dsh-vision-router 2.1.2 → 2.1.3

package/lib/structured-flow-hardening.js

@@ -20,9 +20,7 @@ const MAX_TURN_BUDGET_MS = 600_000
 const MAX_GUIDANCE_OVERRIDES = 14
 const MAX_GUIDANCE_CHARS = 2_000
 const MAX_NON_PROGRESS_ATTEMPTS = 3
-const MAX_EVIDENCE_SIGNATURE_CHARS = 4_096
 const BUDGETED_TIMEOUT_FIELDS = ['timeoutMs', 'visionTaskTimeoutMs', 'ocrTimeoutMs']
-const EVIDENCE_META_KEYS = new Set(['ok', 'code', 'retryable', 'reason', 'phase', 'next'])
 const GUIDANCE_KINDS = new Set([
   'code',
   'document',
@@ -40,25 +38,6 @@ const GUIDANCE_KINDS = new Set([
   'meme',
 ])
 
-const BUILTIN_MIXED_GUIDANCE = {
-  zh: {
-    document: '语义优先；只有逐字引用、合同/表单字段或表格数字确需保真时才使用 OCR。',
-    ui: '优先用 vision_detect / vision_ground 验证元素、状态与位置。',
-    code: '逐字核对代码；对可执行字符、缩进和符号做交叉验证。',
-    chat: '按气泡顺序核对发言者、消息内容与时间/状态信息。',
-    general: '聚焦尚未归类的另一部分可见内容，自行选择最能新增证据的视觉工具。',
-    fallback: '聚焦这个分支新增或验证可见证据。',
-  },
-  en: {
-    document: 'Prioritize semantics; use OCR only when verbatim quotes, contract/form fields, or table numbers require exact transcription.',
-    ui: 'Prefer vision_detect / vision_ground to verify elements, states, and positions.',
-    code: 'Verify code verbatim; cross-check executable characters, indentation, and symbols.',
-    chat: 'Verify speakers, message text, and time/status information in bubble order.',
-    general: 'Focus on the remaining unclassified visible content and choose the vision tool that adds the most useful evidence.',
-    fallback: 'Add or verify visible evidence for this branch.',
-  },
-}
-
 function isObject(value) {
   return value !== null && typeof value === 'object'
 }
@@ -98,36 +77,143 @@ export function producedStructuredEvidence(value) {
   return true
 }
 
-function meaningfulEvidenceValue(value) {
-  if (value === undefined || value === null) return false
-  if (typeof value === 'string') return value.trim() !== ''
-  if (typeof value === 'number') return Number.isFinite(value)
-  if (typeof value === 'boolean') return true
-  if (Array.isArray(value)) return value.length > 0 && value.some(meaningfulEvidenceValue)
-  if (!isObject(value) || value.ok === false) return false
-  return Object.entries(value).some(([key, child]) => (
-    !EVIDENCE_META_KEYS.has(key) && meaningfulEvidenceValue(child)
+function finiteNumber(value) {
+  return typeof value === 'number' && Number.isFinite(value)
+}
+
+function positiveInteger(value) {
+  return Number.isInteger(value) && value > 0
+}
+
+function nonEmptyText(value) {
+  return typeof value === 'string' && value.trim() !== ''
+}
+
+function canonicalBox(value, width, height) {
+  if (!isObject(value) || Array.isArray(value)) return false
+  const { x1, y1, x2, y2 } = value
+  if (![x1, y1, x2, y2].every(Number.isInteger)) return false
+  if (x1 < 0 || y1 < 0 || x2 <= x1 || y2 <= y1) return false
+  if (positiveInteger(width) && x2 > width) return false
+  if (positiveInteger(height) && y2 > height) return false
+  return true
+}
+
+function structuredDescribeEvidence(value) {
+  if (!isObject(value) || Array.isArray(value)) return false
+  if (nonEmptyText(value.summary) || nonEmptyText(value.text)) return true
+  if (Array.isArray(value.layout) && value.layout.some((row) => (
+    isObject(row) && nonEmptyText(row.region) && nonEmptyText(row.content)
+  ))) return true
+  return Array.isArray(value.entities) && value.entities.some((entity) => (
+    isObject(entity) && nonEmptyText(entity.type) && nonEmptyText(entity.label)
+  ))
+}
+
+function groundEvidence(value) {
+  if (!isObject(value) || Array.isArray(value)) return false
+  const { width, height } = value
+  return positiveInteger(width) && positiveInteger(height) && canonicalBox(value, width, height)
+}
+
+function detectEvidence(value) {
+  if (!isObject(value) || Array.isArray(value)) return false
+  const { width, height, elements } = value
+  if (!positiveInteger(width) || !positiveInteger(height) || !Array.isArray(elements)) return false
+  return elements.every((element, index) => (
+    isObject(element) &&
+    element.number === index + 1 &&
+    nonEmptyText(element.label) &&
+    canonicalBox(element.box, width, height)
   ))
 }
 
