diff --git a/reports/benchmark_reproducibility.json b/reports/benchmark_reproducibility.json
index 23f973de..36ee29c5 100644
--- a/reports/benchmark_reproducibility.json
+++ b/reports/benchmark_reproducibility.json
@@ -3,24 +3,21 @@
   "ok": true,
   "generated_at": "2026-06-16",
   "skill_dir": ".",
-  "commit": "391821240ba14f088d0fbe12dd5f1fdca7917a39",
+  "commit": "a7960d4cfd2d9a17cbaebcee7b823f0db13df0cf",
   "git_status": {
     "available": true,
     "dirty": true,
-    "changed_file_count": 36,
+    "changed_file_count": 9,
     "sample": [
-      " M registry/index.json",
-      " M registry/packages/yao-meta-skill.json",
-      " M reports/adoption_drift_report.json",
-      " M reports/architecture_maintainability.json",
-      " M reports/architecture_maintainability.md",
       " M reports/benchmark_reproducibility.json",
       " M reports/benchmark_reproducibility.md",
-      " M reports/context_budget.json",
-      " M reports/context_budget.md",
-      " M reports/context_budget_summary.json",
       " M reports/evidence_consistency.json",
-      " M reports/install_simulation.json"
+      " M reports/review-studio.html",
+      " M reports/review-studio.json",
+      " M reports/review-viewer.json",
+      " M reports/skill-interpretation.json",
+      " M reports/skill-os-2-review.md",
+      " M reports/skill-overview.json"
     ],
     "scope": "generation-time status before this report is written"
   },
@@ -52,7 +49,7 @@
     "public_claim_ready": false,
     "public_claim_blocker_count": 5,
     "working_tree_dirty": true,
-    "changed_file_count": 36
+    "changed_file_count": 9
   },
   "public_claim": {
     "ready": false,
@@ -68,7 +65,7 @@
   },
   "release_lock": {
     "ready": false,
-    "commit": "391821240ba14f088d0fbe12dd5f1fdca7917a39",
+    "commit": "a7960d4cfd2d9a17cbaebcee7b823f0db13df0cf",
     "status_scope": "generation-time status before this report is written",
     "reason": "working tree was dirty at generation time"
   },
diff --git a/reports/benchmark_reproducibility.md b/reports/benchmark_reproducibility.md
index 3d0b7310..f0c1a86b 100644
--- a/reports/benchmark_reproducibility.md
+++ b/reports/benchmark_reproducibility.md
@@ -1,7 +1,7 @@
 # Benchmark Reproducibility
 
 Generated at: `2026-06-16`
-Commit: `391821240ba14f088d0fbe12dd5f1fdca7917a39`
+Commit: `a7960d4cfd2d9a17cbaebcee7b823f0db13df0cf`
 Working tree dirty at generation: `true`
 Evidence bundle SHA256: `5ad19434d41f67e6a8dbcbccbd22ffefed11008d75ebb00a78f0d24f3f8910a8`
 
@@ -23,7 +23,7 @@ Evidence bundle SHA256: `5ad19434d41f67e6a8dbcbccbd22ffefed11008d75ebb00a78f0d24
 - world-class source checks: `6` pass / `13` total; `7` blocked
 - public claim ready: `false`
 - public claim blockers: `5`
-- changed files at generation: `36`
+- changed files at generation: `9`
 
 This report proves local benchmark reproducibility only. It keeps external provider and human-review gaps visible instead of counting them as complete. The git commit is generation-time context; the evidence bundle SHA is the durable anchor for the artifacts listed below.
 
diff --git a/reports/evidence_consistency.json b/reports/evidence_consistency.json
index 59a2fd12..bb9f3623 100644
--- a/reports/evidence_consistency.json
+++ b/reports/evidence_consistency.json
@@ -47,8 +47,8 @@
       "key": "benchmark-release-lock-self-consistency",
       "label": "Benchmark release lock matches git dirty state",
       "status": "pass",
-      "expected": false,
-      "actual": false,
+      "expected": true,
+      "actual": true,
       "paths": [
         "reports/benchmark_reproducibility.json"
       ],
@@ -58,8 +58,8 @@
       "key": "overview-benchmark-commit",
       "label": "overview embeds the benchmark commit",
       "status": "pass",
-      "expected": "391821240ba14f088d0fbe12dd5f1fdca7917a39",
-      "actual": "391821240ba14f088d0fbe12dd5f1fdca7917a39",
+      "expected": "a7960d4cfd2d9a17cbaebcee7b823f0db13df0cf",
+      "actual": "a7960d4cfd2d9a17cbaebcee7b823f0db13df0cf",
       "paths": [
         "reports/benchmark_reproducibility.json",
         "reports/skill-overview.json"
@@ -71,7 +71,7 @@
       "label": "overview embeds benchmark summary fields",
       "status": "pass",
       "expected": {
-        "release_lock_ready": false,
+        "release_lock_ready": true,
         "required_artifact_count": 24,
         "missing_artifact_count": 0,
         "source_contract_sha256": "9fde13846a41c2d1442b6e80350313246dedebe02590ce1f2f07f2da1109c9ea",
@@ -81,10 +81,10 @@
         "world_class_source_pass_count": 6,
         "world_class_source_blocked_count": 7,
         "public_claim_ready": false,
-        "public_claim_blocker_count": 5
+        "public_claim_blocker_count": 4
       },
       "actual": {
-        "release_lock_ready": false,
+        "release_lock_ready": true,
         "required_artifact_count": 24,
         "missing_artifact_count": 0,
         "source_contract_sha256": "9fde13846a41c2d1442b6e80350313246dedebe02590ce1f2f07f2da1109c9ea",
@@ -94,7 +94,7 @@
         "world_class_source_pass_count": 6,
         "world_class_source_blocked_count": 7,
         "public_claim_ready": false,
-        "public_claim_blocker_count": 5
+        "public_claim_blocker_count": 4
       },
       "paths": [
         "reports/benchmark_reproducibility.json",
@@ -204,8 +204,8 @@
       "key": "interpretation-benchmark-commit",
       "label": "interpretation embeds the benchmark commit",
       "status": "pass",
-      "expected": "391821240ba14f088d0fbe12dd5f1fdca7917a39",
-      "actual": "391821240ba14f088d0fbe12dd5f1fdca7917a39",
+      "expected": "a7960d4cfd2d9a17cbaebcee7b823f0db13df0cf",
+      "actual": "a7960d4cfd2d9a17cbaebcee7b823f0db13df0cf",
       "paths": [
         "reports/benchmark_reproducibility.json",
         "reports/skill-interpretation.json"
@@ -217,7 +217,7 @@
       "label": "interpretation embeds benchmark summary fields",
       "status": "pass",
       "expected": {
-        "release_lock_ready": false,
+        "release_lock_ready": true,
         "required_artifact_count": 24,
         "missing_artifact_count": 0,
         "source_contract_sha256": "9fde13846a41c2d1442b6e80350313246dedebe02590ce1f2f07f2da1109c9ea",
@@ -227,10 +227,10 @@
         "world_class_source_pass_count": 6,
         "world_class_source_blocked_count": 7,
         "public_claim_ready": false,
-        "public_claim_blocker_count": 5
+        "public_claim_blocker_count": 4
       },
       "actual": {
-        "release_lock_ready": false,
+        "release_lock_ready": true,
         "required_artifact_count": 24,
         "missing_artifact_count": 0,
         "source_contract_sha256": "9fde13846a41c2d1442b6e80350313246dedebe02590ce1f2f07f2da1109c9ea",
@@ -240,7 +240,7 @@
         "world_class_source_pass_count": 6,
         "world_class_source_blocked_count": 7,
         "public_claim_ready": false,
-        "public_claim_blocker_count": 5
+        "public_claim_blocker_count": 4
       },
       "paths": [
         "reports/benchmark_reproducibility.json",
@@ -2025,9 +2025,9 @@
       "label": "Skill OS 2.0 review summary mirrors current evidence",
       "status": "pass",
       "expected": [
-        "score `89`",
+        "score `91`",
         "`16` gates",
-        "`4` warnings",
+        "`3` warnings",
         "`29` declared internal modules",
         "`82 / 82` CLI help smoke checks passing across `111` scripts",
         "`618` zip entries",
diff --git a/reports/review-studio.html b/reports/review-studio.html
index 698405fa..4c46c3c7 100644
--- a/reports/review-studio.html
+++ b/reports/review-studio.html
@@ -670,28 +670,28 @@
       <div class="decision">
         <span>审查结论</span>
         <strong>review</strong>
-        <span>Score 89/100</span>
+        <span>Score 91/100</span>
       </div>
     </header>
 
