@anionex/dsh-vision-toolkit 0.1.12 → 0.1.14

package/src/runtime-install.ts

@@ -300,6 +300,69 @@ function venvPython(root: string): string {
   return process.platform === 'win32' ? join(root, 'Scripts', 'python.exe') : join(root, 'bin', 'python')
 }
 
+/**
+ * Rewrite a staged venv's `pyvenv.cfg` `home`/`executable` to point at a given
+ * base directory (the app execution alias directory for the Microsoft Store
+ * Python). Pure helper so the transformation is testable cross-platform.
+ */
+export function rewriteVenvConfig(cfg: string, homeDir: string): string {
+  return cfg
+    .replace(/^home = .*$/m, `home = ${homeDir}`)
+    .replace(/^executable = .*$/m, `executable = ${homeDir}\\python.exe`)
+}
+
+/**
+ * Build the Microsoft Store probe environment while preserving Python-variable
+ * tombstones; only the user-directory variables must fall back to the host.
+ */
+export function storePythonProbeEnvironment(installEnv: NodeJS.ProcessEnv): NodeJS.ProcessEnv {
+  return Object.fromEntries(
+    Object.entries(installEnv).filter(([key]) => key !== 'HOME' && key !== 'USERPROFILE' && key !== 'LOCALAPPDATA'),
+  )
+}
+
+/**
+ * Windows-only workaround for the Microsoft Store Python. `python -m venv`
+ * records `home = C:\Program Files\WindowsApps\...` in the new venv, but the
+ * venv launcher (venvlauncher.exe) cannot CreateProcess that `python.exe`
+ * directly — the AppModel package execution restriction denies it (error 5) —
+ * so the venv's pip bootstrap exits 101. Rewrite the staged venv's pyvenv.cfg
+ * to the app execution alias directory, whose `python.exe` is launchable
+ * through Store activation.
+ */
+async function rewriteStorePythonVenvHome(
+  ctx: Context,
+  bootstrap: { command: RuntimeCommand },
+  staging: string,
+  installEnv: NodeJS.ProcessEnv,
+  cwd: string,
+): Promise<void> {
+  if (process.platform !== 'win32') return
+  // The HOME/USERPROFILE/LOCALAPPDATA overrides make the Store alias resolve to
+  // the real Program Files\WindowsApps path; without them sys.executable points
+  // back at the alias directory that venvlauncher can launch.
+  const probeEnv = storePythonProbeEnvironment(installEnv)
+  const probe = await runCollected(
+    ctx,
+    [bootstrap.command.program, ...bootstrap.command.prefix, '-c', 'import os,sys; print(os.path.dirname(sys.executable))'],
+    cwd,
+    { env: probeEnv },
+  )
+  if (probe.exitCode !== 0 || probe.timedOut) return
+  const aliasDir = probe.stdout.trim().split(/\r?\n/)[0]
+  if (aliasDir === undefined || aliasDir.length === 0) return
+  const cfgPath = join(staging, 'pyvenv.cfg')
+  let cfg: string
+  try {
+    cfg = await readFile(cfgPath, 'utf8')
+  } catch {
+    return
+  }
+  // Only touch the known-broken layout; other interpreters are left untouched.
+  if (!/^home = .*Program Files\\WindowsApps.*$/m.test(cfg)) return
+  await writeFile(cfgPath, rewriteVenvConfig(cfg, aliasDir))
+}
+
 async function dependencyVersions(
   ctx: Context,
   python: RuntimeCommand,
@@ -543,13 +606,26 @@ async function prepareManaged(
     if (!created) {
       const create = await runCollected(
         ctx,
-        [bootstrap.command.program, ...bootstrap.command.prefix, '-m', 'venv', staging],
+        [bootstrap.command.program, ...bootstrap.command.prefix, '-m', 'venv', '--without-pip', staging],
         stateRoot,
         { timeoutMs: PREPARE_TIMEOUT_MS, env: installEnv },
       )
       if (create.exitCode !== 0 || create.timedOut) {
         throw new VisionToolkitError('runtime', `Python failed to create the managed runtime: ${create.stderr.trim()}`)
       }
+      // `python -m venv` normally bootstraps pip internally; with the Microsoft
+      // Store Python that step exits 101 (see rewriteStorePythonVenvHome), so pip
+      // is installed explicitly after the venv configuration is repaired.
+      await rewriteStorePythonVenvHome(ctx, bootstrap, staging, installEnv, stateRoot)
+      const pipBootstrap = await runCollected(
+        ctx,
+        [venvPython(staging), '-m', 'ensurepip', '--upgrade', '--default-pip'],
+        stateRoot,
+        { timeoutMs: PREPARE_TIMEOUT_MS, env: installEnv },
+      )
+      if (pipBootstrap.exitCode !== 0 || pipBootstrap.timedOut) {
+        throw new VisionToolkitError('runtime', `Python failed to bootstrap pip in the managed runtime: ${pipBootstrap.stderr.trim()}`)
+      }
       const install = await runCollected(
         ctx,
         [venvPython(staging), '-m', 'pip', 'install', '--disable-pip-version-check', '--no-input', '-r', REQUIREMENTS_PATH],

