",
   "exports": {
@@ -420,17 +420,17 @@
       "types": "./dist-lib/lib/tools/http-request/encode-fields.d.ts",
       "default": "./dist-lib/lib/tools/http-request/encode-fields.js"
     },
+    "./lib/tools/http-request/request.ts": {
+      "types": "./dist-lib/lib/tools/http-request/request.d.ts",
+      "default": "./dist-lib/lib/tools/http-request/request.js"
+    },
     "./lib/tools/http-request/truncate-response.ts": {
       "types": "./dist-lib/lib/tools/http-request/truncate-response.d.ts",
       "default": "./dist-lib/lib/tools/http-request/truncate-response.js"
     },
-    "./lib/tools/http-request/utils.ts": {
-      "types": "./dist-lib/lib/tools/http-request/utils.d.ts",
-      "default": "./dist-lib/lib/tools/http-request/utils.js"
-    },
-    "./lib/tools/mcp/utils.ts": {
-      "types": "./dist-lib/lib/tools/mcp/utils.d.ts",
-      "default": "./dist-lib/lib/tools/mcp/utils.js"
+    "./lib/tools/mcp/protocol.ts": {
+      "types": "./dist-lib/lib/tools/mcp/protocol.d.ts",
+      "default": "./dist-lib/lib/tools/mcp/protocol.js"
     },
     "./lib/tools/model-args.ts": {
       "types": "./dist-lib/lib/tools/model-args.d.ts",
@@ -600,7 +600,7 @@
   "scripts": {
     "build:schema": "bare schema/build.ts",
     "test": "npm run vendor:wheels && npm run test:bare && npm run test:addons",
-    "test:bare": "QVAC_NO_SKILL_BIN_PROBE=1 prettytap -- brittle-bare test/all.ts",
+    "test:bare": "QVAC_NO_SKILL_BIN_PROBE=1 brittle-jobs 'test/{,!(helpers)/}*.ts'",
     "format": "prettier . --write",
     "lint": "prettier . --check && lunte && tsc --noEmit",
     "build:bundle": "node scripts/build-bundle.mjs",
@@ -699,7 +699,7 @@
     "@qvac/sdk": "0.20.0",
     "@qvac/skills": "0.1.9",
     "asenion": "^0.1.0",
-    "b4a": "^1.8.1",
+    "b4a": "^1.9.0",
     "bare-abort-controller": "^1.1.2",
     "bare-buffer": "^3.6.1",
     "bare-channel": "^5.2.4",
@@ -747,21 +747,21 @@
     "zod": "^4.3.6"
   },
   "devDependencies": {
-    "@types/b4a": "^1.6.5",
     "bare-bundle": "^1.10.0",
     "bare-module-lexer": "^1.6.3",
     "bare-runtime": "^1.30.0",
     "bare-semver": "^1.1.0",
     "brittle": "^4.0.2",
+    "brittle-jobs": "^0.2.1",
     "corestore": "^7.11.0",
     "holepunch-types": "^0.2.1",
     "hyperdrive": "^13.3.3",
     "jsdom": "~30.0.1",
-    "lunte": "^1.8.3",
+    "lunte": "^1.8.4",
+    "lunte-plugin-qvac": "^0.1.0",
     "mermaid": "~11.16.1",
     "prettier": "^3.6.2",
     "prettier-config-holepunch": "^2.0.0",
-    "prettytap": "^0.2.0",
     "terser": "^5.51.2",
     "typescript": "^7.0.2"
   },
@@ -771,7 +771,6 @@
     "localwatch@1.4.3": "patches/localwatch@1.4.3.patch",
     "hypercore-blob-server@1.15.0": "patches/hypercore-blob-server@1.15.0.patch",
     "bare-stow@0.1.5": "patches/bare-stow@0.1.5.patch",
-    "brittle@4.1.0": "patches/brittle@4.1.0.patch",
     "@qvac/inference@0.20.0": "patches/@qvac%2Finference@0.20.0.patch"
   },
   "repository": {