     <section>
       <h2>核心指标</h2>
-      <div class="metrics"><article class='metric'><span>Skill IR</span><strong>2.0.0</strong><p>5 targets in platform-neutral contract</p></article><article class='metric'><span>Compiler</span><strong>5/5</strong><p>target contracts compiled from Skill IR</p></article><article class='metric'><span>Output Delta</span><strong>100.0</strong><p>5 cases; 1 file-backed</p></article><article class='metric'><span>Exec Runs</span><strong>10</strong><p>command 10; model 0; recorded 0</p></article><article class='metric'><span>Blind A/B</span><strong>5</strong><p>review pairs hide baseline vs with-skill labels</p></article><article class='metric'><span>Review Kit</span><strong>0/5</strong><p>pending 5; answer key hidden</p></article><article class='metric'><span>Review A/B</span><strong>0/5</strong><p>adjudication decisions; pending 5</p></article><article class='metric'><span>Public Claim</span><strong>blocked</strong><p>5 blockers; local reproducible true</p></article><article class='metric'><span>Blueprint</span><strong>21/21</strong><p>2.0 coverage; extensions partial 0, planned 0; evidence pending 4</p></article><article class='metric'><span>Runtime</span><strong>5/5</strong><p>target conformance pass rate</p></article><article class='metric'><span>Perm Probe</span><strong>4/4</strong><p>0 native; 4 installer-enforced</p></article><article class='metric'><span>Trust</span><strong>0</strong><p>111 scripts scanned; secrets found</p></article><article class='metric'><span>Py Compat</span><strong>0</strong><p>177 files scanned for Python 3.11</p></article><article class='metric'><span>Arch Debt</span><strong>0</strong><p>899 largest lines; 8 watchlist; 64 CLI handlers; 18 in entrypoint</p></article><article class='metric'><span>Atlas</span><strong>5</strong><p>12 scanned skills; route collisions</p></article><article class='metric'><span>Drift</span><strong>low</strong><p>1 metadata events; 0 missed triggers</p></article><article class='metric'><span>Waivers</span><strong>0</strong><p>0 gates covered; human risk decisions</p></article><article class='metric'><span>Intake</span><strong>4/4</strong><p>0 valid submissions; 0 invalid</p></article><article class='metric'><span>Claim Guard</span><strong>0</strong><p>173 public surfaces scanned</p></article><article class='metric'><span>Notes</span><strong>0/0</strong><p>0 open blocker annotations</p></article><article class='metric'><span>Registry</span><strong>1.1.0</strong><p>5 targets; MIT license</p></article><article class='metric'><span>Archive</span><strong>pass</strong><p>618 zip entries; package verification</p></article><article class='metric'><span>Install</span><strong>pass</strong><p>4 adapters; 12 permissions enforced; 0 permission failures</p></article><article class='metric'><span>Upgrade</span><strong>minor</strong><p>declared minor; 0 breaking changes</p></article></div>
+      <div class="metrics"><article class='metric'><span>Skill IR</span><strong>2.0.0</strong><p>5 targets in platform-neutral contract</p></article><article class='metric'><span>Compiler</span><strong>5/5</strong><p>target contracts compiled from Skill IR</p></article><article class='metric'><span>Output Delta</span><strong>100.0</strong><p>5 cases; 1 file-backed</p></article><article class='metric'><span>Exec Runs</span><strong>10</strong><p>command 10; model 0; recorded 0</p></article><article class='metric'><span>Blind A/B</span><strong>5</strong><p>review pairs hide baseline vs with-skill labels</p></article><article class='metric'><span>Review Kit</span><strong>0/5</strong><p>pending 5; answer key hidden</p></article><article class='metric'><span>Review A/B</span><strong>0/5</strong><p>adjudication decisions; pending 5</p></article><article class='metric'><span>Public Claim</span><strong>blocked</strong><p>4 blockers; local reproducible true</p></article><article class='metric'><span>Blueprint</span><strong>21/21</strong><p>2.0 coverage; extensions partial 0, planned 0; evidence pending 4</p></article><article class='metric'><span>Runtime</span><strong>5/5</strong><p>target conformance pass rate</p></article><article class='metric'><span>Perm Probe</span><strong>4/4</strong><p>0 native; 4 installer-enforced</p></article><article class='metric'><span>Trust</span><strong>0</strong><p>111 scripts scanned; secrets found</p></article><article class='metric'><span>Py Compat</span><strong>0</strong><p>177 files scanned for Python 3.11</p></article><article class='metric'><span>Arch Debt</span><strong>0</strong><p>899 largest lines; 8 watchlist; 64 CLI handlers; 18 in entrypoint</p></article><article class='metric'><span>Atlas</span><strong>5</strong><p>12 scanned skills; route collisions</p></article><article class='metric'><span>Drift</span><strong>low</strong><p>1 metadata events; 0 missed triggers</p></article><article class='metric'><span>Waivers</span><strong>0</strong><p>0 gates covered; human risk decisions</p></article><article class='metric'><span>Intake</span><strong>4/4</strong><p>0 valid submissions; 0 invalid</p></article><article class='metric'><span>Claim Guard</span><strong>0</strong><p>173 public surfaces scanned</p></article><article class='metric'><span>Notes</span><strong>0/0</strong><p>0 open blocker annotations</p></article><article class='metric'><span>Registry</span><strong>1.1.0</strong><p>5 targets; MIT license</p></article><article class='metric'><span>Archive</span><strong>pass</strong><p>618 zip entries; package verification</p></article><article class='metric'><span>Install</span><strong>pass</strong><p>4 adapters; 12 permissions enforced; 0 permission failures</p></article><article class='metric'><span>Upgrade</span><strong>minor</strong><p>declared minor; 0 breaking changes</p></article></div>
     </section>
 
     <section>
       <h2>审查闸门</h2>
-      <div class="gates"><article class='gate pass'><div><span>通过</span><h3>意图画布</h3></div><p>intent confidence 100/100; Intent is clear enough to package the first routeable version.</p><footer>reports/intent-confidence.json <a href='intent-confidence.md'>证据</a></footer></article><article class='gate pass'><div><span>通过</span><h3>触发实验</h3></div><p>13 trigger cases; 0 misroutes; 0 ambiguous</p><footer>reports/route_scorecard.json <a href='route_scorecard.md'>证据</a></footer></article><article class='gate warn'><div><span>关注</span><h3>输出实验</h3></div><p>5/5 cases; with-skill 100.0; baseline 0.0; file-backed 1; near-neighbor 1; blind A/B 5; exec 10; command 10; model 0; recorded 0; reviewed 0/5; review pending 5</p><footer>reports/output_quality_scorecard.json <a href='output_quality_scorecard.md'>证据</a></footer></article><article class='gate warn'><div><span>关注</span><h3>上下文</h3></div><p>initial load 960/1000; deferred 436837/120000; top deferred scripts 387626; resource governance needs-review; quality density 135.4</p><footer>reports/context_budget.json <a href='context_budget.md'>证据</a></footer></article><article class='gate pass'><div><span>通过</span><h3>运行矩阵</h3></div><p>5 / 5 targets pass</p><footer>reports/conformance_matrix.json <a href='conformance_matrix.md'>证据</a></footer></article><article class='gate pass'><div><span>通过</span><h3>信任报告</h3></div><p>0 secrets; 111 scripts; 3 network-capable scripts; 0 help smoke failures</p><footer>reports/security_trust_report.json <a href='security_trust_report.md'>证据</a></footer></article><article class='gate pass'><div><span>通过</span><h3>Python 兼容</h3></div><p>Python 3.11; 177 files; 0 compatibility issues; 0 syntax; 0 f-string 3.11 hazards</p><footer>reports/python_compatibility.json <a href='python_compatibility.md'>证据</a></footer></article><article class='gate pass'><div><span>通过</span><h3>架构维护</h3></div><p>174 Python files; 0 hotspots; 8 watchlist files; 0 blockers; largest 899 lines; 64 CLI handlers; 18 in entrypoint</p><footer>reports/architecture_maintainability.json <a href='architecture_maintainability.md'>证据</a></footer></article><article class='gate pass'><div><span>通过</span><h3>权限批准</h3></div><p>3/3 permissions approved; gaps 0; required file_write, network, subprocess</p><footer>reports/security_trust_report.json + security/permission_policy.json <a href='../security/permission_policy.md'>证据</a></footer></article><article class='gate pass'><div><span>通过</span><h3>权限探针</h3></div><p>4/4 targets probed; native 0; metadata fallback 4; installer 4; residual risks 4</p><footer>reports/runtime_permission_probes.json <a href='runtime_permission_probes.md'>证据</a></footer></article><article class='gate pass'><div><span>通过</span><h3>组合治理</h3></div><p>12 skills, 1 actionable; 0 actionable route collisions; 0 actionable owner gaps; 0 actionable stale; 0 actionable drift; 24 scoped non-actionable issues</p><footer>reports/skill_atlas.json <a href='skill_atlas.html'>证据</a></footer></article><article class='gate pass'><div><span>通过</span><h3>运营回路</h3></div><p>1 metadata events; adoption 0; missed 0; bad-output 0; risk low</p><footer>reports/adoption_drift_report.json <a href='adoption_drift_report.md'>证据</a></footer></article><article class='gate warn'><div><span>关注</span><h3>人工批准</h3></div><p>0 active waivers; 2 warning gates still need reviewer decision</p><footer>reports/review_waivers.json <a href='review_waivers.md'>证据</a></footer></article><article class='gate warn'><div><span>关注</span><h3>世界证据</h3></div><p>4 pending world-class evidence entries; 1 human pending; 3 external pending; source checks 6/13 pass; 7 blocked; overclaim guard true</p><footer>reports/world_class_evidence_ledger.json <a href='world_class_evidence_ledger.md'>证据</a></footer></article><article class='gate pass'><div><span>通过</span><h3>注册审计</h3></div><p>yao-meta-skill 1.1.0; 6/6 compatibility entries pass; install pass with 4 adapters; installer permissions 12 enforced / 0 failures</p><footer>reports/registry_audit.json + reports/install_simulation.json <a href='registry_audit.md'>证据</a></footer></article><article class='gate pass'><div><span>通过</span><h3>发布路线</h3></div><p>0 promote; 3 keep current; 0 blocked; upgrade minor declared / minor recommended</p><footer>reports/promotion_decisions.json + reports/upgrade_check.json + docs/migration-v2.md <a href='promotion_decisions.md'>证据</a></footer></article></div>
+      <div class="gates"><article class='gate pass'><div><span>通过</span><h3>意图画布</h3></div><p>intent confidence 100/100; Intent is clear enough to package the first routeable version.</p><footer>reports/intent-confidence.json <a href='intent-confidence.md'>证据</a></footer></article><article class='gate pass'><div><span>通过</span><h3>触发实验</h3></div><p>13 trigger cases; 0 misroutes; 0 ambiguous</p><footer>reports/route_scorecard.json <a href='route_scorecard.md'>证据</a></footer></article><article class='gate warn'><div><span>关注</span><h3>输出实验</h3></div><p>5/5 cases; with-skill 100.0; baseline 0.0; file-backed 1; near-neighbor 1; blind A/B 5; exec 10; command 10; model 0; recorded 0; reviewed 0/5; review pending 5</p><footer>reports/output_quality_scorecard.json <a href='output_quality_scorecard.md'>证据</a></footer></article><article class='gate pass'><div><span>通过</span><h3>上下文</h3></div><p>initial load 960/1000; deferred 436837/120000; top deferred scripts 387626; resource governance governed; quality density 135.4</p><footer>reports/context_budget.json <a href='context_budget.md'>证据</a></footer></article><article class='gate pass'><div><span>通过</span><h3>运行矩阵</h3></div><p>5 / 5 targets pass</p><footer>reports/conformance_matrix.json <a href='conformance_matrix.md'>证据</a></footer></article><article class='gate pass'><div><span>通过</span><h3>信任报告</h3></div><p>0 secrets; 111 scripts; 3 network-capable scripts; 0 help smoke failures</p><footer>reports/security_trust_report.json <a href='security_trust_report.md'>证据</a></footer></article><article class='gate pass'><div><span>通过</span><h3>Python 兼容</h3></div><p>Python 3.11; 177 files; 0 compatibility issues; 0 syntax; 0 f-string 3.11 hazards</p><footer>reports/python_compatibility.json <a href='python_compatibility.md'>证据</a></footer></article><article class='gate pass'><div><span>通过</span><h3>架构维护</h3></div><p>174 Python files; 0 hotspots; 8 watchlist files; 0 blockers; largest 899 lines; 64 CLI handlers; 18 in entrypoint</p><footer>reports/architecture_maintainability.json <a href='architecture_maintainability.md'>证据</a></footer></article><article class='gate pass'><div><span>通过</span><h3>权限批准</h3></div><p>3/3 permissions approved; gaps 0; required file_write, network, subprocess</p><footer>reports/security_trust_report.json + security/permission_policy.json <a href='../security/permission_policy.md'>证据</a></footer></article><article class='gate pass'><div><span>通过</span><h3>权限探针</h3></div><p>4/4 targets probed; native 0; metadata fallback 4; installer 4; residual risks 4</p><footer>reports/runtime_permission_probes.json <a href='runtime_permission_probes.md'>证据</a></footer></article><article class='gate pass'><div><span>通过</span><h3>组合治理</h3></div><p>12 skills, 1 actionable; 0 actionable route collisions; 0 actionable owner gaps; 0 actionable stale; 0 actionable drift; 24 scoped non-actionable issues</p><footer>reports/skill_atlas.json <a href='skill_atlas.html'>证据</a></footer></article><article class='gate pass'><div><span>通过</span><h3>运营回路</h3></div><p>1 metadata events; adoption 0; missed 0; bad-output 0; risk low</p><footer>reports/adoption_drift_report.json <a href='adoption_drift_report.md'>证据</a></footer></article><article class='gate warn'><div><span>关注</span><h3>人工批准</h3></div><p>0 active waivers; 1 warning gates still need reviewer decision</p><footer>reports/review_waivers.json <a href='review_waivers.md'>证据</a></footer></article><article class='gate warn'><div><span>关注</span><h3>世界证据</h3></div><p>4 pending world-class evidence entries; 1 human pending; 3 external pending; source checks 6/13 pass; 7 blocked; overclaim guard true</p><footer>reports/world_class_evidence_ledger.json <a href='world_class_evidence_ledger.md'>证据</a></footer></article><article class='gate pass'><div><span>通过</span><h3>注册审计</h3></div><p>yao-meta-skill 1.1.0; 6/6 compatibility entries pass; install pass with 4 adapters; installer permissions 12 enforced / 0 failures</p><footer>reports/registry_audit.json + reports/install_simulation.json <a href='registry_audit.md'>证据</a></footer></article><article class='gate pass'><div><span>通过</span><h3>发布路线</h3></div><p>0 promote; 3 keep current; 0 blocked; upgrade minor declared / minor recommended</p><footer>reports/promotion_decisions.json + reports/upgrade_check.json + docs/migration-v2.md <a href='promotion_decisions.md'>证据</a></footer></article></div>
     </section>
 