+function ocrEvidence(value) {
+  if (!isObject(value) || Array.isArray(value)) return false
+  if (value.engine !== 'tesseract' && value.engine !== 'vision') return false
+  return typeof value.text === 'string'
+}
+
+function colorEvidence(value) {
+  if (!Array.isArray(value)) return false
+  return value.every((entry) => (
+    isObject(entry) &&
+    typeof entry.hex === 'string' && /^#[0-9a-f]{6}$/i.test(entry.hex) &&
+    positiveInteger(entry.count) &&
+    finiteNumber(entry.share) && entry.share > 0 && entry.share <= 1
+  ))
+}
+
+function pixelDiffEvidence(value) {
+  if (!isObject(value) || Array.isArray(value)) return false
+  const { width, height, differingPixels, totalPixels, diffRatio, worstRegions } = value
+  if (!positiveInteger(width) || !positiveInteger(height) || totalPixels !== width * height) return false
+  if (!Number.isInteger(differingPixels) || differingPixels < 0 || differingPixels > totalPixels) return false
+  if (!finiteNumber(diffRatio) || diffRatio < 0 || diffRatio > 1) return false
+  if (Math.abs(diffRatio - differingPixels / totalPixels) > 0.000051) return false
+  if (!Array.isArray(worstRegions)) return false
+  return worstRegions.every((region) => (
+    isObject(region) &&
+    canonicalBox(region, width, height) &&
+    Number.isInteger(region.differing) && region.differing > 0 &&
+    positiveInteger(region.total) && region.differing <= region.total &&
+    finiteNumber(region.ratio) && region.ratio > 0 && region.ratio <= 1
+  ))
+}
+
+function longOcrEvidence(value) {
+  if (!isObject(value) || Array.isArray(value)) return false
+  if (typeof value.text !== 'string' || !positiveInteger(value.chunks)) return false
+  if (!isObject(value.engines) || Array.isArray(value.engines)) return false
+  if (![value.markdownPath, value.manifestPath, value.artifactsDir].every(nonEmptyText)) return false
+  const engineCounts = Object.values(value.engines)
+  if (engineCounts.length === 0 || !engineCounts.every((count) => Number.isInteger(count) && count >= 0)) return false
+  if (engineCounts.reduce((sum, count) => sum + count, 0) !== value.chunks) return false
+  if (nonEmptyText(value.text)) return true
+  const completed = (value.engines.tesseract ?? 0) + (value.engines.vision ?? 0)
+  return completed > 0
+}
+
 /**
- * Evidence that is strong enough to advance the 1+x flow and consume an
- * explicit deep-dive call cap. Empty arrays/objects and metadata-only success
- * envelopes do not count. Negative facts such as { match: false } or
- * { count: 0 } still count because they are real observations, not emptiness.
+ * Decide whether one registered evidence tool completed a real observation.
+ *
+ * This deliberately follows each tool's output contract instead of treating
+ * arbitrary non-empty containers as evidence. Valid negative observations
+ * (for example zero detections, empty OCR text, or a 0 pixel diff) still count;
+ * transport failures, malformed envelopes, and describe's empty-content
+ * sentinel do not.
  */
-export function producedUsableStructuredEvidence(value) {
-  if (value === undefined || value === null) return false
-  if (typeof value === 'string') {
-    const text = value.trim()
-    if (text === '') return false
-    const parsed = parseJson(text)
-    if (parsed === undefined) return true
-    return producedUsableStructuredEvidence(parsed)
+export function producedUsableToolEvidence(toolName, value, args = {}) {
+  if (!STRUCTURED_EVIDENCE_TOOLS.has(toolName) || value === undefined || value === null) return false
+  const parsed = parseJson(value)
+  if (isObject(parsed) && !Array.isArray(parsed) && parsed.ok === false) return false
+
+  switch (toolName) {
+    case 'vision_describe': {
+      if (typeof value !== 'string') return false
+      const text = value.trim()
+      if (text === '' || text === '(the vision model returned empty content)') return false
+      if (args?.json === true) {
+        if (text.startsWith('vision_describe: the model did not produce valid JSON.')) return false
+        return structuredDescribeEvidence(parsed)
+      }
+      return true
+    }
+    case 'vision_ground':
+      return groundEvidence(parsed)
+    case 'vision_detect':
+      return detectEvidence(parsed)
+    case 'vision_ocr':
+      return ocrEvidence(parsed)
+    case 'vision_colors':
+      return colorEvidence(parsed)
+    case 'vision_pixel_diff':
+      return pixelDiffEvidence(parsed)
+    case 'vision_long_screenshot_ocr':
+      return longOcrEvidence(parsed)
+    default:
+      return false
   }
-  if (Array.isArray(value)) return value.length > 0 && value.some(meaningfulEvidenceValue)
-  if (!isObject(value) || value.ok === false) return false
-  return meaningfulEvidenceValue(value)
 }
 