package/src/runtime.ts

@@ -411,6 +411,7 @@ export interface HtmlScreenshotRequest {
   height?: number
   scale?: number
   waitMs?: number
+  fullPage?: boolean
   output?: string
 }
 
@@ -421,6 +422,8 @@ export interface HtmlScreenshotResult {
   viewport: { width: number; height: number; scale: number }
   width: number
   height: number
+  /** Full document height in CSS pixels; present only for full-page captures. */
+  pageHeight?: number
   artifact: ArtifactDescriptor
 }
 
@@ -1631,6 +1634,7 @@ export class VisionToolkitRuntime {
       const height = integerInRange(request.height, 800, 1, 8192, 'html_screenshot.height')
       const scale = integerInRange(request.scale, 1, 1, 4, 'html_screenshot.scale')
       const waitMs = integerInRange(request.waitMs, 0, 0, 120000, 'html_screenshot.waitMs')
+      const fullPage = request.fullPage === true
       const outputPixels = width * height * scale * scale
       if (!Number.isSafeInteger(outputPixels) || outputPixels > this.config.maxImagePixels) {
         throw new VisionToolkitError('capacity', `HTML screenshot would create ${outputPixels} pixels, exceeding maxImagePixels ${this.config.maxImagePixels}`)
@@ -1657,16 +1661,31 @@ export class VisionToolkitRuntime {
           String(scale),
           '--wait-ms',
           String(waitMs),
+          ...(fullPage ? ['--full-page', '--max-pixels', String(this.config.maxImagePixels)] : []),
         ], operation)
         const parsed = parseHtmlScreenshotOutput(result.stdout)
         const expectedWidth = width * scale
-        const expectedHeight = height * scale
-        if (parsed.outputPath !== staged || parsed.width !== expectedWidth || parsed.height !== expectedHeight) {
+        if (parsed.outputPath !== staged || parsed.width !== expectedWidth) {
+          throw new VisionToolkitError('output', 'html_screenshot: upstream summary does not match the requested output')
+        }
+        if (fullPage !== (parsed.pageHeight !== undefined)) {
+          throw new VisionToolkitError('output', 'html_screenshot: upstream full-page metadata does not match the request')
+        }
+        const pageHeight = parsed.pageHeight
+        const expectedHeight = pageHeight === undefined ? height * scale : pageHeight * scale
+        if (pageHeight !== undefined && (!Number.isInteger(pageHeight) || pageHeight <= 0)) {
+          throw new VisionToolkitError('output', 'html_screenshot: upstream reported an invalid page height')
+        }
+        const outputPixels = expectedWidth * expectedHeight
+        if (!Number.isSafeInteger(outputPixels) || outputPixels > this.config.maxImagePixels) {
+          throw new VisionToolkitError('capacity', `HTML screenshot would create ${outputPixels} pixels, exceeding maxImagePixels ${this.config.maxImagePixels}`)
+        }
+        if (parsed.height !== expectedHeight) {
           throw new VisionToolkitError('output', 'html_screenshot: upstream summary does not match the requested output')
         }
         const generated = await this.probeGeneratedImage(staged, operation, 'html_screenshot')
         if (generated.format !== 'png' || generated.width !== expectedWidth || generated.height !== expectedHeight) {
-          throw new VisionToolkitError('output', 'html_screenshot: generated PNG dimensions do not match the viewport')
+          throw new VisionToolkitError('output', 'html_screenshot: generated PNG dimensions do not match the reported output')
         }
         await commitStagedOutput(staged, finalPath, policy)
         const artifact = await describeArtifact(finalPath, policy, {
@@ -1682,6 +1701,7 @@ export class VisionToolkitRuntime {
           viewport: { width, height, scale },
           width: expectedWidth,
           height: expectedHeight,
+          ...(pageHeight === undefined ? {} : { pageHeight }),
           artifact,
         }
       } finally {