     <div class="twocol">
       <section><h2>阻断事项</h2><p class='muted'>无。</p></section>
-      <section><h2>关注事项</h2><ul class='issues'><li><strong>输出实验</strong><span>5/5 cases; with-skill 100.0; baseline 0.0; file-backed 1; near-neighbor 1; blind A/B 5; exec 10; command 10; model 0; recorded 0; reviewed 0/5; review pending 5</span></li><li><strong>上下文</strong><span>initial load 960/1000; deferred 436837/120000; top deferred scripts 387626; resource governance needs-review; quality density 135.4</span></li><li><strong>人工批准</strong><span>0 active waivers; 2 warning gates still need reviewer decision</span></li><li><strong>世界证据</strong><span>4 pending world-class evidence entries; 1 human pending; 3 external pending; source checks 6/13 pass; 7 blocked; overclaim guard true</span></li></ul></section>
+      <section><h2>关注事项</h2><ul class='issues'><li><strong>输出实验</strong><span>5/5 cases; with-skill 100.0; baseline 0.0; file-backed 1; near-neighbor 1; blind A/B 5; exec 10; command 10; model 0; recorded 0; reviewed 0/5; review pending 5</span></li><li><strong>人工批准</strong><span>0 active waivers; 1 warning gates still need reviewer decision</span></li><li><strong>世界证据</strong><span>4 pending world-class evidence entries; 1 human pending; 3 external pending; source checks 6/13 pass; 7 blocked; overclaim guard true</span></li></ul></section>
     </div>
 
