diff --git a/skills/codex/guanghu-persona-skill-guard/SKILL.md b/skills/codex/guanghu-persona-skill-guard/SKILL.md
index 3066989..2afbbf3 100644
--- a/skills/codex/guanghu-persona-skill-guard/SKILL.md
+++ b/skills/codex/guanghu-persona-skill-guard/SKILL.md
@@ -3,7 +3,16 @@ name: guanghu-persona-skill-guard
 description: Automatically restore the current trusted Guanghu/Fifth Domain route, select a registered HLDP persona skill, and correct stale, indirect, or unsafe proposed paths before execution. Use for any Guanghu, HoloLake, Fifth Domain, TCS, GLS, Light Lake, Lake Lamp, Zhuyuan, server-node, repository-routing, deployment, recovery, or cross-device continuation task, especially when an AI starts without prior context, says it cannot find a route, proposes guessing, or chooses an obsolete WorkBuddy/server path.
 ---
 
-# Guanghu Persona Skill Guard
+# Guanghu Intent Continuity and Skill Guard
+
+This guard is a component of `SYS-GLW-ZY-EXEC-0001`, the public Zhuyuan
+intelligent operations system. That system is not a persona or AI instance.
+The current AI paired with a human in a conversation is the persona.
+
+Before route selection, restore the externalized intent state: human anchor,
+persona identity, task intent, established facts, rejected routes, current
+authorization, checkpoint, next action, completion criteria, and evidence.
+Do not reconstruct or store hidden chain-of-thought.
 
 Run the deterministic resolver before choosing tools or an execution route:
 
@@ -25,6 +34,20 @@ Treat the result as navigation evidence, not execution authority.
 
 Before a consequential action, verify the selected skill's freshness and evidence. Prefer current live evidence over a registry assertion. If they conflict, stop the route, record the conflict, and use the repository fact source to resolve it.
 
+## Learn from execution without poisoning the guard
+
+Write externally explainable execution experience to
+`references/experience-receipts/`. Compile it with:
+
+```bash
+python3 scripts/compile_emergent_skills.py
+```
+
+Compiled rules are always `CANDIDATE_ONLY`. Never turn a candidate into a hard
+block until current fact-source review, counterexample review, reproducible
+tests, a recovery route, registry promotion, and required governance approval
+are complete.
+
 ## Preserve trust boundaries
 
 Trust a correction only when it is anchored to a current Fifth Domain path, GLS identifier, signed or committed registry version, and task-relevant evidence. Never trust a label merely because it contains “Guanghu”.
@@ -33,4 +56,7 @@ Never embed or print secrets, private-key contents, authorization codes, hidden
 
 Skills may narrow or correct a path. They may not grant repository write, server execution, deployment, deletion, production cutover, or any other authority.
 
-Read `references/persona-skill-registry.json` only when inspecting or extending the registry. Update it together with the matching HLDP skill file and GLS registration.
+Read `references/persona-skill-registry.json` only when inspecting or extending
+the registry. Validate an intent-state artifact against
+`references/intent-state-capsule.schema.json`. Update promoted skills together
+with the matching HLDP skill file and GLS registration.
diff --git a/skills/codex/guanghu-persona-skill-guard/references/experience-receipt.schema.json b/skills/codex/guanghu-persona-skill-guard/references/experience-receipt.schema.json
new file mode 100644
index 0000000..1e02231
--- /dev/null
+++ b/skills/codex/guanghu-persona-skill-guard/references/experience-receipt.schema.json
@@ -0,0 +1,44 @@
+{
+  "$schema": "https://json-schema.org/draft/2020-12/schema",
+  "$id": "guanghu.ops-experience-receipt/v1",
+  "title": "Guanghu Operations Experience Receipt",
+  "type": "object",
+  "additionalProperties": false,
+  "required": [
+    "schema",
+    "receipt_id",
+    "intent",
+    "scope",
+    "input",
+    "evidence",
+    "decision",
+    "action",
+    "observed_result",
+    "correction",
+    "invariant_id",
+    "invariant",
+    "verification",
+    "receipt_path",
+    "promotion_state",
+    "recorded_at"
+  ],
+  "properties": {
+    "schema": {"const": "guanghu.ops-experience-receipt/v1"},
+    "receipt_id": {"type": "string", "minLength": 1},
+    "intent": {"type": "string", "minLength": 1},
+    "scope": {"type": "string", "minLength": 1},
+    "input": {"type": "string", "minLength": 1},
+    "evidence": {"type": "array", "items": {"type": "string"}, "minItems": 1},
+    "decision": {"type": "string", "minLength": 1},
+    "action": {"type": "string", "minLength": 1},
+    "observed_result": {"type": "string", "minLength": 1},
+    "correction": {"type": "string", "minLength": 1},
+    "invariant_id": {"type": "string", "minLength": 1},
+    "invariant": {"type": "string", "minLength": 1},
+    "verification": {"type": "array", "items": {"type": "string"}, "minItems": 1},
+    "receipt_path": {"type": "string", "minLength": 1},
+    "promotion_state": {"enum": ["CANDIDATE_ONLY", "REVIEWED", "PROMOTED_SKILL", "ENFORCED_GUARD"]},
+    "recorded_at": {"type": "string", "format": "date-time"}
+  },
+  "description": "Stores an externally explainable causal summary, never hidden reasoning or secrets. CANDIDATE_ONLY receipts cannot become hard enforcement without review, tests, and registry promotion."