package/src/skill.ts

@@ -45,7 +45,7 @@ never enter the model tool set, before or after Skill activation.
 | Split and merge a tall screenshot OCR | vision_long_screenshot_ocr |
 | Extract an icon/logo on transparency | vision_extract_foreground |
 | Measure a palette or choose among exact colors | vision_dominant_colors |
-| Render a local HTML implementation to PNG | vision_html_screenshot |
+| Render a local HTML implementation to PNG (viewport or full document) | vision_html_screenshot |
 
 vision_glance, vision_ground, vision_detect, and non-split long OCR send the
 validated image bytes to the external vision service configured by the user.
@@ -86,7 +86,7 @@ be opened or downloaded by the UI and passed to later tools.
 - vision_long_screenshot_ocr -> merged Markdown, manifest JSON, boundary audit,
   chunk PNGs, and OCR sidecars
 - vision_extract_foreground -> transparent PNG
-- vision_html_screenshot -> PNG
+- vision_html_screenshot -> PNG (fullPage=true also reports the CSS pageHeight)
 
 ## Reliable workflows
 
@@ -109,7 +109,8 @@ request. Reuse the same runName with resume=true to retain matching sidecars.
 2. vision_dominant_colors for measured colors.
 3. vision_extract_foreground or vision_trace for reusable assets.
 4. Implement a local HTML file.
-5. vision_html_screenshot with the target viewport.
+5. vision_html_screenshot with the target viewport; use fullPage=true when the
+   complete document is needed without guessing taller viewports.
 6. vision_pixel_diff against the reference.
 7. Inspect the worst region and iterate until the measured differences are
    acceptable.
@@ -127,6 +128,9 @@ vision_glance with ocr=true when the text content matters.
   invent nested or absolute output paths.
 - vision_html_screenshot accepts local .html/.htm files only, not URLs or data
   URIs.
+- vision_html_screenshot keeps the requested viewport for layout; with
+  fullPage=true, it captures the complete document and returns pageHeight
+  in CSS pixels.
 - Disabling or unloading the plugin cancels its active visual operations before
   unregistering the native tools and skill.
 - Do not infer image contents after a tool error. Fix the path, limits,

package/src/tools.ts

@@ -540,7 +540,7 @@ export function createVisionTools(
         + WORKSPACE_NOTE,
       parameters: {
         source: { type: 'string', required: true, description: 'Local HTML path only.' }, width: { type: 'integer' }, height: { type: 'integer' },
-        scale: { type: 'integer' }, waitMs: { type: 'integer' }, output: { type: 'string', description: 'Artifact filename; .png only.' },
+        scale: { type: 'integer' }, waitMs: { type: 'integer' }, fullPage: { type: 'boolean', description: 'Capture the full document height while preserving the requested viewport.' }, output: { type: 'string', description: 'Artifact filename; .png only.' },
         timeoutMs: { type: 'integer', description: TIMEOUT_NOTE },
       },
       output: {
@@ -548,7 +548,7 @@ export function createVisionTools(
           type: 'object', additionalProperties: false, properties: {
             sourcePath: { type: 'string', required: true }, sourceBytes: { type: 'integer', required: true },
             viewport: { type: 'object', additionalProperties: false, required: true, properties: { width: { type: 'integer', required: true }, height: { type: 'integer', required: true }, scale: { type: 'integer', required: true } } },
-            width: { type: 'integer', required: true }, height: { type: 'integer', required: true }, artifact: requiredArtifactSchema,
+            width: { type: 'integer', required: true }, height: { type: 'integer', required: true }, pageHeight: { type: 'integer' }, artifact: requiredArtifactSchema,
           },
         },
         render: renderJson,
@@ -559,6 +559,7 @@ export function createVisionTools(
           source: args.source,
           ...(args.width === undefined ? {} : { width: args.width }), ...(args.height === undefined ? {} : { height: args.height }),
           ...(args.scale === undefined ? {} : { scale: args.scale }), ...(args.waitMs === undefined ? {} : { waitMs: args.waitMs }),
+          ...(args.fullPage === undefined ? {} : { fullPage: args.fullPage }),
           ...(args.output === undefined ? {} : { output: args.output }),
         }
         return runtimeFrom(source).htmlScreenshot(request, callOptions(exec, args.timeoutMs, lifecycleSignal))
@@ -663,6 +664,7 @@ interface HtmlArgs {
   height?: number
   scale?: number
   waitMs?: number
+  fullPage?: boolean
   output?: string
   timeoutMs?: number
 }