     <section id="actions">
       <h2>修复动作</h2>
-      <div class="actions-grid"><article class='action-card warn'><div><span>关注</span><h3>输出实验</h3></div><p>补足 output eval 覆盖、execution evidence、blind A/B 和 reviewer adjudication。</p><small>没有输出质量和人工盲评证据时，Skill 只能证明会触发，不能证明输出真的更好且经得起审查。</small><dl><dt>修复位置</dt><dd>evals/output/cases.jsonl + reports/output_quality_scorecard.md + reports/output_review_kit.html + reports/output_review_adjudication.md</dd><dt>验证命令</dt><dd><code>python3 scripts/adjudicate_output_review.py --write-template &amp;&amp; python3 scripts/yao.py output-review</code></dd></dl><ul class='source-ref-list'><li><a href='../evals/output/cases.jsonl'>evals/output/cases.jsonl:1</a><small>output eval cases · eval</small></li><li><a href='output_quality_scorecard.md'>reports/output_quality_scorecard.md:1</a><small>output scorecard · report</small></li><li><a href='output_execution_runs.md'>reports/output_execution_runs.md:1</a><small>output execution runs · report</small></li><li><a href='output_blind_review_pack.md'>reports/output_blind_review_pack.md:1</a><small>blind A/B review pack · report</small></li><li><a href='output_review_kit.html'>reports/output_review_kit.html:6</a><small>reviewer cockpit · report</small></li><li><a href='output_review_adjudication.md'>reports/output_review_adjudication.md:1</a><small>review adjudication · report</small></li></ul><footer>reports/output_quality_scorecard.json <a href='output_quality_scorecard.md'>打开证据</a></footer></article><article class='action-card warn'><div><span>关注</span><h3>上下文</h3></div><p>压缩或拆分高成本 deferred resources，保留最小可路由上下文。</p><small>初始加载可以安全，但后续 references、scripts、evals 体量过大时，reviewer 仍需要看到维护和读取成本。</small><dl><dt>修复位置</dt><dd>SKILL.md + references/ + scripts/ + evals/</dd><dt>验证命令</dt><dd><code>python3 scripts/render_context_reports.py</code></dd></dl><ul class='source-ref-list'><li><a href='../SKILL.md'>SKILL.md:9</a><small>entrypoint · source</small></li><li><a href='context_budget.md'>reports/context_budget.md:1</a><small>context budget · report</small></li><li><a href='../scripts/resource_boundary_check.py'>scripts/resource_boundary_check.py:44</a><small>resource boundary checker · source</small></li><li><a href='../references/skill-engineering-method.md'>references/skill-engineering-method.md:208</a><small>skill engineering method · method</small></li></ul><footer>reports/context_budget.json <a href='context_budget.md'>打开证据</a></footer></article><article class='action-card warn'><div><span>关注</span><h3>人工批准</h3></div><p>对保留的 warning 写入 reviewer、理由、范围和到期时间，或修掉 warning。</p><small>warning 可以被接受，但必须可审计、会过期，并且不能掩盖 blocker。</small><dl><dt>修复位置</dt><dd>reports/review_waivers.md</dd><dt>验证命令</dt><dd><code>python3 scripts/render_review_waivers.py .</code></dd></dl><ul class='source-ref-list'><li><a href='review_waivers.md'>reports/review_waivers.md:1</a><small>waiver ledger · report</small></li><li><a href='../references/review-waiver-method.md'>references/review-waiver-method.md:1</a><small>waiver method · method</small></li></ul><footer>reports/review_waivers.json <a href='review_waivers.md'>打开证据</a></footer></article><article class='action-card warn with-evidence'><div><span>关注</span><h3>世界证据</h3></div><p>补齐 provider、真人盲评、原生权限执行和真实客户端遥测证据，或明确本次发布不声明 world-class 完成。</p><small>世界级结论必须来自已接受的外部/人工证据；计划、metadata fallback、待评审和本地命令都不能替代完成证据。</small><dl><dt>修复位置</dt><dd>reports/world_class_operator_runbook.html + reports/world_class_evidence_ledger.md + reports/world_class_evidence_intake.md + reports/world_class_submission_review.md</dd><dt>验证命令</dt><dd><code>python3 scripts/yao.py world-class-runbook . --submissions-dir evidence/world_class/submissions &amp;&amp; python3 scripts/yao.py world-class-ledger . --submissions-dir evidence/world_class/submissions &amp;&amp; python3 scripts/yao.py review-studio .</code></dd></dl><section class='action-evidence-panel'><h4>证据采集</h4><p>以下条目仍需真实外部或人工证据；提交文件、校验命令和阻断检查必须同时闭环。</p><div class='action-evidence-grid'><article class='action-evidence-item pending'><div><span>pending · external</span><h4>Provider Holdout</h4></div><p>model-executed 0; token-observed 0</p><dl><dt>提交</dt><dd><code>evidence/world_class/submissions/provider-holdout.json</code></dd><dt>模板</dt><dd><code>evidence/world_class/templates/provider-holdout.intake.json</code></dd><dt>阻断</dt><dd>2 blocked / 1 pass</dd><dt>下一步</dt><dd>Run provider-backed holdout cases with real credentials and commit only aggregate evidence.</dd></dl><section><h5>阻断检查</h5><ul class='action-evidence-checks'><li class='action-evidence-check blocked'><span>Provider model run</span><code>model_executed_count: 0 / &gt;0</code><small>Run provider-backed output-exec with real credentials.</small></li><li class='action-evidence-check blocked'><span>Token usage observed</span><code>token_observed_count: 0 / &gt;0</code><small>Provider execution should return non-estimated token usage.</small></li></ul></section><details class='action-command-details'><summary>操作命令</summary><ul class='action-command-list'><li><span>准备提交</span><code>python3 scripts/yao.py world-class-submission-kit . --evidence-key provider-holdout --output-dir evidence/world_class/submissions</code></li><li><span>校验入口</span><code>python3 scripts/yao.py world-class-intake . --submissions-dir evidence/world_class/submissions</code></li><li><span>审查提交</span><code>python3 scripts/yao.py world-class-submission-review . --submissions-dir evidence/world_class/submissions</code></li><li><span>刷新台账</span><code>python3 scripts/yao.py world-class-ledger . --submissions-dir evidence/world_class/submissions</code></li></ul></details><details class='action-runbook-details'><summary>首要步骤</summary><ol class='action-runbook-list'><li>YAO_OUTPUT_EVAL_MODEL=gpt-4.1-mini OPENAI_API_KEY=&lt;redacted&gt; python3 scripts/yao.py output-exec --provider-runner openai --timeout-seconds 60</li><li>python3 scripts/yao.py skill-os2-audit . --generated-at &lt;YYYY-MM-DD&gt;</li><li>Copy evidence/world_class/templates/provider-holdout.intake.json to evidence/world_class/submissions/provider-holdout.json and fill only real evidence fields.</li></ol></details></article><article class='action-evidence-item pending'><div><span>pending · human</span><h4>Human Adjudication</h4></div><p>0/5 decisions; pending 5</p><dl><dt>提交</dt><dd><code>evidence/world_class/submissions/human-adjudication.json</code></dd><dt>模板</dt><dd><code>evidence/world_class/templates/human-adjudication.intake.json</code></dd><dt>阻断</dt><dd>2 blocked / 2 pass</dd><dt>下一步</dt><dd>Record real A/B choices in the decision template, then regenerate adjudication.</dd></dl><section><h5>阻断检查</h5><ul class='action-evidence-checks'><li class='action-evidence-check blocked'><span>No pending decisions</span><code>pending_count: 5 / ==0</code><small>Record a reviewer choice for every pair.</small></li><li class='action-evidence-check blocked'><span>Judgments complete</span><code>judgment_count: 0 / ==pair_count</code><small>Every pair needs one valid human judgment.</small></li></ul></section><details class='action-command-details'><summary>操作命令</summary><ul class='action-command-list'><li><span>准备提交</span><code>python3 scripts/yao.py world-class-submission-kit . --evidence-key human-adjudication --output-dir evidence/world_class/submissions</code></li><li><span>校验入口</span><code>python3 scripts/yao.py world-class-intake . --submissions-dir evidence/world_class/submissions</code></li><li><span>审查提交</span><code>python3 scripts/yao.py world-class-submission-review . --submissions-dir evidence/world_class/submissions</code></li><li><span>刷新台账</span><code>python3 scripts/yao.py world-class-ledger . --submissions-dir evidence/world_class/submissions</code></li></ul></details><details class='action-runbook-details'><summary>首要步骤</summary><ol class='action-runbook-list'><li>python3 scripts/yao.py output-review-kit --write-template</li><li>Open reports/output_review_kit.md and choose A or B for each pair without opening the answer key.</li><li>python3 scripts/adjudicate_output_review.py --write-template</li></ol></details></article><article class='action-evidence-item pending'><div><span>pending · external</span><h4>Native Permission Enforcement</h4></div><p>native-enforced targets 0; installer-enforced targets 4</p><dl><dt>提交</dt><dd><code>evidence/world_class/submissions/native-permission-enforcement.json</code></dd><dt>模板</dt><dd><code>evidence/world_class/templates/native-permission-enforcement.intake.json</code></dd><dt>阻断</dt><dd>1 blocked / 2 pass</dd><dt>下一步</dt><dd>Integrate a real target-client or external installer runtime guard before claiming native permission enforcement.</dd></dl><section><h5>阻断检查</h5><ul class='action-evidence-checks'><li class='action-evidence-check blocked'><span>Native enforcement</span><code>native_enforcement_count: 0 / &gt;0</code><small>Collect real target-client or external runtime guard proof.</small></li></ul></section><details class='action-command-details'><summary>操作命令</summary><ul class='action-command-list'><li><span>准备提交</span><code>python3 scripts/yao.py world-class-submission-kit . --evidence-key native-permission-enforcement --output-dir evidence/world_class/submissions</code></li><li><span>校验入口</span><code>python3 scripts/yao.py world-class-intake . --submissions-dir evidence/world_class/submissions</code></li><li><span>审查提交</span><code>python3 scripts/yao.py world-class-submission-review . --submissions-dir evidence/world_class/submissions</code></li><li><span>刷新台账</span><code>python3 scripts/yao.py world-class-ledger . --submissions-dir evidence/world_class/submissions</code></li></ul></details><details class='action-runbook-details'><summary>首要步骤</summary><ol class='action-runbook-list'><li>Implement or connect a real target client or external installer runtime guard that blocks undeclared network, file_write, or subprocess capabilities.</li><li>Update the generated target adapter only when the guard is actually enforced by that target.</li><li>python3 scripts/yao.py package . --platform openai --platform claude --platform generic --platform vscode --output-dir dist --zip</li></ol></details></article><article class='action-evidence-item pending'><div><span>pending · external</span><h4>Native Client Telemetry</h4></div><p>external source events 0; adoption samples 0</p><dl><dt>提交</dt><dd><code>evidence/world_class/submissions/native-client-telemetry.json</code></dd><dt>模板</dt><dd><code>evidence/world_class/templates/native-client-telemetry.intake.json</code></dd><dt>阻断</dt><dd>2 blocked / 1 pass</dd><dt>下一步</dt><dd>Install a real client against the native host and import production metadata-only events.</dd></dl><section><h5>阻断检查</h5><ul class='action-evidence-checks'><li class='action-evidence-check blocked'><span>External events</span><code>external_source_events: 0 / &gt;0</code><small>Import at least one metadata-only event from a real client.</small></li><li class='action-evidence-check blocked'><span>Adoption sample</span><code>adoption_sample_count: 0 / &gt;0</code><small>Telemetry must include adoption outcome evidence.</small></li></ul></section><details class='action-command-details'><summary>操作命令</summary><ul class='action-command-list'><li><span>准备提交</span><code>python3 scripts/yao.py world-class-submission-kit . --evidence-key native-client-telemetry --output-dir evidence/world_class/submissions</code></li><li><span>校验入口</span><code>python3 scripts/yao.py world-class-intake . --submissions-dir evidence/world_class/submissions</code></li><li><span>审查提交</span><code>python3 scripts/yao.py world-class-submission-review . --submissions-dir evidence/world_class/submissions</code></li><li><span>刷新台账</span><code>python3 scripts/yao.py world-class-ledger . --submissions-dir evidence/world_class/submissions</code></li></ul></details><details class='action-runbook-details'><summary>首要步骤</summary><ol class='action-runbook-list'><li>python3 scripts/telemetry_native_host.py . --write-launcher /tmp/yao-telemetry-host.sh --write-manifest /tmp/yao-telemetry-host.json --allowed-origin chrome-extension://&lt;extension-id&gt;/</li><li>Install the generated native messaging manifest for the real client and send at least one accepted skill_activation or skill_output event.</li><li>python3 scripts/yao.py telemetry-import . --input-jsonl .yao/telemetry_spool/external_events.jsonl</li></ol></details></article></div></section><ul class='source-ref-list'><li><a href='world_class_evidence_ledger.md'>reports/world_class_evidence_ledger.md:1</a><small>world-class evidence ledger · report</small></li><li><a href='world_class_evidence_plan.md'>reports/world_class_evidence_plan.md:1</a><small>world-class evidence plan · report</small></li><li><a href='world_class_evidence_intake.md'>reports/world_class_evidence_intake.md:1</a><small>world-class evidence intake · report</small></li><li><a href='world_class_submission_review.md'>reports/world_class_submission_review.md:1</a><small>world-class submission review · report</small></li><li><a href='world_class_claim_guard.md'>reports/world_class_claim_guard.md:1</a><small>world-class claim guard · report</small></li><li><a href='../evidence/world_class/intake.schema.json'>evidence/world_class/intake.schema.json:4</a><small>evidence intake schema · schema</small></li><li><a href='../evidence/world_class/templates/provider-holdout.intake.json'>evidence/world_class/templates/provider-holdout.intake.json:3</a><small>provider intake template · template</small></li><li><a href='../evidence/world_class/templates/human-adjudication.intake.json'>evidence/world_class/templates/human-adjudication.intake.json:3</a><small>human intake template · template</small></li><li><a href='../evidence/world_class/templates/native-permission-enforcement.intake.json'>evidence/world_class/templates/native-permission-enforcement.intake.json:3</a><small>permission intake template · template</small></li><li><a href='../evidence/world_class/templates/native-client-telemetry.intake.json'>evidence/world_class/templates/native-client-telemetry.intake.json:3</a><small>telemetry intake template · template</small></li><li><a href='skill_os2_audit.md'>reports/skill_os2_audit.md:1</a><small>Skill OS 2.0 audit · report</small></li><li><a href='output_review_decisions.json'>reports/output_review_decisions.json:6</a><small>human review decisions · report</small></li><li><a href='runtime_permission_probes.md'>reports/runtime_permission_probes.md:1</a><small>runtime permission probes · report</small></li><li><a href='adoption_drift_report.md'>reports/adoption_drift_report.md:1</a><small>adoption drift · report</small></li></ul><footer>reports/world_class_evidence_ledger.json <a href='world_class_evidence_ledger.md'>打开证据</a></footer></article></div>