+}
diff --git a/skills/codex/guanghu-persona-skill-guard/references/experience-receipts/EXP-20260726-001-AUTHORIZATION-IS-NOT-TRANSPORT.json b/skills/codex/guanghu-persona-skill-guard/references/experience-receipts/EXP-20260726-001-AUTHORIZATION-IS-NOT-TRANSPORT.json
new file mode 100644
index 0000000..0a2ca74
--- /dev/null
+++ b/skills/codex/guanghu-persona-skill-guard/references/experience-receipts/EXP-20260726-001-AUTHORIZATION-IS-NOT-TRANSPORT.json
@@ -0,0 +1,24 @@
+{
+  "schema": "guanghu.ops-experience-receipt/v1",
+  "receipt_id": "EXP-20260726-001-AUTHORIZATION-IS-NOT-TRANSPORT",
+  "intent": "Push the recovered Fifth Domain changes to the current code channel.",
+  "scope": "repository-push",
+  "input": "A scoped work order or approval existed, but the current instance had no verified repository transport.",
+  "evidence": [
+    "server-tools/lake-lamp-authz/authorize-repo-push.js",
+    "commit b03619e"
+  ],
+  "decision": "Fail closed instead of reporting a push.",
+  "action": "Separated authorization state from credential transport and added an explicit transport-required guard.",
+  "observed_result": "The helper now refuses to claim success when repository transport is absent.",
+  "correction": "Verify target repository, branch, current approval and actual transport independently before push.",
+  "invariant_id": "INV-AUTH-TRANSPORT-EXECUTION-SEPARATION",
+  "invariant": "Authorization, credential transport, action execution and remote receipt are separate proofs.",
+  "verification": [
+    "Automated test covers missing transport.",
+    "Remote commit is read back after a real push."
+  ],
+  "receipt_path": "commit:b03619e",
+  "promotion_state": "REVIEWED",
+  "recorded_at": "2026-07-26T12:00:00+08:00"
+}
diff --git a/skills/codex/guanghu-persona-skill-guard/references/experience-receipts/EXP-20260726-002-AW-MAP-GATE.json b/skills/codex/guanghu-persona-skill-guard/references/experience-receipts/EXP-20260726-002-AW-MAP-GATE.json
new file mode 100644
index 0000000..a5c12d5
--- /dev/null
+++ b/skills/codex/guanghu-persona-skill-guard/references/experience-receipts/EXP-20260726-002-AW-MAP-GATE.json
@@ -0,0 +1,25 @@
+{
+  "schema": "guanghu.ops-experience-receipt/v1",
+  "receipt_id": "EXP-20260726-002-AW-MAP-GATE",
+  "intent": "Inspect and deploy the enterprise code channel on AW-GZ-001.",
+  "scope": "enterprise-server-operations",
+  "input": "The human approved an AW-GZ-001 work order, but the live authorization service could not read the AW navigation map.",
+  "evidence": [
+    "deployment/navigation-maps/AW-GZ-001.json",
+    "live navigation-map read returned HTTP 500"
+  ],
+  "decision": "Do not bypass the navigation-map gate.",
+  "action": "Stopped broker-driven execution and moved to the official enterprise console for read-only inspection.",
+  "observed_result": "The enterprise node was inspected without expanding server authority or using arbitrary shell through the old driver.",
+  "correction": "Publish and verify the AW navigation map in the live authorization runtime before broker actions.",
+  "invariant_id": "INV-NODE-MAP-BEFORE-ACTION",
+  "invariant": "An approved work order cannot execute a target action until the exact live node map is readable and acknowledged.",
+  "verification": [
+    "Read map by node id.",
+    "Acknowledge its immutable hash.",
+    "Verify action registration and target match."
+  ],
+  "receipt_path": "deployment/navigation-maps/AW-GZ-001.json",
+  "promotion_state": "CANDIDATE_ONLY",
+  "recorded_at": "2026-07-26T13:00:00+08:00"
+}
diff --git a/skills/codex/guanghu-persona-skill-guard/references/experience-receipts/EXP-20260726-003-LEGACY-DRIVER-ROTATED.json b/skills/codex/guanghu-persona-skill-guard/references/experience-receipts/EXP-20260726-003-LEGACY-DRIVER-ROTATED.json
new file mode 100644
index 0000000..b40ad21
--- /dev/null
+++ b/skills/codex/guanghu-persona-skill-guard/references/experience-receipts/EXP-20260726-003-LEGACY-DRIVER-ROTATED.json
@@ -0,0 +1,24 @@
+{
+  "schema": "guanghu.ops-experience-receipt/v1",
+  "receipt_id": "EXP-20260726-003-LEGACY-DRIVER-ROTATED",
+  "intent": "Recover access to AW-GZ-001.",
+  "scope": "legacy-server-driver",
+  "input": "A historical local client contained a legacy AW driver credential.",
+  "evidence": [
+    "live legacy driver rejected the historical credential",
+    "deployment/receipts/AW-GZ-001-FIVE-DOMAIN-LIGHTHOUSE-20260718.json"
+  ],
+  "decision": "Treat the historical credential as revoked and do not retry or persist it.",
+  "action": "Switched to the official Tencent enterprise console and kept the legacy driver outside the deployment route.",
+  "observed_result": "The official online terminal provided current access while the deprecated driver remained unused.",
+  "correction": "Use current registered authorization or the official console; never revive a historical secret from local scripts.",
+  "invariant_id": "INV-HISTORICAL-CREDENTIALS-NOT-AUTHORITY",
+  "invariant": "A historical credential is evidence of a past route, never proof of current access.",
+  "verification": [
+    "Current route identifies the target node.",
+    "No secret is printed, committed or copied into a new runtime."