package/worker/harness-sidecar.mjs

@@ -1,10 +1,4 @@
-// bare-stow entry: serves the Harness over the ipc duplex bare-stow wires.
-// argv (empty = absent): [2] skillsDir, [3] SDK storage JSON (required: the fs deny list fences it), [4] embedding config JSON,
-// [5] imageModel JSON, [6] imageClamps JSON,
-// [7] '1' to enable harness logging (written to `harness.log`, beside assistant.log),
-// [8] llm config JSON (llmDevice/llmGpuLayers/contextSize/kvCacheK/kvCacheV/completionModelCap),
-// [9] musicModel JSON, [10] musicClamps JSON, [11] vendored python wheels dir,
-// [12] host-supplied GPU stats JSON.
+// bare-stow entry: argv slots are the ones spawnHarness forwards in lib/spawn.ts; empty is absent
 import '../dist-lib/lib/hardware/gpu-probe-entry.js'
 import FileLog from 'bare-file-logger'
 import fs from 'fs'
@@ -41,10 +35,7 @@ const SCOPE = '[harness]'
 // when the file will not open, everything still has somewhere to go that is not the pipe
 const DISCARD = { info() {}, debug() {}, warn() {}, error() {} }
 
-// On Windows the host does not read our stdout, and writing past the OS buffer takes the
-// Bare thread with it: the event loop stops, nothing raises, and a download never produces
-// a byte. Elsewhere the same write only queues in memory (measured), which is why it only
-// showed up there. Every line goes to a file instead, via the FileLog the assistant uses.
+// on Windows unread stdout past the OS buffer silently halts the Bare thread, so log to a file
 function openHarnessLog(path) {
   try {
     fs.mkdirSync(dirname(path), { recursive: true })
@@ -54,8 +45,7 @@ function openHarnessLog(path) {
   }
 }
 
-// Ungated by the host's log flag: the SDK and its addons log through `console`, and a
-// chatty dependency must not be able to wedge this process.
+// ungated by the log flag: the SDK and addons log via console, and chatter must not wedge us
 function captureConsole(log) {
   console.log = console.info = (...args) => log.info(SCOPE, ...args)
   console.debug = (...args) => log.debug(SCOPE, ...args)
@@ -73,17 +63,16 @@ async function alignSdkConsoleWithHost(hostWantsLogs) {
   }
 }
 
+// lunte-disable-next-line qvac/no-default-export -- the stow entry calls this module's default
 export default async function start(ipc, ready) {
   let harness
   let dying = false
-  // close() cancels in-flight SDK model loads (see Harness._close) so exit is not
-  // parked behind a multi-GB load; a wedged native call still ends at the host's
-  // kill deadline (bare-supervisor/stow reap)
+  // close() cancels in-flight SDK loads so exit is not parked; the host's kill deadline backstops
   const die = () => {
     if (dying) return
     dying = true
     Promise.resolve(harness?.close())
-      .catch(() => {})
+      .catch(noop)
       .then(() => Bare.exit(0))
   }
   Bare.IPC.on('end', die).on('close', die)
@@ -92,9 +81,7 @@ export default async function start(ipc, ready) {
   const deniedRoots = deniedRootsFor(sdkStorage.home)
   const hostWantsLogs = Bare.argv[7] === '1'
   setHarnessLog(hostWantsLogs)
-  // before the first await, so nothing logs ahead of the sink. The SDK home sits in
-  // the storage root, which is where the host writes assistant.log — harness.log
-  // belongs next to it.
+  // before the first await so nothing logs ahead of the sink; the host writes assistant.log beside it
   const log = openHarnessLog(join(dirname(sdkStorage.home), 'harness.log')) ?? DISCARD
   setHarnessLogSink((line) => log.info(SCOPE, line))
   captureConsole(log)
@@ -109,7 +96,9 @@ export default async function start(ipc, ready) {
   const musicClamps = Bare.argv[10] ? JSON.parse(Bare.argv[10]) : undefined
   const gpuStats = Bare.argv[12] ? JSON.parse(Bare.argv[12]) : undefined
   // Only the host knows where the shipped wheels landed.
-  if (Bare.argv[11]) process.env.QVAC_PYTHON_WHEELS_DIR = Bare.argv[11]
+  if (Bare.argv[11]) {
+    process.env.QVAC_PYTHON_WHEELS_DIR = Bare.argv[11]
+  }
 