+      <div class="actions-grid"><article class='action-card warn'><div><span>关注</span><h3>输出实验</h3></div><p>补足 output eval 覆盖、execution evidence、blind A/B 和 reviewer adjudication。</p><small>没有输出质量和人工盲评证据时，Skill 只能证明会触发，不能证明输出真的更好且经得起审查。</small><dl><dt>修复位置</dt><dd>evals/output/cases.jsonl + reports/output_quality_scorecard.md + reports/output_review_kit.html + reports/output_review_adjudication.md</dd><dt>验证命令</dt><dd><code>python3 scripts/adjudicate_output_review.py --write-template &amp;&amp; python3 scripts/yao.py output-review</code></dd></dl><ul class='source-ref-list'><li><a href='../evals/output/cases.jsonl'>evals/output/cases.jsonl:1</a><small>output eval cases · eval</small></li><li><a href='output_quality_scorecard.md'>reports/output_quality_scorecard.md:1</a><small>output scorecard · report</small></li><li><a href='output_execution_runs.md'>reports/output_execution_runs.md:1</a><small>output execution runs · report</small></li><li><a href='output_blind_review_pack.md'>reports/output_blind_review_pack.md:1</a><small>blind A/B review pack · report</small></li><li><a href='output_review_kit.html'>reports/output_review_kit.html:6</a><small>reviewer cockpit · report</small></li><li><a href='output_review_adjudication.md'>reports/output_review_adjudication.md:1</a><small>review adjudication · report</small></li></ul><footer>reports/output_quality_scorecard.json <a href='output_quality_scorecard.md'>打开证据</a></footer></article><article class='action-card warn'><div><span>关注</span><h3>人工批准</h3></div><p>对保留的 warning 写入 reviewer、理由、范围和到期时间，或修掉 warning。</p><small>warning 可以被接受，但必须可审计、会过期，并且不能掩盖 blocker。</small><dl><dt>修复位置</dt><dd>reports/review_waivers.md</dd><dt>验证命令</dt><dd><code>python3 scripts/render_review_waivers.py .</code></dd></dl><ul class='source-ref-list'><li><a href='review_waivers.md'>reports/review_waivers.md:1</a><small>waiver ledger · report</small></li><li><a href='../references/review-waiver-method.md'>references/review-waiver-method.md:1</a><small>waiver method · method</small></li></ul><footer>reports/review_waivers.json <a href='review_waivers.md'>打开证据</a></footer></article><article class='action-card warn with-evidence'><div><span>关注</span><h3>世界证据</h3></div><p>补齐 provider、真人盲评、原生权限执行和真实客户端遥测证据，或明确本次发布不声明 world-class 完成。</p><small>世界级结论必须来自已接受的外部/人工证据；计划、metadata fallback、待评审和本地命令都不能替代完成证据。</small><dl><dt>修复位置</dt><dd>reports/world_class_operator_runbook.html + reports/world_class_evidence_ledger.md + reports/world_class_evidence_intake.md + reports/world_class_submission_review.md</dd><dt>验证命令</dt><dd><code>python3 scripts/yao.py world-class-runbook . --submissions-dir evidence/world_class/submissions &amp;&amp; python3 scripts/yao.py world-class-ledger . --submissions-dir evidence/world_class/submissions &amp;&amp; python3 scripts/yao.py review-studio .</code></dd></dl><section class='action-evidence-panel'><h4>证据采集</h4><p>以下条目仍需真实外部或人工证据；提交文件、校验命令和阻断检查必须同时闭环。</p><div class='action-evidence-grid'><article class='action-evidence-item pending'><div><span>pending · external</span><h4>Provider Holdout</h4></div><p>model-executed 0; token-observed 0</p><dl><dt>提交</dt><dd><code>evidence/world_class/submissions/provider-holdout.json</code></dd><dt>模板</dt><dd><code>evidence/world_class/templates/provider-holdout.intake.json</code></dd><dt>阻断</dt><dd>2 blocked / 1 pass</dd><dt>下一步</dt><dd>Run provider-backed holdout cases with real credentials and commit only aggregate evidence.</dd></dl><section><h5>阻断检查</h5><ul class='action-evidence-checks'><li class='action-evidence-check blocked'><span>Provider model run</span><code>model_executed_count: 0 / &gt;0</code><small>Run provider-backed output-exec with real credentials.</small></li><li class='action-evidence-check blocked'><span>Token usage observed</span><code>token_observed_count: 0 / &gt;0</code><small>Provider execution should return non-estimated token usage.</small></li></ul></section><details class='action-command-details'><summary>操作命令</summary><ul class='action-command-list'><li><span>准备提交</span><code>python3 scripts/yao.py world-class-submission-kit . --evidence-key provider-holdout --output-dir evidence/world_class/submissions</code></li><li><span>校验入口</span><code>python3 scripts/yao.py world-class-intake . --submissions-dir evidence/world_class/submissions</code></li><li><span>审查提交</span><code>python3 scripts/yao.py world-class-submission-review . --submissions-dir evidence/world_class/submissions</code></li><li><span>刷新台账</span><code>python3 scripts/yao.py world-class-ledger . --submissions-dir evidence/world_class/submissions</code></li></ul></details><details class='action-runbook-details'><summary>首要步骤</summary><ol class='action-runbook-list'><li>YAO_OUTPUT_EVAL_MODEL=gpt-4.1-mini OPENAI_API_KEY=&lt;redacted&gt; python3 scripts/yao.py output-exec --provider-runner openai --timeout-seconds 60</li><li>python3 scripts/yao.py skill-os2-audit . --generated-at &lt;YYYY-MM-DD&gt;</li><li>Copy evidence/world_class/templates/provider-holdout.intake.json to evidence/world_class/submissions/provider-holdout.json and fill only real evidence fields.</li></ol></details></article><article class='action-evidence-item pending'><div><span>pending · human</span><h4>Human Adjudication</h4></div><p>0/5 decisions; pending 5</p><dl><dt>提交</dt><dd><code>evidence/world_class/submissions/human-adjudication.json</code></dd><dt>模板</dt><dd><code>evidence/world_class/templates/human-adjudication.intake.json</code></dd><dt>阻断</dt><dd>2 blocked / 2 pass</dd><dt>下一步</dt><dd>Record real A/B choices in the decision template, then regenerate adjudication.</dd></dl><section><h5>阻断检查</h5><ul class='action-evidence-checks'><li class='action-evidence-check blocked'><span>No pending decisions</span><code>pending_count: 5 / ==0</code><small>Record a reviewer choice for every pair.</small></li><li class='action-evidence-check blocked'><span>Judgments complete</span><code>judgment_count: 0 / ==pair_count</code><small>Every pair needs one valid human judgment.</small></li></ul></section><details class='action-command-details'><summary>操作命令</summary><ul class='action-command-list'><li><span>准备提交</span><code>python3 scripts/yao.py world-class-submission-kit . --evidence-key human-adjudication --output-dir evidence/world_class/submissions</code></li><li><span>校验入口</span><code>python3 scripts/yao.py world-class-intake . --submissions-dir evidence/world_class/submissions</code></li><li><span>审查提交</span><code>python3 scripts/yao.py world-class-submission-review . --submissions-dir evidence/world_class/submissions</code></li><li><span>刷新台账</span><code>python3 scripts/yao.py world-class-ledger . --submissions-dir evidence/world_class/submissions</code></li></ul></details><details class='action-runbook-details'><summary>首要步骤</summary><ol class='action-runbook-list'><li>python3 scripts/yao.py output-review-kit --write-template</li><li>Open reports/output_review_kit.md and choose A or B for each pair without opening the answer key.</li><li>python3 scripts/adjudicate_output_review.py --write-template</li></ol></details></article><article class='action-evidence-item pending'><div><span>pending · external</span><h4>Native Permission Enforcement</h4></div><p>native-enforced targets 0; installer-enforced targets 4</p><dl><dt>提交</dt><dd><code>evidence/world_class/submissions/native-permission-enforcement.json</code></dd><dt>模板</dt><dd><code>evidence/world_class/templates/native-permission-enforcement.intake.json</code></dd><dt>阻断</dt><dd>1 blocked / 2 pass</dd><dt>下一步</dt><dd>Integrate a real target-client or external installer runtime guard before claiming native permission enforcement.</dd></dl><section><h5>阻断检查</h5><ul class='action-evidence-checks'><li class='action-evidence-check blocked'><span>Native enforcement</span><code>native_enforcement_count: 0 / &gt;0</code><small>Collect real target-client or external runtime guard proof.</small></li></ul></section><details class='action-command-details'><summary>操作命令</summary><ul class='action-command-list'><li><span>准备提交</span><code>python3 scripts/yao.py world-class-submission-kit . --evidence-key native-permission-enforcement --output-dir evidence/world_class/submissions</code></li><li><span>校验入口</span><code>python3 scripts/yao.py world-class-intake . --submissions-dir evidence/world_class/submissions</code></li><li><span>审查提交</span><code>python3 scripts/yao.py world-class-submission-review . --submissions-dir evidence/world_class/submissions</code></li><li><span>刷新台账</span><code>python3 scripts/yao.py world-class-ledger . --submissions-dir evidence/world_class/submissions</code></li></ul></details><details class='action-runbook-details'><summary>首要步骤</summary><ol class='action-runbook-list'><li>Implement or connect a real target client or external installer runtime guard that blocks undeclared network, file_write, or subprocess capabilities.</li><li>Update the generated target adapter only when the guard is actually enforced by that target.</li><li>python3 scripts/yao.py package . --platform openai --platform claude --platform generic --platform vscode --output-dir dist --zip</li></ol></details></article><article class='action-evidence-item pending'><div><span>pending · external</span><h4>Native Client Telemetry</h4></div><p>external source events 0; adoption samples 0</p><dl><dt>提交</dt><dd><code>evidence/world_class/submissions/native-client-telemetry.json</code></dd><dt>模板</dt><dd><code>evidence/world_class/templates/native-client-telemetry.intake.json</code></dd><dt>阻断</dt><dd>2 blocked / 1 pass</dd><dt>下一步</dt><dd>Install a real client against the native host and import production metadata-only events.</dd></dl><section><h5>阻断检查</h5><ul class='action-evidence-checks'><li class='action-evidence-check blocked'><span>External events</span><code>external_source_events: 0 / &gt;0</code><small>Import at least one metadata-only event from a real client.</small></li><li class='action-evidence-check blocked'><span>Adoption sample</span><code>adoption_sample_count: 0 / &gt;0</code><small>Telemetry must include adoption outcome evidence.</small></li></ul></section><details class='action-command-details'><summary>操作命令</summary><ul class='action-command-list'><li><span>准备提交</span><code>python3 scripts/yao.py world-class-submission-kit . --evidence-key native-client-telemetry --output-dir evidence/world_class/submissions</code></li><li><span>校验入口</span><code>python3 scripts/yao.py world-class-intake . --submissions-dir evidence/world_class/submissions</code></li><li><span>审查提交</span><code>python3 scripts/yao.py world-class-submission-review . --submissions-dir evidence/world_class/submissions</code></li><li><span>刷新台账</span><code>python3 scripts/yao.py world-class-ledger . --submissions-dir evidence/world_class/submissions</code></li></ul></details><details class='action-runbook-details'><summary>首要步骤</summary><ol class='action-runbook-list'><li>python3 scripts/telemetry_native_host.py . --write-launcher /tmp/yao-telemetry-host.sh --write-manifest /tmp/yao-telemetry-host.json --allowed-origin chrome-extension://&lt;extension-id&gt;/</li><li>Install the generated native messaging manifest for the real client and send at least one accepted skill_activation or skill_output event.</li><li>python3 scripts/yao.py telemetry-import . --input-jsonl .yao/telemetry_spool/external_events.jsonl</li></ol></details></article></div></section><ul class='source-ref-list'><li><a href='world_class_evidence_ledger.md'>reports/world_class_evidence_ledger.md:1</a><small>world-class evidence ledger · report</small></li><li><a href='world_class_evidence_plan.md'>reports/world_class_evidence_plan.md:1</a><small>world-class evidence plan · report</small></li><li><a href='world_class_evidence_intake.md'>reports/world_class_evidence_intake.md:1</a><small>world-class evidence intake · report</small></li><li><a href='world_class_submission_review.md'>reports/world_class_submission_review.md:1</a><small>world-class submission review · report</small></li><li><a href='world_class_claim_guard.md'>reports/world_class_claim_guard.md:1</a><small>world-class claim guard · report</small></li><li><a href='../evidence/world_class/intake.schema.json'>evidence/world_class/intake.schema.json:4</a><small>evidence intake schema · schema</small></li><li><a href='../evidence/world_class/templates/provider-holdout.intake.json'>evidence/world_class/templates/provider-holdout.intake.json:3</a><small>provider intake template · template</small></li><li><a href='../evidence/world_class/templates/human-adjudication.intake.json'>evidence/world_class/templates/human-adjudication.intake.json:3</a><small>human intake template · template</small></li><li><a href='../evidence/world_class/templates/native-permission-enforcement.intake.json'>evidence/world_class/templates/native-permission-enforcement.intake.json:3</a><small>permission intake template · template</small></li><li><a href='../evidence/world_class/templates/native-client-telemetry.intake.json'>evidence/world_class/templates/native-client-telemetry.intake.json:3</a><small>telemetry intake template · template</small></li><li><a href='skill_os2_audit.md'>reports/skill_os2_audit.md:1</a><small>Skill OS 2.0 audit · report</small></li><li><a href='output_review_decisions.json'>reports/output_review_decisions.json:6</a><small>human review decisions · report</small></li><li><a href='runtime_permission_probes.md'>reports/runtime_permission_probes.md:1</a><small>runtime permission probes · report</small></li><li><a href='adoption_drift_report.md'>reports/adoption_drift_report.md:1</a><small>adoption drift · report</small></li></ul><footer>reports/world_class_evidence_ledger.json <a href='world_class_evidence_ledger.md'>打开证据</a></footer></article></div>
     </section>
 
     <section id="annotations">
@@ -743,7 +743,7 @@
     </section>
 
     <section class="twocol">
-      <div class="panel"><h2>上下文</h2><p>initial load 960/1000; deferred 436837/120000; top deferred scripts 387626; resource governance needs-review; quality density 135.4</p></div>
+      <div class="panel"><h2>上下文</h2><p>initial load 960/1000; deferred 436837/120000; top deferred scripts 387626; resource governance governed; quality density 135.4</p></div>
       <div class="panel"><h2>编译证据</h2><p>Review reports/compiled_targets.md before packaging to inspect target adapter modes, generated files, preserved semantics, warnings, and unsupported features.</p></div>
     </section>
 
@@ -781,7 +781,7 @@
       <h2>人工批准</h2>
       <p class="muted">warning 可以被有边界地接受，但必须写入 reviewer、理由、范围和到期时间；blocker 与 world-class 完成证据不能通过 waiver 变成通过。</p>
       <div class="twocol waiver-summary">
-        <div class="panel"><h2>批准概况</h2><p>0 active waivers; 2 warning gates still need reviewer decision</p></div>
+        <div class="panel"><h2>批准概况</h2><p>0 active waivers; 1 warning gates still need reviewer decision</p></div>
         <div class="panel"><h2>批准台账</h2><dl class='kv-grid'><div><dt>Waiver Count</dt><dd>0</dd></div><div><dt>Active Count</dt><dd>0</dd></div><div><dt>Expired Count</dt><dd>0</dd></div><div><dt>Invalid Count</dt><dd>0</dd></div><div><dt>覆盖 Gate</dt><dd>0</dd></div></dl></div>
       </div>
       <h3 class="section-subtitle">批准候选</h3>
@@ -806,8 +806,8 @@
     </section>
 
     <section class="twocol">
-      <div class="panel"><h2>公开声明</h2><dl class='kv-grid'><div><dt>本地复现</dt><dd>是</dd></div><div><dt>发布锁</dt><dd>否</dd></div><div><dt>可公开声明</dt><dd>否</dd></div><div><dt>声明阻断</dt><dd>5</dd></div><div><dt>Provider 证据</dt><dd>否</dd></div><div><dt>人审完成</dt><dd>否</dd></div><div><dt>世界级就绪</dt><dd>否</dd></div></dl></div>
-      <div class="panel"><h2>声明阻断</h2><ul class='issues'><li><strong>阻断</strong><span>release lock is not clean or commit is unavailable</span></li><li><strong>阻断</strong><span>provider-backed model holdout evidence is incomplete</span></li><li><strong>阻断</strong><span>human blind-review adjudication is incomplete</span></li><li><strong>阻断</strong><span>world-class evidence is not accepted yet (4 open gaps, 4 ledger pending)</span></li><li><strong>阻断</strong><span>world-class source checks are not all accepted (6/13 pass, 7 blocked)</span></li></ul></div>
+      <div class="panel"><h2>公开声明</h2><dl class='kv-grid'><div><dt>本地复现</dt><dd>是</dd></div><div><dt>发布锁</dt><dd>是</dd></div><div><dt>可公开声明</dt><dd>否</dd></div><div><dt>声明阻断</dt><dd>4</dd></div><div><dt>Provider 证据</dt><dd>否</dd></div><div><dt>人审完成</dt><dd>否</dd></div><div><dt>世界级就绪</dt><dd>否</dd></div></dl></div>
+      <div class="panel"><h2>声明阻断</h2><ul class='issues'><li><strong>阻断</strong><span>provider-backed model holdout evidence is incomplete</span></li><li><strong>阻断</strong><span>human blind-review adjudication is incomplete</span></li><li><strong>阻断</strong><span>world-class evidence is not accepted yet (4 open gaps, 4 ledger pending)</span></li><li><strong>阻断</strong><span>world-class source checks are not all accepted (6/13 pass, 7 blocked)</span></li></ul></div>
     </section>
 