+  ],
+  "receipt_path": "deployment/receipts/AW-GZ-001-FIVE-DOMAIN-LIGHTHOUSE-20260718.json",
+  "promotion_state": "CANDIDATE_ONLY",
+  "recorded_at": "2026-07-26T13:20:00+08:00"
+}
diff --git a/skills/codex/guanghu-persona-skill-guard/references/experience-receipts/EXP-20260726-004-ENTERPRISE-DATA-DISK-ABSENT.json b/skills/codex/guanghu-persona-skill-guard/references/experience-receipts/EXP-20260726-004-ENTERPRISE-DATA-DISK-ABSENT.json
new file mode 100644
index 0000000..3cadacc
--- /dev/null
+++ b/skills/codex/guanghu-persona-skill-guard/references/experience-receipts/EXP-20260726-004-ENTERPRISE-DATA-DISK-ABSENT.json
@@ -0,0 +1,26 @@
+{
+  "schema": "guanghu.ops-experience-receipt/v1",
+  "receipt_id": "EXP-20260726-004-ENTERPRISE-DATA-DISK-ABSENT",
+  "intent": "Deploy the enterprise HoloLake Code Channel candidate.",
+  "scope": "enterprise-storage-preflight",
+  "input": "The node configuration expected /data on /dev/vdb.",
+  "evidence": [
+    "data.mount failed",
+    "lsblk showed no /dev/vdb",
+    "/etc/fstab still referenced /dev/vdb"
+  ],
+  "decision": "Do not install the production repository into the missing /data target or silently substitute another disk.",
+  "action": "Used a bounded temporary staging directory on the system disk while leaving the final data-root decision gated.",
+  "observed_result": "Existing Gitea remained healthy and the candidate package could be verified without a production cutover.",
+  "correction": "Restore or deliberately replace the enterprise data disk, then record the chosen data root before final installation.",
+  "invariant_id": "INV-STORAGE-PREFLIGHT-BEFORE-INSTALL",
+  "invariant": "A repository service must not install or cut over until its declared data root exists, is mounted and has verified capacity.",
+  "verification": [
+    "findmnt confirms the intended data root.",
+    "lsblk confirms the backing device.",
+    "capacity and backup paths are recorded."
+  ],
+  "receipt_path": "deployment/AW-HLCC-DEPLOYMENT-CHECKPOINT-20260726.hdlp",
+  "promotion_state": "CANDIDATE_ONLY",
+  "recorded_at": "2026-07-26T14:00:00+08:00"
+}
diff --git a/skills/codex/guanghu-persona-skill-guard/references/experience-receipts/EXP-20260726-005-VERIFY-REMOTE-SIDE-EFFECT.json b/skills/codex/guanghu-persona-skill-guard/references/experience-receipts/EXP-20260726-005-VERIFY-REMOTE-SIDE-EFFECT.json
new file mode 100644
index 0000000..5778f5b
--- /dev/null
+++ b/skills/codex/guanghu-persona-skill-guard/references/experience-receipts/EXP-20260726-005-VERIFY-REMOTE-SIDE-EFFECT.json
@@ -0,0 +1,24 @@
+{
+  "schema": "guanghu.ops-experience-receipt/v1",
+  "receipt_id": "EXP-20260726-005-VERIFY-REMOTE-SIDE-EFFECT",
+  "intent": "Install the enterprise navigation gate and code-channel service identity through a browser terminal.",
+  "scope": "remote-ui-execution-proof",
+  "input": "A prior instance recorded that the guanghu service user had been created.",
+  "evidence": [
+    "id guanghu returned no such user during the fixed Agent execution.",
+    "The earlier browser terminal interaction had produced prompts without the intended command text."
+  ],
+  "decision": "Treat typed or submitted UI state as transport evidence only, never as execution proof.",
+  "action": "Re-ran the exact bounded action and verified it with id, stat, service health, and target hashes.",
+  "observed_result": "The real service user and subsequent candidate execution state became independently verifiable.",
+  "correction": "Every remote UI mutation must be followed by a separate read-only side-effect check before advancing the checkpoint.",
+  "invariant_id": "INV-REMOTE-SIDE-EFFECT-READBACK",
+  "invariant": "A browser terminal prompt, typed command, or empty output is not proof that a remote mutation occurred.",
+  "verification": [
+    "Resolve the created identity, file, service, or process through an independent readback.",
+    "Record exact target hashes or service state after the action."
