opencode-agent-skill 9.0.0 → 11.0.0

package/lib/visual-spec.mjs

@@ -0,0 +1,159 @@
+function number(value, fallback = 0) {
+  const parsed = Number(value)
+  return Number.isFinite(parsed) ? parsed : fallback
+}
+
+function range(value, tolerance = 0) {
+  if (Array.isArray(value) && value.length >= 2) return [number(value[0]), number(value[1])]
+  const exact = number(value)
+  return [exact - tolerance, exact + tolerance]
+}
+
+function within(actual, expected, tolerance) {
+  const [min, max] = range(expected, tolerance)
+  return actual >= Math.min(min, max) && actual <= Math.max(min, max)
+}
+
+function normalizeElement(item = {}) {
+  return {
+    id: String(item.id || "").trim(),
+    role: item.role || null,
+    region: item.region || null,
+    expected: {
+      ...(item.expected || {}),
+      ...(item.x !== undefined ? { x: item.x } : {}),
+      ...(item.y !== undefined ? { y: item.y } : {}),
+      ...(item.width !== undefined ? { width: item.width } : {}),
+      ...(item.height !== undefined ? { height: item.height } : {}),
+    },
+    tolerance: {
+      position: number(item.tolerance?.position, 8),
+      size: number(item.tolerance?.size, 8),
+    },
+    required: item.required !== false,
+  }
+}
+
+export function normalizeVisualSpec(spec = {}) {
+  const viewport = {
+    width: Math.max(1, number(spec.viewport?.width, 1440)),
+    height: Math.max(1, number(spec.viewport?.height, 900)),
+    deviceScaleFactor: Math.max(0.1, number(spec.viewport?.deviceScaleFactor, 1)),
+  }
+  return {
+    schemaVersion: 1,
+    name: spec.name || null,
+    viewport,
+    elements: (spec.elements || []).map(normalizeElement).filter((item) => item.id),
+    regions: Array.isArray(spec.regions) ? spec.regions : [],
+    tokens: spec.tokens || {},
+    metadata: spec.metadata || {},
+  }
+}
+
+export function validateVisualSpec(spec = {}) {
+  const normalized = normalizeVisualSpec(spec)
+  const errors = []
+  const ids = new Set()
+  for (const item of normalized.elements) {
+    if (ids.has(item.id)) errors.push("duplicate element id: " + item.id)
+    ids.add(item.id)
+    for (const key of ["x", "y", "width", "height"]) {
+      if (item.expected[key] === undefined) continue
+      const values = Array.isArray(item.expected[key]) ? item.expected[key] : [item.expected[key]]
+      if (values.some((value) => !Number.isFinite(Number(value)))) errors.push(item.id + ": invalid " + key)
+    }
+  }
+  return { valid: errors.length === 0, errors, spec: normalized }
+}
+
+function actualMap(actual = []) {
+  const values = Array.isArray(actual) ? actual : Object.values(actual || {})
+  return new Map(values.filter((item) => item?.id).map((item) => [String(item.id), item]))
+}
+
+export function createGeometryReceipt(specInput, actualInput, options = {}) {
+  const checked = validateVisualSpec(specInput)
+  if (!checked.valid) throw new Error("Invalid visual spec: " + checked.errors.join("; "))
+  const actual = actualMap(actualInput)
+  const elements = []
+  let failed = 0
+
+  for (const expected of checked.spec.elements) {
+    const found = actual.get(expected.id)
+    if (!found) {
+      const pass = !expected.required
+      if (!pass) failed += 1
+      elements.push({ id: expected.id, verdict: pass ? "PASS" : "FAIL", reason: "missing", expected: expected.expected, actual: null })
+      continue
+    }
+
+    const failures = []
+    const checks = {}
+    for (const key of ["x", "y"]) {
+      if (expected.expected[key] === undefined) continue
+      const ok = within(number(found[key]), expected.expected[key], expected.tolerance.position)
+      checks[key] = ok
+      if (!ok) failures.push(key)
+    }
+    for (const key of ["width", "height"]) {
+      if (expected.expected[key] === undefined) continue
+      const ok = within(number(found[key]), expected.expected[key], expected.tolerance.size)
+      checks[key] = ok
+      if (!ok) failures.push(key)
+    }
+    const pass = failures.length === 0
+    if (!pass) failed += 1
+    elements.push({
+      id: expected.id,
+      verdict: pass ? "PASS" : "FAIL",
+      failures,
+      checks,
+      expected: expected.expected,
+      actual: {
+        x: number(found.x),
+        y: number(found.y),
+        width: number(found.width),
+        height: number(found.height),
+      },
+    })
+  }
+
+  return {
+    schemaVersion: 1,
+    kind: "visual-geometry",
+    verdict: failed === 0 ? "PASS" : "FAIL",
+    viewport: checked.spec.viewport,
+    total: elements.length,
+    passed: elements.length - failed,
+    failed,
+    elements,
+    generatedAt: options.generatedAt || new Date().toISOString(),
+  }
+}
+
+export function buildVisualRepairPlan(receipt, pixelDiff = null) {
+  const failedElements = (receipt?.elements || []).filter((item) => item.verdict !== "PASS")
+  return {
+    schemaVersion: 1,
+    action: failedElements.length || (pixelDiff && pixelDiff.differentPixels > 0) ? "repair" : "accept",
+    failedElementIDs: failedElements.map((item) => item.id),
+    geometryFailures: failedElements.map((item) => ({ id: item.id, failures: item.failures || [item.reason] })),
+    pixelRegion: pixelDiff?.bounds || null,
+    directives: [
+      ...(failedElements.length ? ["edit only the owning component/style for failed geometry unless evidence shows a shared token defect"] : []),
+      ...(pixelDiff?.bounds ? ["inspect and, if vision is available, crop only the largest changed pixel region before another edit"] : []),
+      "re-render the affected viewport and produce a fresh receipt before declaring success",
+    ],
+  }
+}
+
+export function responsiveViewportMatrix(input = null) {
+  if (Array.isArray(input) && input.length) return input
+  return [
+    { id: "mobile", width: 390, height: 844 },
+    { id: "tablet", width: 768, height: 1024 },
+    { id: "desktop", width: 1440, height: 900 },
+    { id: "wide", width: 1920, height: 1080 },
+  ]
+}