 /**
@@ -176,15 +262,6 @@ function languageOf(config) {
   return runtimeLanguageFor(config?.__visionRouterLocale, 'zh')
 }
 
-function guidanceFor(kind, overrides, language = 'zh') {
-  for (let i = overrides.length - 1; i >= 0; i--) {
-    const entry = overrides[i]
-    if (entry.kind === kind) return entry.text
-  }
-  const copy = BUILTIN_MIXED_GUIDANCE[language] ?? BUILTIN_MIXED_GUIDANCE.zh
-  return copy[kind] ?? copy.fallback
-}
-
 function turnBudgetMs(config) {
   const value = Number(config?.visionTurnBudgetMs)
   if (!Number.isFinite(value) || value <= 0) return DEFAULT_TURN_BUDGET_MS
@@ -234,9 +311,7 @@ function freshState(turn, _config) {
     bootstrapDone: false,
     bootstrapResult: undefined,
     successfulEvidenceCalls: 0,
-    mixedBranches: [],
-    completedBranches: new Set(),
-    mixedAttemptSignatures: new Set(),
+    postBootstrapEvidenceCalls: 0,
     consecutiveNonProgressAttempts: 0,
     budgetExhausted: false,
     depthExhausted: false,
@@ -284,56 +359,13 @@ function bootstrapEvidence(raw) {
   return parsed.evidence
 }
 
-function normalizeMixedBranches(evidence, _depth) {
-  if (!isObject(evidence) || evidence.visual_kind !== 'mixed' || !Array.isArray(evidence.mixed_of)) return []
-  const allowed = ['ui', 'document', 'code', 'chat', 'general']
-  const unique = []
-  for (const kind of allowed) {
-    if (evidence.mixed_of.includes(kind)) unique.push(kind)
-  }
-  return unique.slice(0, 2)
-}
-
-function evidenceAttemptSignature(name, args) {
-  let encoded
-  try {
-    encoded = JSON.stringify(args ?? null)
-  } catch {
-    encoded = String(args ?? '')
-  }
-  if (encoded.length <= MAX_EVIDENCE_SIGNATURE_CHARS) return `${name}:${encoded}`
-  const half = Math.floor(MAX_EVIDENCE_SIGNATURE_CHARS / 2)
-  return `${name}:${encoded.length}:${encoded.slice(0, half)}:${encoded.slice(-half)}`
-}
-
-function inferBranchForTool(name, pending) {
-  const prefer = (kinds) => kinds.find((kind) => pending.includes(kind))
-  if (name === 'vision_detect' || name === 'vision_ground') return prefer(['ui', 'general'])
-  if (name === 'vision_long_screenshot_ocr') return prefer(['document', 'chat', 'code', 'general'])
-  if (name === 'vision_ocr') return prefer(['document', 'code', 'chat', 'general'])
-  if (name === 'vision_colors' || name === 'vision_pixel_diff') return prefer(['general', 'ui'])
-  return pending.length === 1 ? pending[0] : undefined
-}
-
-function recordEvidenceSuccess(state, name, args) {
-  state.successfulEvidenceCalls += 1
-  if (state.mixedBranches.length === 0) return true
-  const pending = state.mixedBranches.filter((kind) => !state.completedBranches.has(kind))
-  if (pending.length === 0) return true
-  const signature = evidenceAttemptSignature(name, args)
-  if (state.mixedAttemptSignatures.has(signature)) return false
-  state.mixedAttemptSignatures.add(signature)
-  const branch = inferBranchForTool(name, pending)
-  if (!branch) return false
-  state.completedBranches.add(branch)
-  return true
-}
-
+/**
+ * Structured 1+x has exactly one hard completion condition after bootstrap:
+ * at least one usable task-directed evidence observation. Mixed classifications
+ * remain model-visible advisory evidence; they do not create branch quotas.
+ */
 function missingEvidence(state) {
-  if (state.mixedBranches.length > 0) {
-    return state.mixedBranches.filter((kind) => !state.completedBranches.has(kind)).length
-  }
-  return state.successfulEvidenceCalls >= 1 ? 0 : 1
+  return state.postBootstrapEvidenceCalls >= 1 ? 0 : 1
 }
 
 function hasCoreFollowup(messages) {
@@ -347,6 +379,8 @@ function withoutEvidenceFollowups(messages) {
     const id = message && typeof message.id === 'string' ? message.id : ''
     return !(
       id.includes('vision-router-structured-followup-') ||
+      // Strip historical Round-0 mixed hard guards if one is still present in
+      // a persisted/compatibility message surface. Round 1 never emits them.
       id.includes('vision-router-structured-mixed-guard-') ||
       id.includes('vision-router-structured-evidence-guard-')
     )
@@ -374,30 +408,8 @@ function appendSyntheticGuardOnce(decision, baseMessages, state, message, stripE
 }
 
 function rearmEvidenceGuardAfterAttempt(state) {
-  const mixedPrefix = `vision-router-structured-mixed-guard-${state.turn}-`
   const evidenceId = `vision-router-structured-evidence-guard-${state.turn}`
-  for (const id of state.emittedGuardIds) {
-    if (id === evidenceId || id.startsWith(mixedPrefix)) state.emittedGuardIds.delete(id)
-  }
-}
-
-function mixedReminder(state, config) {
-  const language = languageOf(config)
-  const overrides = normalizeGuidanceOverrides(config?.guidanceOverrides)
-  const pending = state.mixedBranches.filter((kind) => !state.completedBranches.has(kind))
-  const done = state.mixedBranches.filter((kind) => state.completedBranches.has(kind))
-  const separator = language === 'en' ? ': ' : '：'
-  const lines = pending.map((kind) => `- ${kind}${separator}${guidanceFor(kind, overrides, language)}`)
-  if (language === 'en') {
-    const doneText = done.length > 0 ? `Verified branches: ${done.join(' + ')}. ` : ''
-    return `${doneText}This mixed image still has unverified branches: ${pending.join(' + ')}. ` +
-      'Focus each vision call on one unfinished branch; answer only after each branch has produced at least one useful piece of evidence.\n' +
-      lines.join('\n')
-  }
-  const doneText = done.length > 0 ? `已验证分支：${done.join(' + ')}。` : ''
-  return `${doneText}混合图片仍有未验证分支：${pending.join(' + ')}。` +
-    '每次视觉调用只聚焦一个尚未完成的分支；这些分支分别产生至少一次有效证据后再作答。\n' +
-    lines.join('\n')
+  state.emittedGuardIds.delete(evidenceId)
 }
 
 function appendGuardMessage(decision, payload, state, config) {
@@ -421,12 +433,12 @@ function appendGuardMessage(decision, payload, state, config) {
             ? 'The visual turn time budget is exhausted. Do not call more vision tools; answer from the evidence already collected and state any remaining uncertainty.'
             : state.depthExhausted
               ? 'The configured deep-dive call cap has been reached. Do not call more vision tools; answer from the evidence already collected and state any remaining uncertainty.'
-              : 'Repeated follow-up vision attempts did not produce or advance usable evidence. Do not keep retrying vision tools; answer from the evidence already collected and state the remaining uncertainty.'
+              : 'Repeated follow-up vision attempts did not produce usable evidence. Do not keep retrying vision tools; answer from the evidence already collected and state the remaining uncertainty.'
           : state.budgetExhausted
             ? '本轮视觉总时间预算已耗尽。不要再调用视觉工具；请基于已经获得的证据作答，并明确仍存在的不确定性。'
             : state.depthExhausted
               ? '已达到本轮设置的深挖次数上限。不要再调用视觉工具；请基于已经获得的证据作答，并明确仍存在的不确定性。'
-              : '连续多次后续视觉调用都没有产出或推进可用证据。不要继续重复调用视觉工具；请基于已经获得的证据作答，并明确仍存在的不确定性。',
+              : '连续多次后续视觉调用都没有产出可用证据。不要继续重复调用视觉工具；请基于已经获得的证据作答，并明确仍存在的不确定性。',
       }],
       source: { kind: 'plugin', plugin: 'dsh-vision-router' },
     }
@@ -435,16 +447,10 @@ function appendGuardMessage(decision, payload, state, config) {
 