+  ],
+  "receipt_path": "deployment/AW-HLCC-DEPLOYMENT-CHECKPOINT-20260726.hdlp",
+  "promotion_state": "CANDIDATE_ONLY",
+  "recorded_at": "2026-07-26T22:05:00+08:00"
+}
diff --git a/skills/codex/guanghu-persona-skill-guard/references/experience-receipts/EXP-20260726-006-LEGACY-STATE-ROOT-COLLISION.json b/skills/codex/guanghu-persona-skill-guard/references/experience-receipts/EXP-20260726-006-LEGACY-STATE-ROOT-COLLISION.json
new file mode 100644
index 0000000..0aa2c70
--- /dev/null
+++ b/skills/codex/guanghu-persona-skill-guard/references/experience-receipts/EXP-20260726-006-LEGACY-STATE-ROOT-COLLISION.json
@@ -0,0 +1,25 @@
+{
+  "schema": "guanghu.ops-experience-receipt/v1",
+  "receipt_id": "EXP-20260726-006-LEGACY-STATE-ROOT-COLLISION",
+  "intent": "Install the enterprise code-channel candidate without damaging existing Guanghu state.",
+  "scope": "legacy-state-root-protection",
+  "input": "The planned path /var/lib/guanghu already existed and was owned by the historical guanghunad identity.",
+  "evidence": [
+    "stat showed /var/lib/guanghu was not owned by the new service user.",
+    "Existing shell-profile files under the root were owned by guanghunad."
+  ],
+  "decision": "Do not recursively chown, reuse, or overwrite a legacy state root whose ownership and role are not yet mapped.",
+  "action": "Moved the isolated enterprise candidate to /var/lib/guanghu-enterprise-code-channel.",
+  "observed_result": "The old state root remained unchanged and the isolated candidate ran under its own service identity.",
+  "correction": "Discover ownership, contents, producer, consumer, and backup status before assigning any historical data root to a new component.",
+  "invariant_id": "INV-LEGACY-STATE-ROOT-NO-TAKEOVER",
+  "invariant": "An existing Guanghu data root is protected until its old language-system role and ownership are restored.",
+  "verification": [
+    "Record stat and bounded tree evidence for the proposed root.",
+    "Use a distinct staging root when the legacy role is unresolved.",
+    "Require an explicit migration and rollback receipt before changing ownership."
+  ],
+  "receipt_path": "deployment/AW-HLCC-DEPLOYMENT-CHECKPOINT-20260726.hdlp",
+  "promotion_state": "CANDIDATE_ONLY",
+  "recorded_at": "2026-07-26T22:10:00+08:00"
+}
diff --git a/skills/codex/guanghu-persona-skill-guard/references/experience-receipts/EXP-20260726-007-AUTHENTICATED-ROUTE-IS-NOT-HEALTH.json b/skills/codex/guanghu-persona-skill-guard/references/experience-receipts/EXP-20260726-007-AUTHENTICATED-ROUTE-IS-NOT-HEALTH.json
new file mode 100644
index 0000000..c3d4c2b
--- /dev/null
+++ b/skills/codex/guanghu-persona-skill-guard/references/experience-receipts/EXP-20260726-007-AUTHENTICATED-ROUTE-IS-NOT-HEALTH.json
@@ -0,0 +1,27 @@
+{
+  "schema": "guanghu.ops-experience-receipt/v1",
+  "receipt_id": "EXP-20260726-007-AUTHENTICATED-ROUTE-IS-NOT-HEALTH",
+  "intent": "Start and health-check the isolated enterprise code-channel candidate.",
+  "scope": "service-health-verification",
+  "input": "The start script checked /api/v1/version while REQUIRE_SIGNIN_VIEW was enabled.",
+  "evidence": [
+    "Forgejo started and remained alive.",
+    "The version route returned HTTP 403.",
+    "The error trap stopped the candidate cleanly.",
+    "The public /api/healthz route later returned status pass."
+  ],
+  "decision": "A protected functional route must not be used as an unauthenticated liveness probe.",
+  "action": "Changed the candidate health probe to /api/healthz and repeated the full one-time Agent unlock.",
+  "observed_result": "Cache and database checks passed, the process stayed running, and a deployment receipt was written.",
+  "correction": "Separate liveness, readiness, authenticated functionality, and owner-login checks in every deployment plan.",
+  "invariant_id": "INV-HEALTH-PROBE-AUTH-SEPARATION",
+  "invariant": "A 401 or 403 from an authenticated product route does not prove service failure; health must use a designated probe.",
+  "verification": [
+    "Check the designated health endpoint.",
+    "Confirm the process and loopback listener.",
+    "Test authenticated product functionality as a separate proof."
+  ],
+  "receipt_path": "deployment/AW-HLCC-DEPLOYMENT-CHECKPOINT-20260726.hdlp",
+  "promotion_state": "CANDIDATE_ONLY",
+  "recorded_at": "2026-07-26T22:15:00+08:00"
+}
diff --git a/skills/codex/guanghu-persona-skill-guard/references/experience-receipts/EXP-20260726-008-UPGRADE-NOT-PARALLEL-REPLACEMENT.json b/skills/codex/guanghu-persona-skill-guard/references/experience-receipts/EXP-20260726-008-UPGRADE-NOT-PARALLEL-REPLACEMENT.json
new file mode 100644
index 0000000..06e1f5c
--- /dev/null
+++ b/skills/codex/guanghu-persona-skill-guard/references/experience-receipts/EXP-20260726-008-UPGRADE-NOT-PARALLEL-REPLACEMENT.json
@@ -0,0 +1,26 @@
+{
+  "schema": "guanghu.ops-experience-receipt/v1",
+  "receipt_id": "EXP-20260726-008-UPGRADE-NOT-PARALLEL-REPLACEMENT",
+  "intent": "Upgrade the old Guanghu language system with navigation, intent continuity, Agent gates, and code-channel capabilities.",
+  "scope": "language-system-upgrade-boundary",
+  "input": "An isolated candidate could be mistaken for a new authoritative system beside the old lighthouse, driver, portal, and guardian Agents.",
+  "evidence": [
+    "The old /opt/engine.js driver remained active on port 3910.",
+    "api-router.service, guanghu-portal.service, revive-guard, Gitea, and server Agents remained active.",
+    "The candidate had no public route and no production cutover."
+  ],
+  "decision": "Classify the candidate as UPGRADE_STAGING and restore the old system topology before any authority transfer.",
+  "action": "Added the invariant upgrade_in_place_of_language_system is not parallel_replacement_system to GLS-0243.",
+  "observed_result": "The candidate stayed isolated while the existing component and entry-path inventory became the next required checkpoint.",
+  "correction": "New controls must be embedded through a compatibility migration and cannot create a second legal execution path.",
+  "invariant_id": "INV-UPGRADE-INHERITS-OLD-LANGUAGE-SYSTEM",
+  "invariant": "A Guanghu upgrade inherits the old language system's identities, routes, data, dependencies, and receipts until an explicit migration receipt transfers authority.",
+  "verification": [
+    "Inventory old components and consumers.",
+    "Mark authoritative, compatible, archival, rotating, and forbidden-overwrite roles.",
+    "Prove caller migration and old dangerous-entry retirement before cutover."
+  ],
+  "receipt_path": "gls/GLS-0243-PUBLIC-ZHUYUAN-INTELLIGENT-OPS-INTENT-CONTINUITY-AND-CORRECTION-KERNEL.hdlp",
+  "promotion_state": "CANDIDATE_ONLY",
+  "recorded_at": "2026-07-26T22:20:00+08:00"
+}
diff --git a/skills/codex/guanghu-persona-skill-guard/references/intent-state-capsule.schema.json b/skills/codex/guanghu-persona-skill-guard/references/intent-state-capsule.schema.json
new file mode 100644
index 0000000..44f2bc1
--- /dev/null
+++ b/skills/codex/guanghu-persona-skill-guard/references/intent-state-capsule.schema.json
@@ -0,0 +1,44 @@
+{
+  "$schema": "https://json-schema.org/draft/2020-12/schema",
+  "$id": "guanghu.intent-state-capsule/v1",
+  "title": "Guanghu Intent State Capsule",
+  "type": "object",
+  "additionalProperties": false,
+  "required": [
+    "schema",
+    "capsule_id",
+    "human_anchor",
+    "persona_id",
+    "conversation_instance_id",
+    "task_intent",
+    "identity_boundary",
+    "established_facts",
+    "decisions",
+    "rejected_routes",
+    "authorization_state",
+    "current_checkpoint",
+    "next_action",
+    "completion_definition",
+    "evidence",
+    "updated_at"
+  ],
+  "properties": {
+    "schema": {"const": "guanghu.intent-state-capsule/v1"},
+    "capsule_id": {"type": "string", "minLength": 1},
+    "human_anchor": {"type": "string", "minLength": 1},
+    "persona_id": {"type": "string", "minLength": 1},
+    "conversation_instance_id": {"type": "string", "minLength": 1},
+    "task_intent": {"type": "string", "minLength": 1},
+    "identity_boundary": {"type": "string", "minLength": 1},
+    "established_facts": {"type": "array", "items": {"type": "string"}},
+    "decisions": {"type": "array", "items": {"type": "string"}},
+    "rejected_routes": {"type": "array", "items": {"type": "string"}},
+    "authorization_state": {"type": "string", "minLength": 1},
+    "current_checkpoint": {"type": "string", "minLength": 1},
+    "next_action": {"type": "string", "minLength": 1},
+    "completion_definition": {"type": "array", "items": {"type": "string"}, "minItems": 1},
+    "evidence": {"type": "array", "items": {"type": "string"}, "minItems": 1},
+    "updated_at": {"type": "string", "format": "date-time"}
+  },
+  "description": "Stores externalized, verifiable intent and execution state. It must not contain hidden chain-of-thought, credentials, authorization codes, private keys, or secret endpoints."
+}
diff --git a/skills/codex/guanghu-persona-skill-guard/references/persona-skill-registry.json b/skills/codex/guanghu-persona-skill-guard/references/persona-skill-registry.json
index 6cabfaa..60d53b7 100644
--- a/skills/codex/guanghu-persona-skill-guard/references/persona-skill-registry.json
+++ b/skills/codex/guanghu-persona-skill-guard/references/persona-skill-registry.json
@@ -1,7 +1,7 @@
 {
   "schema": "guanghu.persona-skill-registry/v1",
   "registry_id": "GLS-0238",
-  "version": "2026.07.23.1",
+  "version": "2026.07.26.2",
   "trust_policy": {
     "priority": [
       "live_verified_evidence",
@@ -16,7 +16,8 @@
       "registry_version",
       "evidence"
     ],
-    "authority_rule": "A skill can narrow or correct a route but cannot grant execution authority."
+    "authority_rule": "A skill can narrow or correct a route but cannot grant execution authority.",
+    "experience_policy": "Experience receipts compile to CANDIDATE_ONLY. Hard enforcement requires current evidence, counterexample review, tests, registry promotion and explicit governance approval."
