diff --git a/.openclaw-sync/source.json b/.openclaw-sync/source.json
index 444e74f40..34a95f0d7 100644
--- a/.openclaw-sync/source.json
+++ b/.openclaw-sync/source.json
@@ -1,5 +1,5 @@
 {
   "repository": "openclaw/openclaw",
-  "sha": "7b4932f961366f93c87eff1df44648071d0a41b4",
-  "syncedAt": "2026-05-02T15:54:28.907Z"
+  "sha": "563dca82f4290a8d085be5bea10334545c1c7e8d",
+  "syncedAt": "2026-05-02T16:01:12.676Z"
 }
diff --git a/docs/concepts/system-prompt.md b/docs/concepts/system-prompt.md
index 32a3504c0..cb7ae9362 100644
--- a/docs/concepts/system-prompt.md
+++ b/docs/concepts/system-prompt.md
@@ -109,6 +109,19 @@ section when the direct/group chat context already includes the resolved
 conversation-specific `NO_REPLY` behavior. This avoids repeating token mechanics
 in both the global system prompt and channel context.
 
+## Prompt snapshots
+
+OpenClaw keeps committed happy-path prompt snapshots for the Codex/message-tool
+runtime under `test/fixtures/agents/prompt-snapshots/happy-path/`. They render
+the OpenClaw-owned Codex app-server developer instructions, selected thread
+start/resume params, turn user input, and dynamic tool specs for Telegram direct,
+Discord group, and heartbeat turns. The hidden base Codex system prompt and
+turn-scoped Codex collaboration-mode instructions are owned by the Codex runtime
+and are not rendered by OpenClaw.
+
+Regenerate them with `pnpm prompt:snapshots:gen` and verify drift with
+`pnpm prompt:snapshots:check`.
+
 ## Workspace bootstrap injection
 
 Bootstrap files are trimmed and appended under **Project Context** so the model sees identity and profile context without needing explicit reads: