@qijenchen/design-system 0.1.0-beta.13 → 0.1.0-beta.131

package/ds-canonical/hooks/tests/test_check_audit_post_report_validator.sh

@@ -3,7 +3,7 @@
 #
 # Hook(PostToolUse Write/Edit/MultiEdit):驗 audit report quality:
 #   A) NO-SAMPLE keyword 偵測
-#   B) Dim ≥ 46 coverage
+#   B) matrix-derived full Dim coverage
 #   C) audit-prompts.md prompt count
 #   D) @benchmark-unverified-blanket count drift
 #   E) prune-chain-trigger → emit additionalContext JSON
@@ -19,6 +19,84 @@ SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
 HOOK="$SCRIPT_DIR/../check_audit_post_report_validator.sh"
 TMPDIR_TEST=$(mktemp -d)
 trap 'rm -rf "$TMPDIR_TEST"' EXIT
+mkdir -p "$TMPDIR_TEST/.claude/logs"
+mkdir -p "$TMPDIR_TEST/corpus/packages/governance/canonical" \
+  "$TMPDIR_TEST/corpus/canonical/skills/design-system-audit/references" \
+  "$TMPDIR_TEST/scripts"
+cat > "$TMPDIR_TEST/scripts/verify-deep-audit-coverage.mjs" <<'EOF'
+import { existsSync, readFileSync } from 'node:fs'
+import { resolve } from 'node:path'
+
+const rootIndex = process.argv.indexOf('--repo-root')
+const root = rootIndex >= 0 ? process.argv[rootIndex + 1] : process.cwd()
+const modePath = resolve(root, '.fixture-review-mode')
+const mode = existsSync(modePath) ? readFileSync(modePath, 'utf8').trim() : 'missing'
+if (mode === 'waived') {
+  process.stdout.write(JSON.stringify({
+    evidenceKind: 'deep-audit-coverage-verification',
+    secondOpinion: 'waived-by-user',
+    providers: { self: 'alpha', peer: null },
+  }))
+  process.exit(0)
+}
+if (mode === 'selected') {
+  process.stdout.write(JSON.stringify({
+    evidenceKind: 'deep-audit-coverage-verification',
+    secondOpinion: 'completed',
+    providers: { self: 'alpha', peer: 'beta' },
+  }))
+  process.exit(0)
+}
+process.exit(2)
+EOF
+cat > "$TMPDIR_TEST/corpus/packages/governance/canonical/providers.json" <<'EOF'
+{
+  "canonical": {"roots": {"skills": "canonical/skills"}},
+  "providers": [
+    {
+      "id": "alpha",
+      "displayName": "Alpha",
+      "adapter": {"skillBindings": {"deep-audit-cross-codex": {
+        "self": "alpha",
+        "peer": "beta",
+        "certificationContract": "exact-provider-runtime-surface-role-target-v1"
+      }}}
+    },
+    {"id": "beta", "displayName": "Beta", "adapter": {"skillBindings": {}}},
+    {
+      "id": "future-author",
+      "displayName": "Future Author",
+      "adapter": {"skillBindings": {"deep-audit-cross-codex": {
+        "self": "future-author",
+        "peer": "nebula",
+        "certificationContract": "exact-provider-runtime-surface-role-target-v1"
+      }}}
+    },
+    {"id": "nebula", "displayName": "Nebula", "adapter": {"skillBindings": {}}}
+  ]
+}
+EOF
+
+# Model the provider-hook adapter boundary:the hook receives only a concrete peer
+# resolved from a certified independent-review binding,never a guessed provider literal.
+resolve_fixture_peer() {
+  local self_provider="$1" resolved
+  resolved=$(jq -cer --arg self "$self_provider" '
+    . as $registry
+    | first($registry.providers[] | select(.id == $self)) as $selfProvider
+    | $selfProvider.adapter.skillBindings["deep-audit-cross-codex"] as $binding
+    | select(
+        $binding.self == $self
+        and $binding.certificationContract == "exact-provider-runtime-surface-role-target-v1"
+      )
+    | first($registry.providers[] | select(.id == $binding.peer))
+    | {id, displayName}
+  ' "$TMPDIR_TEST/corpus/packages/governance/canonical/providers.json") || return 1
+  export GOVERNANCE_SELF_PROVIDER="$self_provider"
+  GOVERNANCE_PEER_PROVIDER=$(printf '%s' "$resolved" | jq -er '.id') || return 1
+  GOVERNANCE_PEER_DISPLAY_NAME=$(printf '%s' "$resolved" | jq -er '.displayName') || return 1
+  export GOVERNANCE_PEER_PROVIDER GOVERNANCE_PEER_DISPLAY_NAME
+}
 