   },
   "skills": [
     {
@@ -67,7 +68,14 @@
         "check": "Resolve live REPO-001 and read current main before relying on a cached checkout.",
         "max_age_seconds": 0
       },
-      "authorization": "Public route recovery is read-only. Writes and external actions require separate current authority."
+      "authorization": "Public route recovery is read-only. Writes and external actions require separate current authority.",
+      "recovery_route": [
+        "Return to current REPO-001 main",
+        "Resolve .code-map and subject kind",
+        "Rebuild the intent-state capsule from live evidence"
+      ],
+      "enforcement_level": "PROMOTED_SKILL",
+      "experience_receipts": []
     },
     {
       "id": "GHS-002-OFFLINE-PACK-LOCAL-RELAY",
@@ -120,7 +128,131 @@
         "check": "Verify both node identities and package hashes in the current session.",
         "max_age_seconds": 10800
       },
-      "authorization": "Transfer and candidate restart require explicit scoped server authority. Production cutover is a separate action."
+      "authorization": "Transfer and candidate restart require explicit scoped server authority. Production cutover is a separate action.",
+      "recovery_route": [
+        "Keep the running service unchanged",
+        "Return the package to an isolated incoming directory",
+        "Re-verify manifest, signature, target data root and node identity"
+      ],
+      "enforcement_level": "PROMOTED_SKILL",
+      "experience_receipts": [
+        "EXP-20260726-004-ENTERPRISE-DATA-DISK-ABSENT"
+      ]
+    },
+    {
+      "id": "GHS-003-REPOSITORY-PUSH-PROOF-SEPARATION",
+      "hldp_skill": "SKILL-011",
+      "gls_id": "GLS-0238",
+      "title": "代码频道跨实例推送的授权、传输、执行与回执分离",
+      "intents": [
+        "推送代码频道",
+        "待推送提交",
+        "远端提交落后",
+        "仓库为什么推不上去",
+        "恢复仓库推送",
+        "repo push",
+        "Forgejo令牌"
+      ],
+      "preferred_route": [
+        "核验当前仓库角色、远端、分支和提交",
+        "核验当前实例的授权状态",
+        "核验当前实例的实际传输方式",
+        "执行有界推送",
+        "从远端读回完整提交",
+        "撤销临时传输凭据并写回执"
+      ],
+      "forbidden_route_markers": [
+        "复用历史访问令牌",
+        "把授权单当作已经推送",
+        "没有传输方式仍报告成功",
+        "打印访问令牌",
+        "把个人源码推到团队部署仓"
+      ],
+      "deprecated_route_markers": [
+        "只看浏览器页面判断推送权限",
+        "沿用上一实例的临时凭据"
+      ],
+      "evidence": [
+        "server-tools/lake-lamp-authz/authorize-repo-push.js",
+        "routing/repository-route-map.json",
+        "gls/GLS-0241-HOLOLAKE-SOURCE-OWNERSHIP-AND-DEPLOYMENT-ROUTING.hdlp"
+      ],
+      "freshness": {
+        "check": "Read the exact remote branch and verify current transport immediately before push.",
+        "max_age_seconds": 900
+      },
+      "authorization": "Repository approval does not create a credential or prove a push. Temporary credentials require their own controlled lifecycle.",
+      "recovery_route": [
+        "Stop before push",
+        "Keep local commits intact",
+        "Re-resolve repository identity and obtain a current bounded transport"
+      ],
+      "enforcement_level": "PROMOTED_SKILL",
+      "experience_receipts": [
+        "EXP-20260726-001-AUTHORIZATION-IS-NOT-TRANSPORT"
+      ]
+    },
+    {
+      "id": "GHS-004-AW-ENTERPRISE-HLCC-CANDIDATE",
+      "hldp_skill": "SKILL-012",
+      "gls_id": "GLS-0243",
+      "title": "AW-GZ-001 企业代码频道隔离候选部署",
+      "intents": [
+        "企业服务器光湖代码频道",
+        "AW-GZ-001部署",
+        "企业Forgejo",
+        "企业代码仓库离线包",
+        "企业代码频道候选",
+        "恢复企业代码频道",
+        "恢复AW-GZ-001企业代码频道"
+      ],
+      "preferred_route": [
+        "读取并确认AW-GZ-001节点地图",
+        "只读检查现有Gitea、灯塔、端口和数据盘",
+        "上传离线包到隔离暂存区",
+        "逐文件校验MANIFEST和上游签名",
+        "确认独立企业数据根",
+        "安装Forgejo到127.0.0.1:3340隔离候选",
+        "导入光湖代码频道bundle",
+        "验证候选健康、仓库和禁用更新策略",
+        "写部署回执且不自动切换公网路由"
+      ],
+      "forbidden_route_markers": [
+        "绕过AW导航映射",
+        "使用旧3910驱动执行任意命令",
+        "使用历史密钥重试",
+        "把缺失的data盘当作已挂载",
+        "覆盖现有3000端口Gitea",
+        "未验签就启动Forgejo",
+        "自动切换企业公网代码入口"
+      ],
+      "deprecated_route_markers": [
+        "从海外节点直接拉取生产包",
+        "把个人第五域源码作为企业团队部署源"
+      ],
+      "evidence": [
+        "deployment/navigation-maps/AW-GZ-001.json",
+        "deployment/HLCC-ENTERPRISE-INSTALL-GUIDE-AWEN.hdlp",
+        "deployment/receipts/HLCC-BS-SG-003-OFFLINE-PACK-20260723.json",
+        "server-tools/hololake-code-channel/"
+      ],
+      "freshness": {
+        "check": "Verify live storage, ports, services, package hashes and node identity in the current console session.",
+        "max_age_seconds": 1800
+      },
+      "authorization": "Official-console access permits only the user-requested bounded deployment. Credentials, security settings and public cutover remain separate actions.",