   if (!state.bootstrapDone || missingEvidence(state) === 0) return decision
 
-  if (state.mixedBranches.length > 0) {
-    const message = {
-      role: 'user',
-      id: `vision-router-structured-mixed-guard-${state.turn}-${state.successfulEvidenceCalls}`,
-      content: [{ type: 'text', text: mixedReminder(state, config) }],
-      source: { kind: 'plugin', plugin: 'dsh-vision-router' },
-    }
-    return appendSyntheticGuardOnce(decision, baseMessages, state, message, true)
-  }
-
+  // Core owns bootstrap ordering and its one-shot follow-up guidance. This
+  // outer boundary owns the hard x>=1 completion rule. If Core's guidance is
+  // already present, do not duplicate it; if Core's legacy presentation latch
+  // stopped emitting it after an empty/failed attempt, re-assert the hard guard.
   if (hasCoreFollowup(baseMessages)) return decision
   const message = {
     role: 'user',
@@ -452,8 +458,8 @@ function appendGuardMessage(decision, payload, state, config) {
     content: [{
       type: 'text',
       text: language === 'en'
-        ? 'The structured bootstrap is complete, but the previous follow-up tool did not produce usable evidence. Call at least one more vision tool that can add or verify evidence, then answer after it succeeds.'
-        : '结构化预识别已经完成，但上一条后续工具没有产出可用证据。请再调用至少一个能新增或验证证据的视觉工具，成功后再作答。',
+        ? 'The structured bootstrap is complete, but no usable task-directed visual evidence has been produced yet. Call at least one vision tool that can add or verify evidence, then answer after it succeeds.'
+        : '结构化预识别已经完成，但目前还没有产出可用的任务定向视觉证据。请至少调用一个能新增或验证证据的视觉工具，成功后再作答。',
     }],
     source: { kind: 'plugin', plugin: 'dsh-vision-router' },
   }
@@ -542,9 +548,11 @@ function wrapRegisteredTool(def, states, getConfig) {
           if (evidence !== undefined) {
             state.bootstrapDone = true
             state.bootstrapResult = result
-            state.mixedBranches = normalizeMixedBranches(evidence, depthOf(getConfig()))
-            state.completedBranches.clear()
-            state.mixedAttemptSignatures.clear()
+            // Structured 1+x is explicitly ordered: x starts only after the
+            // bootstrap succeeds. Keep the all-turn successful-evidence count
+            // intact for the user's explicit call cap; post-bootstrap progress
+            // has its own counter so an early model call cannot satisfy x>=1.
+            state.postBootstrapEvidenceCalls = 0
             resetNonProgress(state)
             state.budgetExhausted = false
             refreshDepthExhaustion(state, getConfig())
@@ -565,7 +573,7 @@ function wrapRegisteredTool(def, states, getConfig) {
         if (state.bootstrapDone && missingEvidence(state) > 0 && state.nonProgressExhausted) {
           return failure(
             'VISION_NO_PROGRESS_LIMIT',
-            `the structured follow-up stopped after ${MAX_NON_PROGRESS_ATTEMPTS} consecutive attempts without usable evidence progress`,
+            `the structured follow-up stopped after ${MAX_NON_PROGRESS_ATTEMPTS} consecutive attempts without usable evidence`,
           )
         }
 
@@ -577,18 +585,11 @@ function wrapRegisteredTool(def, states, getConfig) {
         )
         const code = resultCode(result)
         if (code === 'VISION_TURN_BUDGET_EXCEEDED') state.budgetExhausted = true
-        if (producedUsableStructuredEvidence(result)) {
-          const progressed = recordEvidenceSuccess(state, def.name, args)
+        if (producedUsableToolEvidence(def.name, result, args)) {
+          state.successfulEvidenceCalls += 1
+          if (state.bootstrapDone) state.postBootstrapEvidenceCalls += 1
           refreshDepthExhaustion(state, active)
-          if (state.bootstrapDone && missingEvidence(state) > 0) {
-            if (progressed) resetNonProgress(state)
-            else {
-              recordNonProgress(state)
-              if (!state.nonProgressExhausted) rearmEvidenceGuardAfterAttempt(state)
-            }
-          } else {
-            resetNonProgress(state)
-          }
+          resetNonProgress(state)
         } else if (code !== 'VISION_TURN_BUDGET_EXCEEDED' && state.bootstrapDone && missingEvidence(state) > 0) {
           recordNonProgress(state)
           if (!state.nonProgressExhausted) rearmEvidenceGuardAfterAttempt(state)

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "dsh-vision-router",
-  "version": "2.1.2",
+  "version": "2.1.3",
   "description": "Eyes for text-only DeepSeek Harness agents: built-in free vision chain (no key) + pixel-level vision tools (Q&A, grounding, crop, pixel diff, colors, OCR, SVG trace, cutout, screenshots). One-command install, no Python, image turns work like ordinary tool-calling turns.",
   "license": "MIT",
   "repository": {
@@ -69,13 +69,13 @@
     "sharp": "^0.35.3"
   },
   "scripts": {
-    "test": "node --test tests/adapter-prepare-call-compat.test.js tests/core.test.js tests/issue-374-delegated-call-config.test.js tests/grounding-coordinate-frame.test.js tests/grounding-coordinate-runtime.test.js tests/capability-advisory.test.js tests/vision-execution-policy.test.js tests/vision-backend-diagnostics.test.js tests/live-model-discovery.test.js tests/live-model-credential-resolution.test.js tests/vision-model-registry.test.js tests/live-model-client-stability.test.js tests/providers-persistence.test.js tests/provider-directory-fallback.test.js tests/model-catalog-policy.test.js tests/settings-ia-client-prelude.test.js tests/settings-ia-fj.test.js tests/settings-ia-acceptance-regression.test.js tests/settings-migration.test.js tests/settings-section-order.test.js tests/attachment-admission-policy.test.js tests/vision-resilience.test.js tests/vision-resilience-lifecycle.test.js tests/vision-breaker-readonly.test.js tests/alpha1-browser-lifecycle-integration.test.js tests/issue-367-remote-session-inject.test.js tests/client.test.js tests/issue-307-regression.test.js tests/issue-284-vision-mode-toggle.test.js tests/issue-284-vision-mode-hardening.test.js tests/issue-284-attachment-id-hint.test.js tests/issue-284-model-visibility.test.js tests/issue-284-vision-selection-effort.test.js tests/client-presentation-boundary.test.js tests/remote-settings-bridge.test.js tests/remote-settings-v2-contract.test.js tests/remote-settings-risk-confirmation.test.js tests/local-remote-settings-permission.test.js tests/rc6-real-settings-persistence.test.js tests/artifact-path-security.test.js tests/http-compat.test.js tests/catalog-corrections.test.js tests/update-check.test.js tests/update-check-signal-lifecycle.test.js tests/self-update.test.js tests/doctor.test.js tests/doctor-cli.test.js tests/legacy-session-repair.test.js tests/profile-pnpm-diagnostics.test.js tests/doctor-v2.test.js tests/doctor-cli-v2.test.js tests/doctor-runtime.test.js tests/dsh-host-capabilities.test.js tests/doctor-host-capabilities.test.js tests/session-repair-v2.test.js tests/bundle-defaults.test.js tests/file-logger.test.js tests/replay-delegation.test.js tests/twin-image-capability-fallback.test.js tests/twin-image-custom-wrapper.test.js tests/adversarial-hardening.test.js tests/vision-adversarial-hardening-v2.test.js tests/vision-capability-benchmark-visual-proof.test.js tests/wrapper-directory.test.js tests/logging-ui.test.js tests/manifest-dependencies.test.js tests/structured-bootstrap.test.js tests/structured-bootstrap-gate.test.js tests/local-ollama.test.js tests/ollama-cold-start.test.js tests/ollama-cold-start-loaded.test.js tests/zero-regression-gate.test.js tests/runtime-e2e.test.js tests/local-vision-stabilizer.test.js tests/local-connection-probe.test.js tests/repetition-guard.test.js tests/guide-scroll-gate.test.js tests/settings-scroll-jank-regression.test.js tests/android-attachment-compat.test.js tests/free-cloud-first.test.js tests/rc6-rc7-compat.test.js tests/native-image-coexistence.test.js tests/issue-276-session-image-ownership.test.js tests/issue-276-entry-policy-bridge.test.js tests/issue-276-policy-hardening.test.js tests/issue-289-native-nonintervention.test.js tests/pi-ai-bridge-wire-compat.test.js tests/mixed-router.test.js tests/depth-tier.test.js tests/depth-quota-behavior.test.js tests/structured-flow-hardening.test.js tests/structured-flow-adversarial.test.js tests/vision-budget-activation.test.js tests/runtime-boundary-fixes.test.js tests/runtime-i18n.test.js tests/tesseract-node24-boot.test.js tests/session-vision-state.test.js tests/session-vision-state-integration.test.js tests/image-resource-governor.test.js tests/resource-retention-lifecycle.test.js tests/pixel-diff-stream.test.js tests/large-image-resource-integration.test.js tests/structured-guard-idempotency.test.js tests/vision-routing-product.test.js tests/vision-runtime-performance.test.js tests/vision-routing-settings-prelude.test.js tests/v2-settings-ia-integration.test.js tests/vision-capability-router.test.js tests/vision-capability-reference.test.js tests/vision-capability-benchmark.test.js tests/vision-capability-grounding-proof-contract.test.js tests/vision-capability-probe.test.js tests/vision-capability-axis-freshness.test.js tests/vision-capability-shadow-health.test.js tests/vision-capability-benchmark-service.test.js tests/vision-capability-preflight-hardening.test.js tests/vision-capability-benchmark-contract.test.js tests/vision-capability-benchmark-client.test.js tests/vision-capability-benchmark-client-productization.test.js