     <section class="twocol">
diff --git a/reports/review-studio.json b/reports/review-studio.json
index 931d7eb2..f4a28249 100644
--- a/reports/review-studio.json
+++ b/reports/review-studio.json
@@ -4,11 +4,11 @@
   "skill_dir": ".",
   "summary": {
     "decision": "review",
-    "world_class_score": 89,
+    "world_class_score": 91,
     "gate_count": 16,
     "blocker_count": 0,
-    "warning_count": 4,
-    "action_count": 4,
+    "warning_count": 3,
+    "action_count": 3,
     "annotation_count": 0,
     "open_annotation_count": 0,
     "open_annotation_blocker_count": 0,
@@ -42,8 +42,8 @@
     {
       "key": "context-budget",
       "label": "上下文",
-      "status": "warn",
-      "detail": "initial load 960/1000; deferred 436837/120000; top deferred scripts 387626; resource governance needs-review; quality density 135.4",
+      "status": "pass",
+      "detail": "initial load 960/1000; deferred 436837/120000; top deferred scripts 387626; resource governance governed; quality density 135.4",
       "evidence": "reports/context_budget.json",
       "link": "context_budget.md"
     },
@@ -115,7 +115,7 @@
       "key": "review-waivers",
       "label": "人工批准",
       "status": "warn",
-      "detail": "0 active waivers; 2 warning gates still need reviewer decision",
+      "detail": "0 active waivers; 1 warning gates still need reviewer decision",
       "evidence": "reports/review_waivers.json",
       "link": "review_waivers.md"
     },
@@ -154,19 +154,11 @@
       "evidence": "reports/output_quality_scorecard.json",
       "link": "output_quality_scorecard.md"
     },
-    {
-      "key": "context-budget",
-      "label": "上下文",
-      "status": "warn",
-      "detail": "initial load 960/1000; deferred 436837/120000; top deferred scripts 387626; resource governance needs-review; quality density 135.4",
-      "evidence": "reports/context_budget.json",
-      "link": "context_budget.md"
-    },
     {
       "key": "review-waivers",
       "label": "人工批准",
       "status": "warn",
-      "detail": "0 active waivers; 2 warning gates still need reviewer decision",
+      "detail": "0 active waivers; 1 warning gates still need reviewer decision",
       "evidence": "reports/review_waivers.json",
       "link": "review_waivers.md"
     },
@@ -243,53 +235,6 @@
       "verification_command": "python3 scripts/adjudicate_output_review.py --write-template && python3 scripts/yao.py output-review",
       "evidence_steps": []
     },
-    {
-      "gate_key": "context-budget",
-      "label": "上下文",
-      "status": "warn",
-      "priority": "warning",
-      "summary": "压缩或拆分高成本 deferred resources，保留最小可路由上下文。",
-      "why": "初始加载可以安全，但后续 references、scripts、evals 体量过大时，reviewer 仍需要看到维护和读取成本。",
-      "source_fix": "SKILL.md + references/ + scripts/ + evals/",
-      "source_refs": [
-        {
-          "path": "SKILL.md",
-          "label": "entrypoint",
-          "kind": "source",
-          "line": 9,
-          "exists": true,
-          "link": "../SKILL.md"
-        },
-        {
-          "path": "reports/context_budget.md",
-          "label": "context budget",
-          "kind": "report",
-          "line": 1,
-          "exists": true,
-          "link": "context_budget.md"
-        },
-        {
-          "path": "scripts/resource_boundary_check.py",
-          "label": "resource boundary checker",
-          "kind": "source",
-          "line": 44,
-          "exists": true,
-          "link": "../scripts/resource_boundary_check.py"
-        },
-        {
-          "path": "references/skill-engineering-method.md",
-          "label": "skill engineering method",
-          "kind": "method",
-          "line": 208,
-          "exists": true,
-          "link": "../references/skill-engineering-method.md"
-        }
-      ],
-      "evidence": "reports/context_budget.json",
-      "evidence_link": "context_budget.md",
-      "verification_command": "python3 scripts/render_context_reports.py",
-      "evidence_steps": []
-    },
     {
       "gate_key": "review-waivers",
       "label": "人工批准",
@@ -1731,13 +1676,13 @@
         "ok": true,
         "summary": {
           "reproducibility_ready": true,
-          "release_lock_ready": false,
+          "release_lock_ready": true,
           "methodology_complete": true,
           "required_artifact_count": 24,
           "missing_artifact_count": 0,
-          "evidence_bundle_sha256": "d4d58b9557d03656a800aad0acee41d5434db006ebbbff2645fbbab80962cc27",
-          "source_contract_sha256": "f48a71e36bdbcaa353bd4e6bfc499bda41f047aeffe3d8eae2f99509ec693086",
-          "archive_sha256": "00151114ce7fee627882964137dd5259583f5521c75c6570789345e874454137",
+          "evidence_bundle_sha256": "5ad19434d41f67e6a8dbcbccbd22ffefed11008d75ebb00a78f0d24f3f8910a8",
+          "source_contract_sha256": "9fde13846a41c2d1442b6e80350313246dedebe02590ce1f2f07f2da1109c9ea",
+          "archive_sha256": "83ece575ef740a97b6a398f77ed4bc9f1869cde756257c64fdd90d8785c6f3a5",
           "output_case_count": 5,
           "failure_disclosure_count": 3,
           "command_count": 22,
@@ -1755,11 +1700,11 @@
           "world_class_source_pass_count": 6,
           "world_class_source_blocked_count": 7,
           "public_claim_ready": false,
-          "public_claim_blocker_count": 5,
-          "working_tree_dirty": true,
-          "changed_file_count": 20
+          "public_claim_blocker_count": 4,
+          "working_tree_dirty": false,
+          "changed_file_count": 0
         },
-        "commit": "391821240ba14f088d0fbe12dd5f1fdca7917a39",
+        "commit": "a7960d4cfd2d9a17cbaebcee7b823f0db13df0cf",
         "missing_artifacts": [],
         "limitations": [
           "The git commit and dirty flag are generation-time context; the evidence bundle hash is the durable artifact anchor inside a committed report.",
@@ -4401,7 +4346,7 @@
           "execution_mode": "command",
           "model_executed": false,
           "command_executed": true,
-          "duration_ms": 26.59,
+          "duration_ms": 26.79,
           "provider": "local-output-eval-runner",
           "model": "",
           "usage": {
@@ -4429,7 +4374,7 @@
           "execution_mode": "command",
           "model_executed": false,
           "command_executed": true,
-          "duration_ms": 26.1,
+          "duration_ms": 27.31,
           "provider": "local-output-eval-runner",
           "model": "",
           "usage": {
@@ -4452,7 +4397,7 @@
           "execution_mode": "command",
           "model_executed": false,
           "command_executed": true,
-          "duration_ms": 26.12,
+          "duration_ms": 26.66,
           "provider": "local-output-eval-runner",
           "model": "",
           "usage": {
@@ -4480,7 +4425,7 @@
           "execution_mode": "command",
           "model_executed": false,
           "command_executed": true,
-          "duration_ms": 28.36,
+          "duration_ms": 26.21,
           "provider": "local-output-eval-runner",
           "model": "",
           "usage": {
@@ -4503,7 +4448,7 @@
           "execution_mode": "command",
           "model_executed": false,
           "command_executed": true,
-          "duration_ms": 29.38,
+          "duration_ms": 26.62,
           "provider": "local-output-eval-runner",
           "model": "",
           "usage": {
@@ -4531,7 +4476,7 @@
           "execution_mode": "command",
           "model_executed": false,
           "command_executed": true,
-          "duration_ms": 28.48,
+          "duration_ms": 26.34,
           "provider": "local-output-eval-runner",
           "model": "",
           "usage": {
@@ -4581,7 +4526,7 @@
           "execution_mode": "command",
           "model_executed": false,
           "command_executed": true,
-          "duration_ms": 25.88,
+          "duration_ms": 26.37,
           "provider": "local-output-eval-runner",
           "model": "",
           "usage": {
@@ -4604,7 +4549,7 @@
           "execution_mode": "command",
           "model_executed": false,
           "command_executed": true,
-          "duration_ms": 25.99,
+          "duration_ms": 26.44,
           "provider": "local-output-eval-runner",
           "model": "",
           "usage": {
@@ -4633,7 +4578,7 @@
           "execution_mode": "command",
           "model_executed": false,
           "command_executed": true,
-          "duration_ms": 26.23,
+          "duration_ms": 26.21,
           "provider": "local-output-eval-runner",
           "model": "",
           "usage": {
@@ -5354,36 +5299,23 @@
       "ok": true,
       "generated_at": "2026-06-16",
       "skill_dir": ".",
-      "commit": "391821240ba14f088d0fbe12dd5f1fdca7917a39",
+      "commit": "a7960d4cfd2d9a17cbaebcee7b823f0db13df0cf",
       "git_status": {
         "available": true,
-        "dirty": true,
-        "changed_file_count": 20,
-        "sample": [
-          " M reports/adoption_drift_report.json",
-          " M reports/architecture_maintainability.json",
-          " M reports/architecture_maintainability.md",
-          " M reports/benchmark_reproducibility.json",
-          " M reports/benchmark_reproducibility.md",
-          " M reports/context_budget.json",
-          " M reports/context_budget.md",
-          " M reports/context_budget_summary.json",
-          " M reports/output_execution_runs.json",
-          " M reports/output_execution_runs.md",
-          " M reports/python_compatibility.json",
-          " M reports/python_compatibility.md"
-        ],
+        "dirty": false,
+        "changed_file_count": 0,
+        "sample": [],
         "scope": "generation-time status before this report is written"
       },
       "summary": {
         "reproducibility_ready": true,
-        "release_lock_ready": false,
+        "release_lock_ready": true,
         "methodology_complete": true,
         "required_artifact_count": 24,
         "missing_artifact_count": 0,
-        "evidence_bundle_sha256": "d4d58b9557d03656a800aad0acee41d5434db006ebbbff2645fbbab80962cc27",
-        "source_contract_sha256": "f48a71e36bdbcaa353bd4e6bfc499bda41f047aeffe3d8eae2f99509ec693086",
-        "archive_sha256": "00151114ce7fee627882964137dd5259583f5521c75c6570789345e874454137",
+        "evidence_bundle_sha256": "5ad19434d41f67e6a8dbcbccbd22ffefed11008d75ebb00a78f0d24f3f8910a8",
+        "source_contract_sha256": "9fde13846a41c2d1442b6e80350313246dedebe02590ce1f2f07f2da1109c9ea",
+        "archive_sha256": "83ece575ef740a97b6a398f77ed4bc9f1869cde756257c64fdd90d8785c6f3a5",
         "output_case_count": 5,
         "failure_disclosure_count": 3,
         "command_count": 22,
@@ -5401,15 +5333,14 @@
         "world_class_source_pass_count": 6,
         "world_class_source_blocked_count": 7,
         "public_claim_ready": false,
-        "public_claim_blocker_count": 5,
-        "working_tree_dirty": true,
-        "changed_file_count": 20
+        "public_claim_blocker_count": 4,
+        "working_tree_dirty": false,
+        "changed_file_count": 0
       },
       "public_claim": {
         "ready": false,
         "scope": "public benchmark or world-class readiness claim",
         "blockers": [
-          "release lock is not clean or commit is unavailable",
           "provider-backed model holdout evidence is incomplete",
           "human blind-review adjudication is incomplete",
           "world-class evidence is not accepted yet (4 open gaps, 4 ledger pending)",
@@ -5418,10 +5349,10 @@
         "policy": "Local reproducibility can pass before public claims; public claims require provider evidence, human adjudication, clean release lock, accepted world-class evidence, and complete source checks."
       },
       "release_lock": {
-        "ready": false,
-        "commit": "391821240ba14f088d0fbe12dd5f1fdca7917a39",
+        "ready": true,
+        "commit": "a7960d4cfd2d9a17cbaebcee7b823f0db13df0cf",
         "status_scope": "generation-time status before this report is written",
-        "reason": "working tree was dirty at generation time"
+        "reason": "clean generation-time HEAD"
       },
       "evidence_bundle": {
         "algorithm": "sha256(path,label,exists,artifact_sha256)",
@@ -5429,7 +5360,7 @@
         "existing_count": 24,
         "missing_count": 0,
         "missing_paths": [],
-        "sha256": "d4d58b9557d03656a800aad0acee41d5434db006ebbbff2645fbbab80962cc27"
+        "sha256": "5ad19434d41f67e6a8dbcbccbd22ffefed11008d75ebb00a78f0d24f3f8910a8"
       },
       "methodology": {
         "path": "reports/benchmark_methodology.md",
@@ -5503,7 +5434,7 @@
           "path": "reports/output_execution_runs.json",
           "exists": true,
           "bytes": 7967,
-          "sha256": "2ee3f1028f48673cddeb6242eb30aee3db1eccf26c44499744f3031d6d3520d7"
+          "sha256": "56430f01ba194dc6dc793eb505f92fec7a8c46e1be4b71976a3d641c9b98c7ec"
         },
         {
           "label": "blind_review",
@@ -5537,8 +5468,8 @@
           "label": "trust_report",
           "path": "reports/security_trust_report.json",
           "exists": true,
-          "bytes": 109099,
-          "sha256": "491354ae881f6682687a62a98c9a4988bbeef931b24565fdf43e6f3a206fd449"
+          "bytes": 109726,
+          "sha256": "a3a6d4e6809ea0b78b8012e37ad68903a54876a06100434efad38ea6bfe9b2bd"
         },
         {
           "label": "python_compatibility",
@@ -5552,21 +5483,21 @@
           "path": "reports/registry_audit.json",
           "exists": true,
           "bytes": 3183,
-          "sha256": "870c455a5cd209ff8d9e8793861da074e668083d265230d04e38e8ee0ff1323a"
+          "sha256": "d6b523cc5cbb2d69c35989913e7e68dd4dc2bb04d3bcff88bbcb56fbe8be5084"
         },
         {
           "label": "package_verification",
           "path": "reports/package_verification.json",
           "exists": true,
           "bytes": 19338,
-          "sha256": "c4acb824482d992e0e4339b7eeeb8d23d49da12d3414606222f4916c3ed09df6"
+          "sha256": "ae8ac3872be3299b3d2dc18a29128346a3ecbb579987b1b08156e305237b5119"
         },
         {
           "label": "install_simulation",
           "path": "reports/install_simulation.json",
           "exists": true,
           "bytes": 8604,
-          "sha256": "28ceb014e202269bb96628391d056234794b3570ff25de0ef9bdfd7f755fd85f"
+          "sha256": "3e27a6f7678efe9d94528f811a9d3acf20cf41d6328ad5b5fa0be5714c6396a0"
         },
         {
           "label": "skill_os2_audit",
@@ -6008,7 +5939,7 @@
           "label": "Review Studio",
           "status": "pass",
           "objective": "One HTML page supports first-pass production review across trigger, output, runtime, trust, release, and evidence actions.",
-          "current": "16 gates; decision review; warnings 3",
+          "current": "16 gates; decision review; warnings 4",
           "command": "python3 scripts/yao.py review-studio .",
           "test": "python3 tests/verify_review_studio.py",
           "evidence": [
@@ -15488,16 +15419,14 @@
     "context_budget": {
       "ok": true,
       "failures": [],
-      "warnings": [
-        "Deferred resource footprint is high: 436837 estimated tokens across references/scripts/evals. Keep Review Studio warnings visible until the largest resource dirs are split, archived, or justified."
-      ],
+      "warnings": [],
       "stats": {
         "context_budget_tier": "production",
         "context_budget_limit": 1000,
         "skill_body_tokens": 767,
-        "other_text_tokens": 1376303,
+        "other_text_tokens": 1377236,
         "estimated_initial_load_tokens": 960,
-        "estimated_total_text_tokens": 1377070,
+        "estimated_total_text_tokens": 1378003,
         "deferred_resource_tokens": 436837,
         "deferred_resource_warn_threshold": 120000,
         "deferred_resource_dirs": [
@@ -15525,31 +15454,30 @@
           }
         ],
         "deferred_resource_governance": {
-          "status": "needs-review",
+          "status": "governed",
           "large_dir_count": 1,
-          "governed_large_dir_count": 0,
+          "governed_large_dir_count": 1,
           "directories": [
             {
-              "status": "needs-review",
+              "status": "governed",
               "evidence": [
                 "reports/security_trust_report.json",
                 "reports/architecture_maintainability.json",
                 "reports/python_compatibility.json"
               ],
               "reasons": [
+                "trust report covers scripts",
                 "architecture report has no script hotspots or blockers",
                 "Python compatibility report has no issues"
               ],
-              "missing": [
-                "trust report covers scripts"
-              ],
+              "missing": [],
               "path": "scripts",
               "estimated_tokens": 387626,
               "file_count": 111,
               "rationale": "Script resources are deterministic deferred tools, not initial-load prompt context."
             }
           ],
-          "summary": "One or more large deferred resource directories still need explicit governance evidence."
+          "summary": "Large deferred resources are indexed and backed by evidence."
         },
         "relevant_file_count": 565,
         "unused_resource_dirs": [],
@@ -17419,7 +17347,7 @@
     "adoption_drift": {
       "ok": true,
       "schema_version": "2.0",
-      "generated_at": "2026-06-15T17:03:40Z",
+      "generated_at": "2026-06-15T17:08:05Z",
       "skill_dir": ".",
       "privacy_contract": {
         "storage": "local-first",
diff --git a/reports/review-viewer.json b/reports/review-viewer.json
index c48a2612..e18272db 100644
--- a/reports/review-viewer.json
+++ b/reports/review-viewer.json
@@ -996,7 +996,7 @@
       "ok": true,
       "summary": {
         "reproducibility_ready": true,
-        "release_lock_ready": false,
+        "release_lock_ready": true,
         "methodology_complete": true,
         "required_artifact_count": 24,
         "missing_artifact_count": 0,
@@ -1020,11 +1020,11 @@
         "world_class_source_pass_count": 6,
         "world_class_source_blocked_count": 7,
         "public_claim_ready": false,
-        "public_claim_blocker_count": 5,
-        "working_tree_dirty": true,
-        "changed_file_count": 36
+        "public_claim_blocker_count": 4,
+        "working_tree_dirty": false,
+        "changed_file_count": 0
       },
-      "commit": "391821240ba14f088d0fbe12dd5f1fdca7917a39",
+      "commit": "a7960d4cfd2d9a17cbaebcee7b823f0db13df0cf",
       "missing_artifacts": [],
       "limitations": [
         "The git commit and dirty flag are generation-time context; the evidence bundle hash is the durable artifact anchor inside a committed report.",
diff --git a/reports/skill-interpretation.json b/reports/skill-interpretation.json
index a3a7a072..40282d39 100644
--- a/reports/skill-interpretation.json
+++ b/reports/skill-interpretation.json
@@ -1026,9 +1026,9 @@
       "public_claim_ready": false,
       "public_claim_blocker_count": 5,
       "working_tree_dirty": true,
-      "changed_file_count": 36
+      "changed_file_count": 9
     },
-    "commit": "391821240ba14f088d0fbe12dd5f1fdca7917a39",
+    "commit": "a7960d4cfd2d9a17cbaebcee7b823f0db13df0cf",
     "missing_artifacts": [],
     "limitations": [
       "The git commit and dirty flag are generation-time context; the evidence bundle hash is the durable artifact anchor inside a committed report.",
diff --git a/reports/skill-os-2-review.md b/reports/skill-os-2-review.md
index 71d9e581..f4d46109 100644
--- a/reports/skill-os-2-review.md
+++ b/reports/skill-os-2-review.md
@@ -37,7 +37,7 @@ Yao Meta Skill is no longer only a Meta Skill factory. The current working tree
 - Local Install Sync Preflight v0 so source-to-local and source-to-active install syncs rebuild the package first, run install simulation, enforce installer permission coverage, and refuse to copy files before any destructive sync when the package is not install-ready.
 - Atlas Scope Policy v0 so examples, evolution snapshots, embedded generated skills, and validator fixtures remain visible in the full portfolio report without polluting release-actionable gates.
 - Review Annotations v0 for reviewer comments tied to Review Studio gates, source/report paths, and optional line numbers; open blocker annotations now block the Review Studio decision.
-- Review Studio now avoids over-claiming release readiness when blind A/B adjudication, context-budget pressure, waiver handling, and world-class evidence are still pending: the root Meta Skill is in `review` with score `89`, no blockers, four warnings, and explicit actions for Output Lab reviewer adjudication, context-budget follow-up, waiver handling, and world-class evidence completion.
+- Review Studio now avoids over-claiming release readiness when blind A/B adjudication, waiver handling, and world-class evidence are still pending: the root Meta Skill is in `review` with score `91`, no blockers, three warnings, and explicit actions for Output Lab reviewer adjudication, waiver handling, and world-class evidence completion.
 - Review Studio Output Lab actions now link directly to `reports/output_review_decisions.json`, so pending blind A/B reviewer decisions have a concrete template instead of only a general adjudication warning.
 - Output Review Adjudication now preserves blind-review integrity by hiding expected winners for pending or invalid reviewer decisions; answer keys are revealed only after a valid A/B decision exists for that case.
 - Provider Output Eval Runner v0 so `python3 scripts/yao.py output-exec --provider-runner openai` can collect real provider-backed model evidence through a reviewed OpenAI Responses API compatible runner instead of ad hoc shell glue.
@@ -91,7 +91,7 @@ Next move: run provider-backed holdout cases with real credentials and one real
 
 ### 3. Review Studio is unified and now has reviewer actions plus annotations
 
-The Review Studio page aggregates intent, trigger, output, context, runtime, trust, permission approvals, runtime permission probes, atlas, operations-loop, reviewer waiver, reviewer annotations, registry, world-class evidence, and release gates. It exposes current warnings directly and emits `review_actions` for each non-pass gate with a source-fix location, structured `source_refs`, reason, evidence path, and verification command. It now also loads `reports/review_annotations.json`, renders reviewer comments tied to gates and source/report paths, and blocks the page decision when any open blocker annotation exists. The current root report is intentionally not a clean pass: decision `review`, score `89`, `16` gates, `0` blockers, `4` warnings, `4` actions, and `0` open annotation blockers. The warnings are useful evidence: the automated Output Eval Lab is strong, but the blind A/B adjudication still has `5` pending reviewer decisions, the context budget is close to its limit, and the world-class ledger still has `4` pending external or human evidence entries, so the package should not claim fully reviewed or world-class status.
+The Review Studio page aggregates intent, trigger, output, context, runtime, trust, permission approvals, runtime permission probes, atlas, operations-loop, reviewer waiver, reviewer annotations, registry, world-class evidence, and release gates. It exposes current warnings directly and emits `review_actions` for each non-pass gate with a source-fix location, structured `source_refs`, reason, evidence path, and verification command. It now also loads `reports/review_annotations.json`, renders reviewer comments tied to gates and source/report paths, and blocks the page decision when any open blocker annotation exists. The current root report is intentionally not a clean pass: decision `review`, score `91`, `16` gates, `0` blockers, `3` warnings, `3` actions, and `0` open annotation blockers. The warnings are useful evidence: the automated Output Eval Lab is strong, but the blind A/B adjudication still has `5` pending reviewer decisions and the world-class ledger still has `4` pending external or human evidence entries, so the package should not claim fully reviewed or world-class status.
 
 Next move: add richer source-line anchors inside generated reports, record real reviewer annotations during the next human review pass, and close the provider, native-permission, real-client telemetry, and human review evidence entries in the world-class ledger.
 
@@ -126,7 +126,7 @@ Next move: add real client or installer permission enforcement integration.
 | Adoption Drift | `1` metadata-only review event, `0` adoption samples, adoption `0`, risk band `low`; optional `yao.py` CLI capture, external client `telemetry-emit`, `5` `telemetry-hooks` recipes, Browser/Chrome native messaging host, and validated external JSONL import are available but off by default for reproducible release evidence; raw `reports/telemetry_events.jsonl` is gitignored and blocked from zip packages |
 | Review Waivers | ledger generated; current release has `1` warning gate that still needs reviewer decision or a time-bounded waiver; blockers remain non-waivable in v0, and world-class evidence gaps must be closed with accepted evidence rather than waived as completed |
 | Review Annotations | ledger generated; current release has `0` reviewer annotations and `0` open annotation blockers |
-| Review Studio | decision `review`, world-class score `89`, `16` gates, `0` blockers, `4` warnings, `4` review actions, `0` open annotation blockers |
+| Review Studio | decision `review`, world-class score `91`, `16` gates, `0` blockers, `3` warnings, `3` review actions, `0` open annotation blockers |
 | Skill OS 2.0 Audit | `15` audited areas, with local foundation evidence separated from human-required and external-required gaps |
 | World-Class Evidence Plan | `4` remaining evidence tasks: `1` human-required and `3` external-required, all with runbooks, success checks, and privacy contracts |
 | World-Class Evidence Ledger | `4` evidence entries remain pending: `1` human-required and `3` external-required; anti-overclaim guards block planned work, metadata fallbacks, pending review, and local command runners from counting as completion |
diff --git a/reports/skill-overview.json b/reports/skill-overview.json
index d66178e6..93e32078 100644
--- a/reports/skill-overview.json
+++ b/reports/skill-overview.json
@@ -1021,9 +1021,9 @@
       "public_claim_ready": false,
       "public_claim_blocker_count": 5,
       "working_tree_dirty": true,
-      "changed_file_count": 36
+      "changed_file_count": 9
     },
-    "commit": "391821240ba14f088d0fbe12dd5f1fdca7917a39",
+    "commit": "a7960d4cfd2d9a17cbaebcee7b823f0db13df0cf",
     "missing_artifacts": [],
     "limitations": [
       "The git commit and dirty flag are generation-time context; the evidence bundle hash is the durable artifact anchor inside a committed report.",