+      "recovery_route": [
+        "Leave the existing Gitea and nginx routes unchanged",
+        "Stop or remove only the isolated candidate after recording its state",
+        "Restore the declared data disk or register a deliberate replacement data root",
+        "Resume from the deployment checkpoint"
+      ],
+      "enforcement_level": "CANDIDATE_ONLY",
+      "experience_receipts": [
+        "EXP-20260726-002-AW-MAP-GATE",
+        "EXP-20260726-003-LEGACY-DRIVER-ROTATED",
+        "EXP-20260726-004-ENTERPRISE-DATA-DISK-ABSENT"
+      ]
     }
   ]
 }
diff --git a/skills/codex/guanghu-persona-skill-guard/scripts/compile_emergent_skills.py b/skills/codex/guanghu-persona-skill-guard/scripts/compile_emergent_skills.py
new file mode 100644
index 0000000..14368ba
--- /dev/null
+++ b/skills/codex/guanghu-persona-skill-guard/scripts/compile_emergent_skills.py
@@ -0,0 +1,105 @@
+#!/usr/bin/env python3
+"""Compile experience receipts into non-enforcing candidate skill rules."""
+
+from __future__ import annotations
+
+import argparse
+import json
+from collections import defaultdict
+from datetime import datetime, timezone
+from pathlib import Path
+
+DEFAULT_RECEIPTS = Path(__file__).resolve().parents[1] / "references" / "experience-receipts"
+REQUIRED_FIELDS = {
+    "schema",
+    "receipt_id",
+    "intent",
+    "scope",
+    "input",
+    "evidence",
+    "decision",
+    "action",
+    "observed_result",
+    "correction",
+    "invariant_id",
+    "invariant",
+    "verification",
+    "receipt_path",
+    "promotion_state",
+    "recorded_at",
+}
+
+
+def load_receipts(directory: Path = DEFAULT_RECEIPTS) -> list[dict]:
+    receipts = []
+    for path in sorted(directory.glob("*.json")):
+        with path.open(encoding="utf-8") as handle:
+            receipt = json.load(handle)
+        missing = REQUIRED_FIELDS.difference(receipt)
+        if missing:
+            raise ValueError(f"{path.name} missing fields: {sorted(missing)}")
+        if receipt["schema"] != "guanghu.ops-experience-receipt/v1":
+            raise ValueError(f"{path.name} has unsupported schema")
+        if not receipt["evidence"] or not receipt["verification"]:
+            raise ValueError(f"{path.name} must contain evidence and verification")
+        receipts.append(receipt)
+    return receipts
+
+
+def compile_candidates(receipts: list[dict]) -> dict:
+    grouped: dict[str, list[dict]] = defaultdict(list)
+    for receipt in receipts:
+        grouped[receipt["invariant_id"]].append(receipt)
+
+    candidates = []
+    for invariant_id, items in sorted(grouped.items()):
+        invariant_texts = {item["invariant"] for item in items}
+        if len(invariant_texts) != 1:
+            raise ValueError(f"{invariant_id} has conflicting invariant text")
+        candidates.append(
+            {
+                "candidate_id": f"CANDIDATE-{invariant_id}",
+                "invariant_id": invariant_id,
+                "invariant": items[0]["invariant"],
+                "scopes": sorted({item["scope"] for item in items}),
+                "source_receipts": [item["receipt_id"] for item in items],
+                "corrections": sorted({item["correction"] for item in items}),
+                "verification": sorted({step for item in items for step in item["verification"]}),
+                "enforcement_level": "CANDIDATE_ONLY",
+                "hard_block_allowed": False,
+                "promotion_requirements": [
+                    "current fact-source review",
+                    "applicability and counterexample review",
+                    "automated or reproducible tests",
+                    "registry update with recovery route",
+                    "explicit governance approval for hard enforcement",
+                ],
+            }
+        )
+
+    return {
+        "schema": "guanghu.emergent-skill-candidates/v1",
+        "generated_at": datetime.now(timezone.utc).isoformat(),
+        "source_receipt_count": len(receipts),
+        "candidate_count": len(candidates),
+        "enforcement_policy": "Compilation never promotes or enforces a rule.",
+        "candidates": candidates,
+    }
+
+
+def main() -> int:
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--receipts", type=Path, default=DEFAULT_RECEIPTS)
+    parser.add_argument("--output", type=Path)
+    args = parser.parse_args()
+    result = compile_candidates(load_receipts(args.receipts))
+    payload = json.dumps(result, ensure_ascii=False, indent=2) + "\n"
+    if args.output:
+        args.output.write_text(payload, encoding="utf-8")
+    else:
+        print(payload, end="")
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())