package/src/upstream.ts

@@ -165,6 +165,7 @@ export interface HtmlScreenshotOutput {
   outputPath: string
   width: number
   height: number
+  pageHeight?: number
 }
 
 const BOX_SUFFIX = /x1:\s*(\d+),\s*y1:\s*(\d+),\s*x2:\s*(\d+),\s*y2:\s*(\d+)\s*$/
@@ -448,9 +449,14 @@ export function parseDominantColorsOutput(stdout: string): DominantColorsOutput
 
 /** Parse the local Chrome screenshot summary. */
 export function parseHtmlScreenshotOutput(stdout: string): HtmlScreenshotOutput {
-  const wrote = /^wrote\s+(.+?)\s+\((\d+)x(\d+)\)\s*$/.exec(stdout.trim())
+  const wrote = /^wrote\s+(.+?)\s+\((\d+)x(\d+)(?:;\s*pageHeight=(\d+))?\)\s*$/.exec(stdout.trim())
   if (wrote === null) throw new VisionToolkitError('output', 'html_screenshot: upstream did not report a written PNG')
-  return { outputPath: wrote[1] ?? '', width: Number(wrote[2]), height: Number(wrote[3]) }
+  return {
+    outputPath: wrote[1] ?? '',
+    width: Number(wrote[2]),
+    height: Number(wrote[3]),
+    ...(wrote[4] === undefined ? {} : { pageHeight: Number(wrote[4]) }),
+  }
 }
 
 const REQUIRED_TOOLS = ['glance', 'ground', 'detect', 'crop', 'trace'] as const
@@ -527,20 +533,24 @@ const VISION_MODEL_GUARD = [
 ].join('\n')
 
 const HTML_SCREENSHOT_GUARD = [
-  'import runpy,subprocess,sys,tempfile',
+  'import os,runpy,subprocess,sys,tempfile',
   'script=sys.argv[1]',
   'sys.argv=[script,*sys.argv[2:]]',
-  'original_run=subprocess.run',
+  'original_popen=subprocess.Popen',
   'with tempfile.TemporaryDirectory(prefix="dsh-vision-chrome-") as profile:',
-  '    def guarded_run(command,*args,**kwargs):',
+  '    original_profile=os.environ.get("DSH_VISION_CHROME_PROFILE")',
+  '    os.environ["DSH_VISION_CHROME_PROFILE"]=profile',
+  '    def guarded_popen(command,*args,**kwargs):',
   '        command=list(command)',
   '        command[1:1]=["--use-mock-keychain",f"--user-data-dir={profile}","--incognito","--disable-background-networking","--proxy-server=http://127.0.0.1:9","--proxy-bypass-list=<-loopback>"]',
-  '        return original_run(command,*args,**kwargs)',
-  '    subprocess.run=guarded_run',
+  '        return original_popen(command,*args,**kwargs)',
+  '    subprocess.Popen=guarded_popen',
   '    try:',
   '        runpy.run_path(script,run_name="__main__")',
   '    finally:',
-  '        subprocess.run=original_run',
+  '        subprocess.Popen=original_popen',
+  '        if original_profile is None: os.environ.pop("DSH_VISION_CHROME_PROFILE",None)',
+  '        else: os.environ["DSH_VISION_CHROME_PROFILE"]=original_profile',
 ].join('\n')
 