 if [ ! -f "$HOOK" ]; then echo "FATAL: hook not found: $HOOK"; exit 1; fi
 
@@ -33,7 +111,9 @@ run_hook() {
     '{tool_name: $tn, tool_input: {file_path: $fp}}')
   STDOUT=$(mktemp); STDERR=$(mktemp)
   set +e
-  export CLAUDE_PROJECT_DIR="$TMPDIR_TEST"
+  export GOVERNANCE_PROJECT_DIR="$TMPDIR_TEST"
+  export GOVERNANCE_CORPUS_ROOT="$TMPDIR_TEST/corpus"
+  export GOVERNANCE_READ_ONLY=1
   printf '%s' "$payload" | bash "$HOOK" >"$STDOUT" 2>"$STDERR"
   EXIT=$?
   set -e
@@ -63,6 +143,29 @@ expect_stdout_contains() {
   fi
 }
 
+# 2026-05-31:Validator C/F/G critical fail 現 exit 2 真 block(原只 additionalContext soft)。
+expect_block_stderr() {
+  local name="$1"; local needle="$2"
+  if [ "$EXIT" = "2" ] && echo "$STDERR_TEXT" | grep -qF "$needle"; then
+    echo "  PASS  $name"; PASS=$((PASS+1))
+  else
+    echo "  FAIL  $name (expected exit 2 + stderr '$needle', got exit $EXIT)"
+    FAIL=$((FAIL+1)); FAILED_TESTS="${FAILED_TESTS}\n  - $name"
+  fi
+}
+
+expect_integrity() {
+  local name="$1"; local needle="$2"
+  if [ "$EXIT" = "70" ] && [ -z "$STDOUT_TEXT" ] \
+    && echo "$STDERR_TEXT" | grep -qF "GOVERNANCE_INTEGRITY:" \
+    && echo "$STDERR_TEXT" | grep -qF "$needle"; then
+    echo "  PASS  $name"; PASS=$((PASS+1))
+  else
+    echo "  FAIL  $name (expected integrity exit 70 + '$needle', got exit $EXIT)"
+    FAIL=$((FAIL+1)); FAILED_TESTS="${FAILED_TESTS}\n  - $name"
+  fi
+}
+
 echo "=== check_audit_post_report_validator tests ==="
 
 # 1. Non-audit file path → skip silently
@@ -78,33 +181,208 @@ run_hook "$TMPDIR_TEST/.claude/logs/audit-report-missing.json" "Write"
 expect_silent "3. missing file → skip"
 
 # 4. project_audit_progress.md with NO-SAMPLE keyword but no audit-prompts.md → no prune trigger, silent stdout
-mkdir -p "$TMPDIR_TEST/.claude/memory"
-cat > "$TMPDIR_TEST/.claude/memory/project_audit_progress.md" <<'EOF'
+mkdir -p "$TMPDIR_TEST/governance/memory"
+cat > "$TMPDIR_TEST/governance/memory/project_audit_progress.md" <<'EOF'
 # Audit progress
 Dim 1: pass
 Dim 2: pass
 Dim 3: pass
 EOF
-run_hook "$TMPDIR_TEST/.claude/memory/project_audit_progress.md" "Write"
-# B fires (DIM_COUNT < 46) but no TRIGGER_PRUNE since audit-prompts.md absent + no bench debt
+run_hook "$TMPDIR_TEST/governance/memory/project_audit_progress.md" "Write"
+# B fires (DIM_COUNT < matrix-derived total) but no TRIGGER_PRUNE since audit-prompts.md absent + no bench debt
 # stdout should be silent (TRIGGER_PRUNE=0)
 expect_silent "4. valid file no prune trigger → silent stdout"
 
-# 5. audit-prompts.md present but only 2 prompts → TRIGGER_PRUNE=1, stdout JSON
-mkdir -p "$TMPDIR_TEST/.claude/skills/design-system-audit/references"
+# 5. judgment dim(dim 99 PURE-JUDGMENT)缺 `## 99.` prompt → Validator C CRITICAL → exit 2 真 block
+#    (2026-05-31:原測 additionalContext soft;改測 set-membership C + exit-2 block)
+mkdir -p "$TMPDIR_TEST/.claude/skills/design-system-audit/references" "$TMPDIR_TEST/generated/governance"
+cat > "$TMPDIR_TEST/generated/governance/audit-coverage-matrix.json" <<'EOF'
+{"coverage_by_dim":{"99":{"tier":"PURE-JUDGMENT","mechanism":"test judgment dim"}}}
+EOF
+cat > "$TMPDIR_TEST/corpus/canonical/skills/design-system-audit/references/audit-prompts.md" <<'EOF'
+## 1. foo
+prompt
+EOF
+# Poisoned generated Claude view claims dim 99 exists. Canonical authority must ignore it.
 cat > "$TMPDIR_TEST/.claude/skills/design-system-audit/references/audit-prompts.md" <<'EOF'
-### Dim 1: foo
+## 1. foo
 prompt
-
-### Dim 2: bar
+## 99. poisoned generated view
 prompt
 EOF
-cat > "$TMPDIR_TEST/.claude/memory/project_audit_progress.md" <<'EOF'
+cat > "$TMPDIR_TEST/governance/memory/project_audit_progress.md" <<'EOF'
 # Audit progress
 Dim 1: pass
 EOF
-run_hook "$TMPDIR_TEST/.claude/memory/project_audit_progress.md" "Edit"
-expect_stdout_contains "5. audit-prompts coverage low → prune trigger JSON" "prune-chain-trigger"
+run_hook "$TMPDIR_TEST/governance/memory/project_audit_progress.md" "Edit"
+expect_block_stderr "5. judgment dim 缺 prompt → Validator C exit-2 block" "VALIDATOR BLOCK"
+
+# 6/7. Validator K:when review is required,a decision needs SSOT-check /
+#    world-class cite / registry-resolved peer verdict / design-fit.
+#    First satisfy Validator C so these cases isolate K.
+cat > "$TMPDIR_TEST/corpus/canonical/skills/design-system-audit/references/audit-prompts.md" <<'EOF'
+## 1. foo
+prompt
+## 99. test judgment dim
+prompt
+EOF
+mkdir -p "$TMPDIR_TEST/.claude/logs/deep-audit-test"
+resolve_fixture_peer alpha || {
+  echo "FATAL:fixture lacks a concrete certified independent-review peer"
+  exit 1
+}
+cat > "$TMPDIR_TEST/.claude/logs/deep-audit-test/C1-final-report.md" <<'EOF'
+# Deep Audit 報告
+Dim 1: pass
+
+### 待你拍板
+1. 決策一:Popover 內距是否改 12px
+   - SSOT 理由:改跨元件 canonical 視覺結構
+   - SSOT-check:grep spec + memory,無既有拍板
+   - Beta verdict:agree(辯論共識)
+   - 設計語言 fit:符合 density canonical
+EOF
+run_hook "$TMPDIR_TEST/.claude/logs/deep-audit-test/C1-final-report.md" "Write"
+expect_block_stderr "6. 決策缺世界級 cite → Validator K exit-2 block" "決策品質要件不全"
+
+cat > "$TMPDIR_TEST/.claude/logs/deep-audit-test/C1-final-report.md" <<'EOF'
+# Deep Audit 報告
+Dim 1: pass
+
+### 待你拍板
+1. 決策一:Popover 內距是否改 12px
+   - SSOT 理由:改跨元件 canonical 視覺結構
+   - SSOT-check:grep spec + memory,無既有拍板
+   - 世界級:Polaris https://polaris.shopify.com/tokens/space / Material / Ant(3 家)
+   - Beta verdict:agree(辯論共識)
+   - 設計語言 fit:符合 density canonical
+EOF
+run_hook "$TMPDIR_TEST/.claude/logs/deep-audit-test/C1-final-report.md" "Write"
+if [ "$EXIT" = "0" ]; then
+  echo "  PASS  7. 四要件齊 → Validator K pass(exit 0)"; PASS=$((PASS+1))
+else
+  echo "  FAIL  7. 四要件齊 → Validator K pass (expected exit 0, got $EXIT, stderr=[$STDERR_TEXT])"
+  FAIL=$((FAIL+1)); FAILED_TESTS="${FAILED_TESTS}\n  - 7. 四要件齊 pass"
+fi
+
+# 8. The independent-review requirement follows the registry-resolved peer,
+#    not a provider literal baked into the validator.
+cat > "$TMPDIR_TEST/.claude/logs/deep-audit-test/C1-final-report.md" <<'EOF'
+# Deep Audit 報告
+Dim 1: pass
+
+### 待你拍板
+1. 決策一:Popover 內距是否改 12px
+   - SSOT 理由:改跨元件 canonical 視覺結構
+   - SSOT-check:grep spec + memory,無既有拍板
+   - 世界級:Polaris https://polaris.shopify.com/tokens/space / Material / Ant(3 家)
+   - Nebula verdict:agree
+   - 設計語言 fit:符合 density canonical
+EOF
+resolve_fixture_peer future-author || {
+  echo "  FAIL  8. future peer fixture lacks a certified registry binding"
+  FAIL=$((FAIL+1)); FAILED_TESTS="${FAILED_TESTS}\n  - 8. future peer fixture binding"
+}
+run_hook "$TMPDIR_TEST/.claude/logs/deep-audit-test/C1-final-report.md" "Write"
+if [ "$EXIT" = "0" ]; then
+  echo "  PASS  8. future peer verdict satisfies the same four requirements"; PASS=$((PASS+1))
+else
+  echo "  FAIL  8. future peer was coupled to a Codex literal (exit=$EXIT, stderr=[$STDERR_TEXT])"
+  FAIL=$((FAIL+1)); FAILED_TESTS="${FAILED_TESTS}\n  - 8. future peer verdict"
+fi
+
+# 9. Report prose alone cannot authorize a waiver. The same receipt passes only
+#    when the current immutable run verifier proves the exact one-run waiver.
+unset GOVERNANCE_PEER_PROVIDER GOVERNANCE_PEER_DISPLAY_NAME
+cat > "$TMPDIR_TEST/.claude/logs/deep-audit-test/C1-final-report.md" <<'EOF'
+# Deep Audit 報告
+Dim 1: pass
+second opinion: waived by user
+
+### 待你拍板
+1. 決策一:Popover 內距是否改 12px
+   - SSOT 理由:改跨元件 canonical 視覺結構
+   - SSOT-check:grep spec + memory,無既有拍板
+   - 世界級:Polaris https://polaris.shopify.com/tokens/space / Material / Ant(3 家)
+   - 設計語言 fit:符合 density canonical
+EOF
+run_hook "$TMPDIR_TEST/.claude/logs/deep-audit-test/C1-final-report.md" "Write"
+expect_block_stderr "9. text-only user waiver → block" "Report 文字不是 second-opinion waiver authority"
+
+printf 'waived\n' > "$TMPDIR_TEST/.fixture-review-mode"
+run_hook "$TMPDIR_TEST/.claude/logs/deep-audit-test/C1-final-report.md" "Write"
+expect_silent "9b. immutable user-waived run → no peer context required"
+
+printf 'selected\n' > "$TMPDIR_TEST/.fixture-review-mode"
+run_hook "$TMPDIR_TEST/.claude/logs/deep-audit-test/C1-final-report.md" "Write"
+expect_block_stderr "9c. selected run cannot be waived by report prose" "Report 文字不是 second-opinion waiver authority"
+
+printf 'tampered\n' > "$TMPDIR_TEST/.fixture-review-mode"
+run_hook "$TMPDIR_TEST/.claude/logs/deep-audit-test/C1-final-report.md" "Write"
+expect_block_stderr "9d. tampered/missing active waiver proof → block" "Report 文字不是 second-opinion waiver authority"
+
+printf 'waived\n' > "$TMPDIR_TEST/.fixture-review-mode"
+cat > "$TMPDIR_TEST/.claude/logs/deep-audit-test/C1-final-report.md" <<'EOF'
+# Deep Audit 報告
+Dim 1: pass
+
+### 待你拍板
+1. 決策一:Popover 內距是否改 12px
+   - SSOT 理由:改跨元件 canonical 視覺結構
+   - SSOT-check:grep spec + memory,無既有拍板
+   - 世界級:Polaris https://polaris.shopify.com/tokens/space / Material / Ant(3 家)
+   - 設計語言 fit:符合 density canonical
+EOF
+run_hook "$TMPDIR_TEST/.claude/logs/deep-audit-test/C1-final-report.md" "Write"
+expect_block_stderr "9e. active waiver still requires explicit report receipt" "registry-resolved peer provider"
+
+# 10. A general report with no decision and an explicit task-level receipt also
+#     has no peer dependency.
+cat > "$TMPDIR_TEST/.claude/logs/deep-audit-test/C1-final-report.md" <<'EOF'
+# Scoped Audit 報告
+Dim 1: pass
+second opinion: not required by task
+EOF
+run_hook "$TMPDIR_TEST/.claude/logs/deep-audit-test/C1-final-report.md" "Write"
+expect_silent "10. no-decision scoped report → no peer context required"
+
+# 11. If the report does require independent review,it must carry the adapter's
+#     registry-resolved peer context.
+cat > "$TMPDIR_TEST/.claude/logs/deep-audit-test/C1-final-report.md" <<'EOF'
+# Scoped Audit 報告
+Dim 1: pass
+second opinion: required
+EOF
+run_hook "$TMPDIR_TEST/.claude/logs/deep-audit-test/C1-final-report.md" "Write"
+expect_block_stderr "11. required review without peer → block" "registry-resolved peer provider"
+
+# 12/13. A deep final report must close knowledge prune and governance coverage
+#        in the same run. The recovery trigger is not a final-report receipt.
+cat > "$TMPDIR_TEST/.claude/logs/deep-audit-test/C1-final-report.md" <<'EOF'
+# Deep Audit 報告
+Dim 1: pass
+Dim 2: pass
+Dim 3: pass
+Dim 4: pass
+Dim 5: pass
+Dim 6: pass
+Dim 7: pass
+Dim 8: pass
+Dim 9: pass
+Dim 10: pass
+A.1b story-vs-code: files scanned: 1; 0 findings
+performance-audit: N/A; ux-audit: N/A; visual-audit: N/A
+second opinion: waived by user
+governance-coverage: same-run reconciled; unavailable telemetry = UNOBSERVED
+EOF
+run_hook "$TMPDIR_TEST/.claude/logs/deep-audit-test/C1-final-report.md" "Write"
+expect_block_stderr "12. deep final missing same-run prune receipt → block" "Same-run knowledge-prune receipt"
+
+cat >> "$TMPDIR_TEST/.claude/logs/deep-audit-test/C1-final-report.md" <<'EOF'
+knowledge-prune: same-run complete
+EOF
+run_hook "$TMPDIR_TEST/.claude/logs/deep-audit-test/C1-final-report.md" "Write"
+expect_silent "13. deep final with all same-run receipts → pass"
 
 echo ""
 echo "=== Summary ==="

package/ds-canonical/hooks/tests/test_check_benchmark_citation.sh

@@ -3,8 +3,8 @@
 #
 # Hook(PreToolUse Edit/Write/MultiEdit):spec.md / .tsx in packages/design-system/src/
 # 含 world-class benchmark claim(Ant / Material / Polaris ...)必附 inline citation
-# (URL / GitHub #L / snapshots/ / @benchmark-unverified)。違 P1 soft warn(exit 1)。
-# 整檔 escape:前 5 行含 @benchmark-citation-allow: 或 @benchmark-unverified-blanket:。
+# (URL / GitHub #L / snapshots/ / @benchmark-unverified)。違 P0 BLOCKER(exit 2,2026-07-07 方向 2 升級)。
+# Component specs 是 zero-debt scope,禁止逐句與檔級 escape；其他 scope 保留既有 escape。
 
 set -u
 
@@ -45,10 +45,10 @@ expect_pass_silent() {
 
 expect_warn() {
   local name="$1"; local needle="$2"
-  if [ "$EXIT" = "1" ] && echo "$STDERR_TEXT" | grep -qF "$needle"; then
+  if [ "$EXIT" = "2" ] && echo "$STDERR_TEXT" | grep -qF "$needle"; then
     echo "  PASS  $name"; PASS=$((PASS+1))
   else
-    echo "  FAIL  $name (expected exit=1 + needle '$needle', got exit $EXIT)"
+    echo "  FAIL  $name (expected exit=2 + needle '$needle', got exit $EXIT)"
     echo "  --- stderr ---"; echo "$STDERR_TEXT" | sed 's/^/    /'; echo "  --- end ---"
     FAIL=$((FAIL+1)); FAILED_TESTS="${FAILED_TESTS}\n  - $name"
   fi
@@ -89,22 +89,54 @@ run_hook_write "/repo/packages/design-system/src/components/Foo/foo.spec.md" '
 '
 expect_warn "5. Polaris claim missing cite → warn" "check_benchmark_citation"
 
-# 6. DS path Atlassian claim with @benchmark-unverified → silent (撤回 marker)
+# 6. Non-component spec can still use the legacy per-claim retraction
+run_hook_write "/repo/packages/design-system/src/patterns/Foo/foo.spec.md" '
+# Foo spec
+
+對齊 Atlassian Design @benchmark-unverified 待補 source
+'
+expect_pass_silent "6. non-component claim with @benchmark-unverified → silent"
+
+# 7. Non-component file-level @benchmark-citation-allow escape → silent
+run_hook_write "/repo/packages/design-system/src/patterns/Foo/foo.spec.md" '// @benchmark-citation-allow: legacy migration backlog
+# Foo
+
+對齊 Carbon Design 內部共識。
+對齊 Polaris 派系。
+'
+expect_pass_silent "7. non-component file-level @benchmark-citation-allow → silent"
+
+# 8. Component per-claim retraction cannot recreate evidence debt
 run_hook_write "/repo/packages/design-system/src/components/Foo/foo.spec.md" '
 # Foo spec
 
 對齊 Atlassian Design @benchmark-unverified 待補 source
 '
-expect_pass_silent "6. claim with @benchmark-unverified → silent"
+expect_warn "8. component @benchmark-unverified → BLOCK" "component spec benchmark evidence debt"
 
-# 7. File-level @benchmark-citation-allow escape → silent
-run_hook_write "/repo/packages/design-system/src/components/Foo/foo.spec.md" '// @benchmark-citation-allow: legacy migration backlog
+# 9. Component blanket retraction cannot bypass the zero-debt rule
+run_hook_write "/repo/packages/design-system/src/components/Foo/foo.spec.md" '<!-- @benchmark-unverified-blanket: legacy backlog -->
 # Foo
 
 對齊 Carbon Design 內部共識。
+'
+expect_warn "9. component @benchmark-unverified-blanket → BLOCK" "zero-debt scope"
+
+# 10. Component file-level citation allow cannot bypass the zero-debt rule
+run_hook_write "/repo/packages/design-system/src/components/Foo/foo.spec.md" '<!-- @benchmark-citation-allow: legacy backlog -->
+# Foo
+
 對齊 Polaris 派系。
 '
-expect_pass_silent "7. file-level @benchmark-citation-allow → silent"
+expect_warn "10. component @benchmark-citation-allow → BLOCK" "zero-debt scope"
+
+# 11. Verified component claim remains valid
+run_hook_write "/repo/packages/design-system/src/components/Foo/foo.spec.md" '
+# Foo
+
+對齊 Ant Design，來源 https://ant-design.com/components/button
+'
+expect_pass_silent "11. component claim + direct citation → silent"
 
 echo ""
 echo "=== Summary ==="

package/ds-canonical/hooks/tests/test_check_canonical_propagation.sh

@@ -14,19 +14,57 @@ run_hook() {
   STDOUT=$(mktemp); STDERR=$(mktemp)
   set +e
   jq -n --arg tool "$tool" --arg fp "$file_path" --arg c "$content" \
-    '{tool_name:$tool, tool_input:{file_path:$fp, content:$c}}' \
+    '{hook_event_name:"PreToolUse", tool_name:$tool, tool_input:{file_path:$fp, content:$c}}' \
     | bash "$HOOK" >"$STDOUT" 2>"$STDERR"
   EXIT=$?
   set -e
-  STDERR_TEXT=$(cat "$STDERR"); rm -f "$STDOUT" "$STDERR"
+  STDOUT_TEXT=$(cat "$STDOUT"); STDERR_TEXT=$(cat "$STDERR"); rm -f "$STDOUT" "$STDERR"
+}
+
+run_raw() {
+  local payload="$1"
+  STDOUT=$(mktemp); STDERR=$(mktemp)
+  set +e
+  printf '%s' "$payload" | bash "$HOOK" >"$STDOUT" 2>"$STDERR"
+  EXIT=$?
+  set -e
+  STDOUT_TEXT=$(cat "$STDOUT"); STDERR_TEXT=$(cat "$STDERR"); rm -f "$STDOUT" "$STDERR"
+}
+
+is_exact_context() {
+  local needle="$1" compact
+  compact=$(printf '%s' "$STDOUT_TEXT" | jq -c . 2>/dev/null) || return 1
+  [ "$STDOUT_TEXT" = "$compact" ] || return 1
+  printf '%s' "$STDOUT_TEXT" | jq -se --arg needle "$needle" '
+    length == 1
+    and (.[0] | type == "object" and (keys | sort) == ["governanceContext"])
+    and (.[0].governanceContext |
+      type == "object"
+      and (keys | sort) == ["hookEventName", "message"]
+      and .hookEventName == "PreToolUse"
+      and (.message | type == "string" and length > 0 and contains($needle)))
+  ' >/dev/null
 }
 
 expect_exit() {
   local name="$1" expect="$2" needle="${3:-}"
-  if [ "$EXIT" = "$expect" ] && { [ -z "$needle" ] || echo "$STDERR_TEXT" | grep -qF "$needle"; }; then
+  local ok=1
+  if [ "$expect" = "0" ] && [ -n "$needle" ]; then
+    [ "$EXIT" = "0" ] && [ -z "$STDERR_TEXT" ] && is_exact_context "$needle" && ok=0
+  elif [ "$expect" = "0" ]; then
+    [ "$EXIT" = "0" ] && [ -z "$STDOUT_TEXT" ] && [ -z "$STDERR_TEXT" ] && ok=0
+  elif [ "$expect" = "2" ]; then
+    [ "$EXIT" = "2" ] && [ -z "$STDOUT_TEXT" ] \
+      && { [ -z "$needle" ] || echo "$STDERR_TEXT" | grep -qF "$needle"; } && ok=0
+  elif [ "$expect" = "70" ]; then
+    [ "$EXIT" = "70" ] && [ -z "$STDOUT_TEXT" ] \
+      && { [ -z "$needle" ] || echo "$STDERR_TEXT" | grep -qF "$needle"; } && ok=0
+  fi
+  if [ "$ok" = "0" ]; then
     echo "  PASS  $name"; PASS=$((PASS+1))
   else
     echo "  FAIL  $name (expected exit $expect${needle:+ + needle '$needle'}, got exit $EXIT)"
+    echo "  --- stdout ---"; echo "$STDOUT_TEXT" | sed 's/^/    /'; echo "  --- end ---"
     echo "  --- stderr ---"; echo "$STDERR_TEXT" | sed 's/^/    /'; echo "  --- end ---"
     FAIL=$((FAIL+1)); FAILED_TESTS="${FAILED_TESTS}\n  - $name"
   fi
@@ -106,12 +144,12 @@ expect_exit "E.2.4 no L3 import → silent" 0 ""
 echo ""
 echo "=== E.3 spec-impl default alignment ==="
 
-# 10. spec.md with 「預設」keyword → WARN stderr
+# 10. spec.md with 「預設」keyword → governance context
 run_hook "/r/packages/design-system/src/components/Foo/foo.spec.md" '
 ## Width
 預設 = trigger-width(同 popover anchor 寬)
 ' Write
-expect_exit "E.3.1 預設 keyword → WARN stderr" 0 "spec-impl default"
+expect_exit "E.3.1 預設 keyword → governance context" 0 "spec-impl default"
 
 # 11. spec.md without default keyword → silent
 run_hook "/r/packages/design-system/src/components/Foo/foo.spec.md" '
@@ -127,6 +165,19 @@ run_hook "/r/packages/design-system/src/components/Foo/foo.spec.md" '// @spec-im
 ' Write
 expect_exit "E.3.3 spec allowlist → silent" 0 ""
 
+echo ""
+echo "=== Output contract ==="
+
+# Quotes/newlines in warning data must remain one safely encoded compact JSON document.
+run_hook '/r/packages/design-system/src/components/Foo/"quoted".spec.md' '
+## "Quoted default"
+預設 = "trigger-width"
+' Write
+expect_exit "O.1 adversarial quotes/newlines → exact encoded context" 0 "spec-impl default"
+
+run_raw '{'
+expect_exit "O.2 malformed input → stderr-only integrity failure" 70 "GOVERNANCE_INTEGRITY:"
+
 echo ""
 echo "═══ Results: $PASS PASS, $FAIL FAIL ═══"
 [ "$FAIL" -gt 0 ] && { printf "Failed:%b\n" "$FAILED_TESTS"; exit 1; }