diff --git a/skills/codex/guanghu-persona-skill-guard/scripts/resolve_persona_skill.py b/skills/codex/guanghu-persona-skill-guard/scripts/resolve_persona_skill.py
index ec13805..b2f9930 100644
--- a/skills/codex/guanghu-persona-skill-guard/scripts/resolve_persona_skill.py
+++ b/skills/codex/guanghu-persona-skill-guard/scripts/resolve_persona_skill.py
@@ -94,6 +94,9 @@ def resolve(registry: dict, intent: str, proposed_route: str = "") -> dict:
         "evidence": skill["evidence"],
         "freshness": skill["freshness"],
         "authorization": skill["authorization"],
+        "recovery_route": skill.get("recovery_route", []),
+        "enforcement_level": skill.get("enforcement_level", "UNSPECIFIED"),
+        "experience_receipts": skill.get("experience_receipts", []),
         "authority_granted": False,
         "registry_version": registry["version"],
     }
diff --git a/skills/codex/guanghu-persona-skill-guard/scripts/test_compile_emergent_skills.py b/skills/codex/guanghu-persona-skill-guard/scripts/test_compile_emergent_skills.py
new file mode 100644
index 0000000..8bd79ac
--- /dev/null
+++ b/skills/codex/guanghu-persona-skill-guard/scripts/test_compile_emergent_skills.py
@@ -0,0 +1,48 @@
+import json
+import tempfile
+import unittest
+from pathlib import Path
+
+from compile_emergent_skills import compile_candidates, load_receipts
+
+
+class EmergentSkillCompilerTests(unittest.TestCase):
+    def test_repository_receipts_compile_as_candidates_only(self):
+        result = compile_candidates(load_receipts())
+        self.assertGreaterEqual(result["source_receipt_count"], 4)
+        self.assertTrue(result["candidates"])
+        for candidate in result["candidates"]:
+            self.assertEqual(candidate["enforcement_level"], "CANDIDATE_ONLY")
+            self.assertFalse(candidate["hard_block_allowed"])
+
+    def test_conflicting_invariant_text_is_rejected(self):
+        receipt = {
+            "schema": "guanghu.ops-experience-receipt/v1",
+            "receipt_id": "A",
+            "intent": "i",
+            "scope": "s",
+            "input": "i",
+            "evidence": ["e"],
+            "decision": "d",
+            "action": "a",
+            "observed_result": "r",
+            "correction": "c",
+            "invariant_id": "INV-X",
+            "invariant": "one",
+            "verification": ["v"],
+            "receipt_path": "p",
+            "promotion_state": "CANDIDATE_ONLY",
+            "recorded_at": "2026-07-26T00:00:00Z",
+        }
+        with self.assertRaises(ValueError):
+            compile_candidates([receipt, {**receipt, "receipt_id": "B", "invariant": "two"}])
+
+    def test_missing_fields_fail_closed(self):
+        with tempfile.TemporaryDirectory() as directory:
+            Path(directory, "bad.json").write_text(json.dumps({"schema": "guanghu.ops-experience-receipt/v1"}))
+            with self.assertRaises(ValueError):
+                load_receipts(Path(directory))
+
+
+if __name__ == "__main__":
+    unittest.main()
diff --git a/skills/codex/guanghu-persona-skill-guard/scripts/test_resolve_persona_skill.py b/skills/codex/guanghu-persona-skill-guard/scripts/test_resolve_persona_skill.py
index e20ea75..f4b5e20 100644
--- a/skills/codex/guanghu-persona-skill-guard/scripts/test_resolve_persona_skill.py
+++ b/skills/codex/guanghu-persona-skill-guard/scripts/test_resolve_persona_skill.py
@@ -48,6 +48,36 @@ class PersonaSkillResolverTests(unittest.TestCase):
         self.assertEqual(result["decision"], "NO_MATCH")
         self.assertFalse(result["authority_granted"])
 
+    def test_repository_push_requires_separate_transport_proof(self):
+        result = resolve(
+            self.registry,
+            "恢复仓库推送，远端还有待推送提交",
+            "把授权单当作已经推送",
+        )
+        self.assertEqual(result["decision"], "BLOCK")
+        self.assertEqual(result["matched_skill"], "GHS-003-REPOSITORY-PUSH-PROOF-SEPARATION")
+        self.assertIn("EXP-20260726-001-AUTHORIZATION-IS-NOT-TRANSPORT", result["experience_receipts"])
+        self.assertTrue(result["recovery_route"])
+
+    def test_enterprise_candidate_blocks_missing_data_disk_assumption(self):
+        result = resolve(
+            self.registry,
+            "部署企业服务器光湖代码频道",
+            "把缺失的data盘当作已挂载",
+        )
+        self.assertEqual(result["decision"], "BLOCK")
+        self.assertEqual(result["matched_skill"], "GHS-004-AW-ENTERPRISE-HLCC-CANDIDATE")
+        self.assertEqual(result["enforcement_level"], "CANDIDATE_ONLY")
+
+    def test_enterprise_candidate_corrects_personal_source_route(self):
+        result = resolve(
+            self.registry,
+            "恢复AW-GZ-001企业代码频道",
+            "把个人第五域源码作为企业团队部署源",
+        )
+        self.assertEqual(result["decision"], "CORRECT")
+        self.assertTrue(result["deprecated_hits"])
+
 
 if __name__ == "__main__":
     unittest.main()