 const LONG_OCR_PINNED_GLANCE = [
@@ -853,6 +863,9 @@ export class UpstreamAdapter {
     if (/Missing config VISION_/i.test(result.stderr)) {
       return new VisionToolkitError('config', message)
     }
+    if (/maxImagePixels|exceed(?:s|ing).*pixels/i.test(result.stderr)) {
+      return new VisionToolkitError('capacity', message)
+    }
     if (/not found|only PNG|unsupported|cannot open|empty region|must be|expects|invalid colour|needs at least/i.test(result.stderr)) {
       return new VisionToolkitError('input', message)
     }

package/src/web.ts

@@ -28,6 +28,13 @@ import {
   type VisionToolkitConfig,
 } from './config.ts'
 import type { VisionToolkitHealthResult } from './runtime.ts'
+import {
+  PluginUpdateError,
+  VisionToolkitPluginUpdateService,
+  type PluginUpdateCapability,
+  type PluginUpdateCheck,
+  type PluginUpdateResult,
+} from './plugin-update.ts'
 import {
   VisionToolkitRuntimeManager,
   type PreparedRuntimeGeneration,
@@ -62,6 +69,7 @@ export interface VisionToolkitSettingsSnapshot {
     upstreamRepository: string
     upstreamVersion: string
     upstreamCommit: string
+    update: PluginUpdateCapability
   }
   artifactRouteAvailable: boolean
 }
@@ -85,7 +93,16 @@ interface CredentialRequest {
   value: string
 }
 
-type SettingsRequest = SaveRequest | HealthRequest | CredentialRequest
+interface CheckUpdateRequest {
+  action: 'check-update'
+}
+
+interface ApplyUpdateRequest {
+  action: 'apply-update'
+  expectedVersion: string
+}
+
+type SettingsRequest = SaveRequest | HealthRequest | CredentialRequest | CheckUpdateRequest | ApplyUpdateRequest
 
 interface JsonError {
   ok: false
@@ -109,6 +126,14 @@ export interface WebRuntimeManager {
   status(): RuntimeManagerStatus
 }
 
+/** Minimal self-update face used by the Web route and its tests. */
+export interface WebPluginUpdater {
+  configureWebServer?(host: string, port: number): void
+  capability(): Promise<PluginUpdateCapability>
+  check(): Promise<PluginUpdateCheck>
+  installAndRestart(expectedVersion: string): Promise<PluginUpdateResult>
+}
+
 /** Callback invoked when a Settings save makes the first runtime available. */
 export type RuntimeActivated = () => void
 
@@ -195,6 +220,13 @@ function parseRequest(value: unknown): SettingsRequest {
       value: secret,
     }
   }
+  if (value.action === 'check-update') return { action: 'check-update' }
+  if (value.action === 'apply-update') {
+    if (typeof value.expectedVersion !== 'string' || value.expectedVersion.trim().length === 0) {
+      throw new TypeError('apply-update.expectedVersion must be a non-empty string')
+    }
+    return { action: 'apply-update', expectedVersion: value.expectedVersion.trim() }
+  }
   throw new TypeError(`unsupported action: ${value.action}`)
 }
 
@@ -205,12 +237,24 @@ function publicMessage(error: unknown): string {
 
 /** Same-origin Settings and health handler. */
 export class VisionToolkitWebBackend {
+  private readonly updater: WebPluginUpdater
+
   constructor(
     private readonly ctx: Context,
     private readonly manager: WebRuntimeManager,
     private readonly artifacts: ArtifactAccessController,
     private readonly onRuntimeActivated: RuntimeActivated,
-  ) {}
+    updater?: WebPluginUpdater,
+  ) {
+    this.updater = updater ?? new VisionToolkitPluginUpdateService(ctx, PLUGIN_VERSION, {
+      runtimeReady: () => this.manager.status().ready,
+    })
+  }
+
+  /** Supply the active listener address before the Settings route becomes reachable. */
+  configureWebServer(host: string, port: number): void {
+    this.updater.configureWebServer?.(host, port)
+  }
 