tests/vision-capability-adapter-route.test.js tests/vision-background-benchmark.test.js tests/vision-background-lifecycle.test.js tests/vision-background-benchmark-productization.test.js tests/vision-capability-benchmark-budget.test.js tests/vision-image-input-verdict.test.js tests/vision-capability-failure-classification.test.js tests/vision-capability-gate-client.test.js tests/v2-product-review-fixes.test.js tests/qa-top5-reliability.test.js tests/qa-screenshot-runtime.test.js tests/qa-runtime-identity.test.js tests/qa-turn-budget-cancellation.test.js tests/qa-cancellation-publication.test.js tests/qa-reliability-tail.test.js tests/qa-endpoint-route-alias.test.js tests/settings-native-card-semantics.test.js tests/settings-native-card-lazy-data.test.js tests/settings-invalid-staged-dirty.test.js tests/vision-toggle-root-hardening.test.js && node --test tests/vision-backend-runtime-policy.test.js",
+    "test": "node --test tests/adapter-prepare-call-compat.test.js tests/core.test.js tests/issue-374-delegated-call-config.test.js tests/grounding-coordinate-frame.test.js tests/grounding-coordinate-runtime.test.js tests/capability-advisory.test.js tests/vision-execution-policy.test.js tests/vision-backend-diagnostics.test.js tests/live-model-discovery.test.js tests/live-model-credential-resolution.test.js tests/vision-model-registry.test.js tests/live-model-client-stability.test.js tests/providers-persistence.test.js tests/provider-directory-fallback.test.js tests/model-catalog-policy.test.js tests/settings-ia-client-prelude.test.js tests/settings-ia-fj.test.js tests/settings-ia-acceptance-regression.test.js tests/settings-migration.test.js tests/settings-section-order.test.js tests/attachment-admission-policy.test.js tests/vision-resilience.test.js tests/vision-resilience-lifecycle.test.js tests/vision-breaker-readonly.test.js tests/alpha1-browser-lifecycle-integration.test.js tests/issue-367-remote-session-inject.test.js tests/vision-mode-toggle-injection.test.js tests/client.test.js tests/issue-307-regression.test.js tests/issue-284-vision-mode-toggle.test.js tests/issue-284-vision-mode-hardening.test.js tests/issue-284-attachment-id-hint.test.js tests/issue-284-model-visibility.test.js tests/issue-284-vision-selection-effort.test.js tests/client-presentation-boundary.test.js tests/clipboard-image-paste-compat.test.js tests/remote-settings-bridge.test.js tests/remote-settings-v2-contract.test.js tests/remote-settings-risk-confirmation.test.js tests/local-remote-settings-permission.test.js tests/rc6-real-settings-persistence.test.js tests/artifact-path-security.test.js tests/http-compat.test.js tests/catalog-corrections.test.js tests/update-check.test.js tests/update-check-signal-lifecycle.test.js tests/self-update.test.js tests/doctor.test.js tests/doctor-cli.test.js tests/legacy-session-repair.test.js tests/profile-pnpm-diagnostics.test.js tests/doctor-v2.test.js tests/doctor-cli-v2.test.js tests/doctor-runtime.test.js tests/dsh-host-capabilities.test.js tests/doctor-host-capabilities.test.js tests/session-repair-v2.test.js tests/bundle-defaults.test.js tests/file-logger.test.js tests/replay-delegation.test.js tests/twin-image-capability-fallback.test.js tests/twin-image-custom-wrapper.test.js tests/adversarial-hardening.test.js tests/vision-adversarial-hardening-v2.test.js tests/vision-capability-benchmark-visual-proof.test.js tests/wrapper-directory.test.js tests/logging-ui.test.js tests/manifest-dependencies.test.js tests/structured-bootstrap.test.js tests/structured-bootstrap-gate.test.js tests/vision-model-guidance-contract.test.js tests/local-ollama.test.js tests/ollama-cold-start.test.js tests/ollama-cold-start-loaded.test.js tests/zero-regression-gate.test.js tests/runtime-e2e.test.js tests/local-vision-stabilizer.test.js tests/local-connection-probe.test.js tests/repetition-guard.test.js tests/guide-scroll-gate.test.js tests/settings-scroll-jank-regression.test.js tests/android-attachment-compat.test.js tests/free-cloud-first.test.js tests/rc6-rc7-compat.test.js tests/native-image-coexistence.test.js tests/issue-276-session-image-ownership.test.js tests/issue-276-entry-policy-bridge.test.js tests/issue-276-policy-hardening.test.js tests/issue-289-native-nonintervention.test.js tests/pi-ai-bridge-wire-compat.test.js tests/mixed-router.test.js tests/depth-tier.test.js tests/depth-quota-behavior.test.js tests/structured-flow-hardening.test.js tests/structured-flow-adversarial.test.js tests/vision-budget-activation.test.js tests/runtime-boundary-fixes.test.js tests/runtime-i18n.test.js tests/tesseract-node24-boot.test.js tests/session-vision-state.test.js tests/session-vision-state-integration.test.js tests/image-resource-governor.test.js tests/resource-retention-lifecycle.test.js tests/pixel-diff-stream.test.js tests/large-image-resource-integration.test.js tests/structured-guard-idempotency.test.js tests/vision-routing-product.test.js tests/vision-runtime-performance.test.js tests/vision-routing-settings-prelude.test.js tests/v2-settings-ia-integration.test.js tests/vision-capability-router.test.js