   configureSdkStorage(sdkStorage)
   process.env.QVAC_PYODIDE_CACHE_DIR = `${sdkStorage.home}/pyodide-cache`
@@ -148,14 +137,16 @@ export default async function start(ipc, ready) {
   if (skillCatalog) {
     tools.push(...compileOperationTools(skillCatalog, { httpRequestTool, mcpCallTool }))
   }
+
   if (skillsDir) {
     tools.push(
       createSkillTool({
         skillsDir,
-        ...(credentialed ? { credentialStore: credentialed.credentialStore } : {})
+        credentialStore: credentialed?.credentialStore
       })
     )
   }
+
   harness = new Harness({
     gpuStats,
     tools,
@@ -180,3 +171,5 @@ export default async function start(ipc, ready) {
 
   return () => {}
 }
+
+function noop() {}

package/worker/stow-entry.mjs

@@ -1,6 +1,5 @@
 /* global Bare */
-// bare-stow's generated bundle shim (lib/shim.js) as a plain file, so the
-// sidecar can be spawned without stowing one.
+// bare-stow's generated shim (lib/shim.js) as a plain file, so the sidecar spawns unstowed
 import '../dist-lib/lib/hardware/gpu-probe-entry.js'
 import protocol from 'bare-stow/protocol'
 
@@ -22,12 +21,16 @@ async function shutdown(code, err = null) {
   exited = true
 
   try {
-    if (stop) await stop()
+    if (stop) {
+      await stop()
+    }
   } catch (e) {
     err = err || e
   }
 
-  if (err) ipc.send('error', { message: err.message, stack: err.stack })
+  if (err) {
+    ipc.send('error', { message: err.message, stack: err.stack })
+  }
 
   await ipc.send('exit', { code })
 

package/dist-lib/lib/tools/mcp/utils.d.ts

@@ -1,47 +0,0 @@
-import type { AbortSignal as BareAbortSignal } from 'bare-abort-controller';
-import type { ToolExecuteContext } from '../../capabilities/completion.ts';
-import { type FetchInit } from '../http-request/utils.ts';
-export declare const MCP_PROTOCOL_VERSION = "2025-03-26";
-export declare const TIMEOUT_MS = 30000;
-export type McpFetchFn = (url: URL, init: FetchInit) => Promise<{
-    status: number;
-    statusText?: string;
-    headers: {
-        get(name: string): string | null;
-    };
-    text(): Promise<string> | string;
-}>;
-export declare function isNotification(method: string): boolean;
-export declare function isMetadataMethod(method: unknown): boolean;
-export declare function normalizeMethod(method: string, params: Record<string, unknown> | undefined): {
-    method: string;
-    params: Record<string, unknown> | undefined;
-};
-export declare function flattenMcpResponse(body: string): {
-    error: string;
-} | {
-    body: string;
-};
-export declare function sessionKey(url: string, ctx: ToolExecuteContext | undefined, credentialKey?: string): string;
-export declare function getSession(key: string): string;
-export declare function deleteSession(key: string): void;
-export declare function cacheSession(key: string, sessionId: string): void;
-export declare function post(fetch: McpFetchFn, url: URL, method: string, params: Record<string, unknown> | undefined, authHeader: string | null, sessionId: string | undefined, signal: BareAbortSignal): Promise<{
-    status: number;
-    statusText?: string;
-    headers: {
-        get(name: string): string | null;
-    };
-    text(): Promise<string> | string;
-}>;
-export declare function readBody(response: {
-    text(): Promise<string> | string;
-    headers: {
-        get(name: string): string | null;
-    };
-}): Promise<string>;
-export declare function needsSession(status: number, cached: string): boolean;
-export declare function establishSession(fetch: McpFetchFn, url: URL, authHeader: string | null, signal: BareAbortSignal): Promise<string | {
-    error: string;
-    stateless?: true;
-}>;