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "opencode-agent-skill",
-  "version": "9.0.0",
-  "description": "Evidence-first engineering runtime with adaptive orchestration, incremental semantic context, bounded ACI, sandboxed verification, replayable traces and benchmark-gated learning for OpenCode",
+  "version": "11.0.0",
+  "description": "Perception-aware evidence-first engineering runtime with adaptive context, capability routing, visual/browser verification, weak-model recovery and benchmark-gated execution for OpenCode",
   "type": "module",
   "bin": {
     "ocskill": "bin/ocskill.mjs"
@@ -23,7 +23,7 @@
     "evals": "node scripts/eval-skills.mjs",
     "evals:live:validate": "node scripts/validate-live-suite.mjs",
     "test": "node --test test/*.test.mjs",
-    "ci": "npm run syntax && npm run validate && npm run evals && npm run evals:router && npm run evals:live:validate && npm run evals:long:validate && npm run evals:polyglot:validate && npm test && npm pack --dry-run && npm run smoke:pack && npm run smoke:plain-install",
+    "ci": "npm run syntax && npm run validate && npm run evals && npm run evals:router && npm run evals:v11:validate && npm run evals:live:validate && npm run evals:long:validate && npm run evals:polyglot:validate && npm test && npm pack --dry-run && npm run smoke:pack && npm run smoke:plain-install",
     "postinstall": "node scripts/install.mjs",
     "preuninstall": "node scripts/uninstall.mjs",
     "prepublishOnly": "npm run ci",
@@ -45,7 +45,10 @@
     "release:check-tag": "node scripts/check-release-tag.mjs",
     "evals:polyglot:validate": "node scripts/validate-live-suite.mjs --suite polyglot",
     "evals:polyglot": "node scripts/eval-live.mjs --suite polyglot",
-    "evals:matrix:gate": "node scripts/eval-matrix.mjs --require-confidence"
+    "evals:matrix:gate": "node scripts/eval-matrix.mjs --require-confidence",
+    "evals:ablation": "node scripts/eval-ablation.mjs",
+    "evals:v11": "node --test test/*-v11.test.mjs",
+    "evals:v11:validate": "node scripts/validate-v11-suite.mjs"
   },
   "keywords": [
     "opencode",
@@ -55,7 +58,10 @@
     "engineering-workflow",
     "code-review",
     "debugging",
-    "verification"
+    "verification",
+    "visual-testing",
+    "browser-automation",
+    "multimodal-agents"
   ],
   "engines": {
     "node": ">=20"

package/scripts/eval-ablation.mjs

@@ -0,0 +1,47 @@
+#!/usr/bin/env node
+import { readFile } from "node:fs/promises"
+import path from "node:path"
+import { compareEvalSummaries } from "../lib/eval-ablation.mjs"
+
+const args = process.argv.slice(2)
+
+function option(name, fallback) {
+  const index = args.indexOf(name)
+  return index >= 0 && index + 1 < args.length ? args[index + 1] : fallback
+}
+
+function positional() {
+  const values = []
+  for (let index = 0; index < args.length; index += 1) {
+    if (args[index].startsWith("--")) {
+      if (["--pass-rate-tolerance", "--min-initial-reduction", "--max-token-ratio", "--max-duration-ratio", "--min-cacheable-ratio", "--min-evidence-reuse-ratio", "--max-repeated-stable-ratio"].includes(args[index])) index += 1
+      continue
+    }
+    values.push(args[index])
+  }
+  return values
+}
+
+const [referenceFile, candidateFile] = positional()
+if (!referenceFile || !candidateFile) {
+  console.error("Usage: node scripts/eval-ablation.mjs <reference-summary.json> <candidate-summary.json> [--require-gate] [--min-initial-reduction 0.10]")
+  process.exit(2)
+}
+
+const [reference, candidate] = await Promise.all([
+  readFile(path.resolve(referenceFile), "utf8").then(JSON.parse),
+  readFile(path.resolve(candidateFile), "utf8").then(JSON.parse),
+])
+
+const report = compareEvalSummaries(reference, candidate, {
+  passRateTolerance: Number(option("--pass-rate-tolerance", "0")),
+  minInitialInputReduction: Number(option("--min-initial-reduction", "0.10")),
+  maxTotalTokenRatio: Number(option("--max-token-ratio", "1.05")),
+  maxDurationRatio: Number(option("--max-duration-ratio", "1.10")),
+  minCacheableRatio: option("--min-cacheable-ratio", null) == null ? null : Number(option("--min-cacheable-ratio", null)),
+  minEvidenceReuseRatio: option("--min-evidence-reuse-ratio", null) == null ? null : Number(option("--min-evidence-reuse-ratio", null)),
+  maxRepeatedStableRatio: option("--max-repeated-stable-ratio", null) == null ? null : Number(option("--max-repeated-stable-ratio", null)),
+})
+
+console.log(JSON.stringify(report, null, 2))
+if (args.includes("--require-gate") && !report.gateEligible) process.exitCode = 1

package/scripts/eval-matrix.mjs

@@ -32,6 +32,8 @@ const onlyLive = has("--standard-only")
 const onlyPolyglot = has("--polyglot-only")
 const withoutPolyglot = has("--without-polyglot")
 const requireConfidence = has("--require-confidence")
+const maxInitialInputRatio = Number(argValue("--max-initial-input-ratio", "1.5"))
+const maxTokenRatio = Number(argValue("--max-token-ratio", "1.75"))
 
 if (!model) {
   console.error("Usage: node scripts/eval-matrix.mjs --model provider/model [--trials 3] [--auth current|env-only] [--variant high] [--without-polyglot|--long-only|--standard-only|--polyglot-only]")
@@ -110,7 +112,10 @@ const baselineCount = results.filter((item) => item.mode === "baseline").length
 const uesCount = results.filter((item) => item.mode === "ues").length
 const coverageComplete = baselineCount === expectedPerMode && uesCount === expectedPerMode
 const summary = summarizeEvalResults(results)
-const confidence = pairedBenchmarkConfidence(results)
+const confidence = pairedBenchmarkConfidence(results, {
+  maxInitialInputRatio: Number.isFinite(maxInitialInputRatio) ? maxInitialInputRatio : 1.5,
+  maxTokenRatio: Number.isFinite(maxTokenRatio) ? maxTokenRatio : 1.75,
+})
 
 const report = {
   schemaVersion: 1,
@@ -134,7 +139,11 @@ const report = {
     mode: item.mode,
     passed: item.passed === true,
     durationMs: item.durationMs ?? null,
-    telemetry: item.telemetry?.costSamples > 0 ? { cost: item.telemetry.cost } : {},
+    telemetry: {
+      ...(item.telemetry?.costSamples > 0 ? { cost: item.telemetry.cost } : {}),
+      ...(item.telemetry?.firstUsage ? { firstUsage: item.telemetry.firstUsage } : {}),
+      ...(item.telemetry?.tokens ? { tokens: item.telemetry.tokens } : {}),
+    },
   })),
   summary,
   confidence,
@@ -149,6 +158,8 @@ console.log("- UES: " + (summary.modes.ues?.passed || 0) + "/" + (summary.modes.
 console.log("- pass-rate delta: " + (summary.passRateDelta == null ? "n/a" : (summary.passRateDelta * 100).toFixed(1) + " pp"))
 console.log("- coverage: " + (coverageComplete ? "COMPLETE" : "INCOMPLETE"))
 console.log("- paired confidence: " + (confidence.promotionEligible ? "SUPPORTED" : "NOT YET SUPPORTED") + " (pairs=" + confidence.pairs + ", p=" + confidence.pValue.toFixed(4) + ")")
+console.log("- initial-input ratio UES/baseline: " + (confidence.initialInput.ratio == null ? "n/a" : confidence.initialInput.ratio.toFixed(3)) + " (max " + confidence.initialInput.maxRatio + ")")
+console.log("- total-token ratio UES/baseline: " + (confidence.tokens.ratio == null ? "n/a" : confidence.tokens.ratio.toFixed(3)) + " (max " + confidence.tokens.maxRatio + ")")
 console.log("- confidence gate: " + (requireConfidence ? "REQUIRED" : "report-only"))
 console.log("- report: " + reportFile)
 

package/scripts/validate-v11-suite.mjs

@@ -0,0 +1,58 @@
+import { existsSync } from "node:fs"
+import { readFile } from "node:fs/promises"
+import path from "node:path"
+import { fileURLToPath } from "node:url"
+
+const root = path.resolve(path.dirname(fileURLToPath(import.meta.url)), "..")
+const file = path.join(root, "evals", "v11", "tasks.json")
+const suite = JSON.parse(await readFile(file, "utf8"))
+const errors = []
+
+const REQUIRED_V11_FILES = [
+  "lib/context-engine-v11.mjs",
+  "lib/evidence-store.mjs",
+  "lib/evidence-budget.mjs",
+  "lib/prompt-cache.mjs",
+  "lib/capability-registry.mjs",
+  "lib/browser-adapter.mjs",
+  "lib/browser-runtime.mjs",
+  "lib/png-diff.mjs",
+  "lib/visual-spec.mjs",
+  "lib/dynamic-workflow.mjs",
+  "lib/skill-quality.mjs",
+  "lib/ui-inspector.mjs",
+  "lib/v11-metrics.mjs",
+  "global-config/agents/visual-verifier.md",
+  "global-config/agents/merge-arbiter.md",
+]
+for (const relative of REQUIRED_V11_FILES) {
+  if (!existsSync(path.join(root, relative))) errors.push("missing V11 runtime file: " + relative)
+}
+const ids = new Set()
+const categories = new Set()
+
+if (!Array.isArray(suite.tasks) || suite.tasks.length < 10) errors.push("V11 suite must contain at least 10 tasks")
+
+for (const task of suite.tasks || []) {
+  if (!task.id || ids.has(task.id)) errors.push("missing/duplicate id: " + (task.id || "<missing>"))
+  ids.add(task.id)
+  if (!task.category) errors.push((task.id || "<missing>") + ": missing category")
+  categories.add(task.category)
+  if (!task.objective || task.objective.length < 30) errors.push((task.id || "<missing>") + ": objective is too weak")
+  if (!Array.isArray(task.requiredFiles) || !task.requiredFiles.length) errors.push((task.id || "<missing>") + ": requiredFiles missing")
+  for (const relative of task.requiredFiles || []) {
+    if (!existsSync(path.join(root, relative))) errors.push((task.id || "<missing>") + ": missing required file " + relative)
+  }
+}
+
+for (const category of ["context","routing","visual","browser","workflow","sidecar"]) {
+  if (!categories.has(category)) errors.push("V11 suite missing category: " + category)
+}
+
+if (errors.length) {
+  console.error("V11 suite validation failed:")
+  for (const error of errors) console.error("- " + error)
+  process.exit(1)
+}
+
+console.log("Validated " + suite.tasks.length + " V11 contract tasks across " + categories.size + " categories and " + REQUIRED_V11_FILES.length + " required runtime files.")

package/scripts/validate.mjs

@@ -87,9 +87,12 @@ for (const entry of await readdir(agentsRoot, { withFileTypes: true })) {
   if (!/mode:\s*subagent/.test(source)) errors.push(`${entry.name}: agent must use mode: subagent`)
 }
 
-if (ids.size < 39) errors.push(`expected at least 39 skills, found ${ids.size}`)
+const V11_REQUIRED_SKILLS = ["visual-fidelity","browser-qa","design-source","responsive-verification","component-visual-testing","browser-security","skill-authoring","skill-evaluation","dynamic-workflow"]
+for (const id of V11_REQUIRED_SKILLS) if (!ids.has(id)) errors.push(`missing V11 skill ${id}`)
+
+if (ids.size < 48) errors.push(`expected at least 48 skills for V11, found ${ids.size}`)
 if (commands < 11) errors.push(`expected at least 11 commands, found ${commands}`)
-if (agents < 10) errors.push(`expected at least 10 subagents, found ${agents}`)
+if (agents < 12) errors.push(`expected at least 12 subagents for V11, found ${agents}`)
 
 const routerIndex = path.join(pluginsRoot, "ues-router", "index.js")
 const routerCore = path.join(pluginsRoot, "ues-router", "router.js")
@@ -115,6 +118,15 @@ if (existsSync(routerIndex)) {
   if (!source.includes('name: "task_policy"')) errors.push("v8 router plugin must expose adaptive task policy")
   if (!source.includes('name: "cancel_task"')) errors.push("v8 router plugin must expose executor cancellation")
   if (!source.includes('name: "recover_task"')) errors.push("v8 router plugin must expose task-scoped recovery")
+  if (!source.includes('name: "capability_requirements"')) errors.push("v11 router plugin must expose capability requirements")
+  if (!source.includes('name: "evidence_get"')) errors.push("v11 router plugin must expose bounded evidence retrieval")
+  if (!source.includes('name: "browser_plan"')) errors.push("v11 router plugin must expose browser verification planning")
+  if (!source.includes('name: "browser_inspect"')) errors.push("v11 router plugin must expose bounded browser inspection")
+  if (!source.includes('name: "visual_geometry"')) errors.push("v11 router plugin must expose visual geometry receipts")
+  if (!source.includes('name: "visual_compare"')) errors.push("v11 router plugin must expose deterministic PNG comparison")
+  if (!source.includes('name: "workflow_plan"')) errors.push("v11 router plugin must expose dynamic workflow planning")
+  if (!source.includes('name: "ui_tokens"')) errors.push("v11 router plugin must expose compact design-token extraction")
+  if (!source.includes('name: "ui_layout"')) errors.push("v11 router plugin must expose responsive UI layout verification")
   if (!source.includes("ctx.session.interrupt")) errors.push("v8 task dispatch must interrupt timed-out executors")
   if (!source.includes('"work", "heartbeat"')) errors.push("v8 task dispatch must refresh task leases")
 }
@@ -138,11 +150,11 @@ if (existsSync(plainInstallSmoke)) {
   }
 }
 
-for (const name of ["process-runner.mjs","evidence-receipt.mjs","gate-receipt.mjs","runtime-events.mjs","context-manifest.mjs","orchestrator-policy.mjs","worktree-sandbox.mjs","learning-engine.mjs","hermes-bridge.mjs","control-center.mjs"]) {
+for (const name of ["process-runner.mjs","evidence-receipt.mjs","gate-receipt.mjs","runtime-events.mjs","context-manifest.mjs","orchestrator-policy.mjs","worktree-sandbox.mjs","learning-engine.mjs","hermes-bridge.mjs","control-center.mjs","evidence-store.mjs","evidence-budget.mjs","prompt-cache.mjs","capability-registry.mjs","visual-spec.mjs","png-diff.mjs","browser-adapter.mjs","browser-runtime.mjs","dynamic-workflow.mjs","skill-quality.mjs","ui-inspector.mjs","context-engine-v11.mjs","v11-metrics.mjs"]) {
   if (!existsSync(path.join(root, "lib", name))) errors.push(`missing V8 core module ${name}`)
 }
 
-for (const name of ["codebase-mapper.md","plan-checker.md","executor.md","integration-verifier.md"]) {
+for (const name of ["codebase-mapper.md","plan-checker.md","executor.md","integration-verifier.md","visual-verifier.md","merge-arbiter.md"]) {
   if (!existsSync(path.join(agentsRoot, name))) errors.push(`missing V6 subagent ${name}`)
 }