tests/vision-capability-reference.test.js tests/vision-capability-benchmark.test.js tests/vision-capability-grounding-proof-contract.test.js tests/vision-capability-probe.test.js tests/vision-capability-axis-freshness.test.js tests/vision-capability-shadow-health.test.js tests/vision-capability-benchmark-service.test.js tests/vision-capability-preflight-hardening.test.js tests/vision-capability-benchmark-contract.test.js tests/vision-capability-benchmark-client.test.js tests/vision-capability-benchmark-client-productization.test.js tests/vision-capability-adapter-route.test.js tests/vision-background-benchmark.test.js tests/vision-background-lifecycle.test.js tests/vision-background-benchmark-productization.test.js tests/vision-capability-benchmark-budget.test.js tests/vision-image-input-verdict.test.js tests/vision-capability-failure-classification.test.js tests/vision-capability-gate-client.test.js tests/v2-product-review-fixes.test.js tests/qa-top5-reliability.test.js tests/qa-screenshot-runtime.test.js tests/qa-runtime-identity.test.js tests/qa-turn-budget-cancellation.test.js tests/qa-cancellation-publication.test.js tests/qa-reliability-tail.test.js tests/qa-endpoint-route-alias.test.js tests/settings-native-card-semantics.test.js tests/settings-native-card-lazy-data.test.js tests/settings-invalid-staged-dirty.test.js tests/vision-toggle-root-hardening.test.js && node --test tests/vision-backend-runtime-policy.test.js",
     "test:core": "node --test tests/core.test.js tests/runtime-e2e.test.js tests/structured-bootstrap.test.js tests/structured-flow-hardening.test.js tests/structured-flow-adversarial.test.js tests/vision-execution-policy.test.js tests/vision-backend-runtime-policy.test.js",
     "test:routing": "node --test tests/issue-374-delegated-call-config.test.js tests/twin-image-capability-fallback.test.js tests/twin-image-custom-wrapper.test.js tests/vision-routing-product.test.js tests/vision-capability-router.test.js tests/vision-capability-probe.test.js tests/vision-capability-shadow-health.test.js tests/vision-runtime-performance.test.js tests/vision-background-benchmark.test.js tests/vision-image-input-verdict.test.js",
     "test:session": "node --test tests/session-vision-state.test.js tests/session-vision-state-integration.test.js tests/replay-delegation.test.js tests/legacy-session-repair.test.js tests/session-repair-v2.test.js tests/issue-276-session-image-ownership.test.js",
     "test:resources": "node --test tests/artifact-path-security.test.js tests/image-resource-governor.test.js tests/resource-retention-lifecycle.test.js tests/pixel-diff-stream.test.js tests/large-image-resource-integration.test.js tests/qa-turn-budget-cancellation.test.js tests/qa-cancellation-publication.test.js",
     "test:compat": "node --test tests/adapter-prepare-call-compat.test.js tests/attachment-admission-policy.test.js tests/rc6-rc7-compat.test.js tests/rc6-real-settings-persistence.test.js tests/android-attachment-compat.test.js tests/http-compat.test.js tests/pi-ai-bridge-wire-compat.test.js tests/native-image-coexistence.test.js tests/tesseract-node24-boot.test.js tests/dsh-host-capabilities.test.js",
-    "test:web": "node --test tests/alpha1-browser-lifecycle-integration.test.js tests/issue-367-remote-session-inject.test.js tests/client.test.js tests/client-presentation-boundary.test.js tests/settings-ia-client-prelude.test.js tests/settings-ia-acceptance-regression.test.js tests/remote-settings-bridge.test.js tests/remote-settings-v2-contract.test.js tests/local-remote-settings-permission.test.js tests/v2-settings-ia-integration.test.js tests/logging-ui.test.js",
+    "test:web": "node --test tests/alpha1-browser-lifecycle-integration.test.js tests/issue-367-remote-session-inject.test.js tests/vision-mode-toggle-injection.test.js tests/client.test.js tests/client-presentation-boundary.test.js tests/clipboard-image-paste-compat.test.js tests/settings-ia-client-prelude.test.js tests/settings-ia-acceptance-regression.test.js tests/remote-settings-bridge.test.js tests/remote-settings-v2-contract.test.js tests/local-remote-settings-permission.test.js tests/v2-settings-ia-integration.test.js tests/logging-ui.test.js",
     "test:contract": "node --test tests/manifest-dependencies.test.js tests/bundle-defaults.test.js tests/zero-regression-gate.test.js tests/issue-289-native-nonintervention.test.js tests/issue-374-delegated-call-config.test.js tests/runtime-boundary-fixes.test.js tests/runtime-i18n.test.js tests/adapter-prepare-call-compat.test.js tests/rc6-rc7-compat.test.js tests/attachment-admission-policy.test.js tests/dsh-host-capabilities.test.js tests/doctor-host-capabilities.test.js tests/structured-bootstrap.test.js tests/structured-bootstrap-gate.test.js",
     "test:grounding": "node --test tests/grounding-coordinate-frame.test.js tests/grounding-coordinate-runtime.test.js tests/vision-capability-grounding-proof-contract.test.js",
     "test:stress": "node scripts/image-resource-stress.mjs && node --test tests/large-image-resource-integration.test.js tests/resource-retention-lifecycle.test.js"