   private async credential(config: ResolvedVisionToolkitConfig): Promise<CredentialInfo> {
     if (isBuiltInFreeVisionProvider(config.provider)) {
@@ -225,6 +269,7 @@ export class VisionToolkitWebBackend {
     const value = descriptor.value as VisionToolkitConfig
     const resolved = resolveConfig(value)
     const credential = await this.credential(resolved)
+    const update = await this.updater.capability()
     return {
       schemaVersion: 1,
       writable: this.ctx.settings.writable,
@@ -247,6 +292,7 @@ export class VisionToolkitWebBackend {
         upstreamRepository: UPSTREAM_REPOSITORY,
         upstreamVersion: UPSTREAM_VERSION,
         upstreamCommit: UPSTREAM_COMMIT,
+        update,
       },
       artifactRouteAvailable: this.artifacts.routeAvailable,
     }
@@ -350,20 +396,39 @@ export class VisionToolkitWebBackend {
         case 'credential':
           responseJson(res, 200, { ok: true, value: await this.saveCredential(parsed) })
           break
+        case 'check-update':
+          responseJson(res, 200, { ok: true, value: await this.updater.check() })
+          break
+        case 'apply-update':
+          responseJson(res, 200, { ok: true, value: await this.updater.installAndRestart(parsed.expectedVersion) })
+          break
       }
     } catch (error) {
       const settingsConflict = error instanceof SettingsConflictError
       const credentialConflict = error instanceof CredentialReferenceConflictError
+      const updateError = error instanceof PluginUpdateError
       const code = settingsConflict
         ? 'settings-conflict'
         : credentialConflict
           ? 'credential-conflict'
+          : updateError
+            ? error.code
           : parsed.action === 'health'
             ? 'health-failed'
             : parsed.action === 'credential'
               ? 'credential-rejected'
               : 'settings-rejected'
-      const status = settingsConflict || credentialConflict ? 409 : parsed.action === 'health' ? 503 : 400
+      const updateConflict = updateError && ['update-in-progress', 'update-stale', 'update-unavailable', 'already-current'].includes(error.code)
+      const updateGateway = updateError && error.code === 'update-check-failed'
+      const status = settingsConflict || credentialConflict || updateConflict
+        ? 409
+        : parsed.action === 'health'
+          ? 503
+          : updateGateway
+            ? 502
+            : updateError
+              ? 500
+              : 400
       this.ctx.logger.warn('dsh-vision-toolkit Web action=%s failed: %s', parsed.action, publicMessage(error))
       requestError(res, status, code, publicMessage(error))
     }
@@ -454,6 +519,7 @@ export function installVisionToolkitWeb(
 ): void {
   ctx.inject(['webServer'], (webCtx) => {
     webCtx.effect(() => {
+      backend.configureWebServer(webCtx.webServer.host, webCtx.webServer.port)
       const detach = artifacts.attachRoute()
       const disposeArtifact = webCtx.webServer.register({
         kind: 'prefix',

package/vendor/agent-vision-toolkit/UPSTREAM_MANIFEST.json

@@ -3,7 +3,7 @@
   "repository": "https://github.com/Anionex/agent-vision-toolkit",
   "version": "v0.1.0+snapshot.bc9803d",
   "commit": "bc9803d7d6300c864d17460ecbb33540b26638e0",
-  "contentSha256": "6936f8fba3ed40f54267245e717603c2622649ce4bd7ccf3c1b23f39047b1f3f",
+  "contentSha256": "aee26acedf76b083addc31acfc8817eb2a2044d144b495f4437adc9e20d18750",
   "files": [
     {
       "path": "CHANGELOG.md",
@@ -67,8 +67,8 @@
     },
     {
       "path": "skills/vision-tools/scripts/html_shot.py",
-      "bytes": 4317,
-      "sha256": "22f796dc6afaa791bbd9acc9bf87bc4914c4157e726f0132f7465f0e3d4d2894"
+      "bytes": 17573,
+      "sha256": "242e62e5d8bc02c4a707c57fe782cefb7f55abc0652792c1881a3f5d36c9302f"
     },
     {
       "path": "skills/vision-tools/scripts/long_screenshot_ocr.py",