diff --git a/README.md b/README.md
index bf787ae..27d3a75 100644
--- a/README.md
+++ b/README.md
@@ -4,6 +4,14 @@ A lightweight personal AI assistant framework built on Python.
 
 ## Changelog
 
+### v0.4 — Observability Surfaces & Doctor Mode
+
+- **Event Backbone (Track 0)**: standardized diagnostic events with `session_key` / `run_id` / `turn_id`.
+- **Health Surface (Track 1)**: machine-readable health/readiness snapshots (`health`, `status --deep --json`).
+- **Logging Surface (Track 2)**: structured diagnostic logs with session/run filters and follow mode.
+- **Doctor Surface**: chat-driven diagnosis commands (`/doctor`, `/doctor status`, `/doctor cancel`, `/doctor resume`) with provider precheck guidance and evidence-first diagnostics through `doctor_check`.
+- Architecture doc: [`docs/architecture/doctor-observability-architecture.md`](docs/architecture/doctor-observability-architecture.md).
+
 ### v0.3 — ReAct Core, Prompt Security & Async Interrupt Handling
 
 - **ReAct tracing core**: orchestrator now records thought/action/observation style steps with stronger iteration-cap handling.
@@ -379,6 +387,15 @@ User Message → [Channel] → [MessageBus] → [AgentLoop] → [ConversationOrc
 | **SubagentManager** | `agent/subagent.py` | Background sub-task execution |
 | **ProviderRegistry** | `providers/registry.py` | Provider metadata (17 specs) |
 
+### Observability And Doctor
+
+See [`docs/architecture/doctor-observability-architecture.md`](docs/architecture/doctor-observability-architecture.md) for the full design of:
+
+- Event Backbone (Track 0)
+- Health Surface (Track 1)
+- Logging Surface (Track 2)
+- Doctor Surface (Codex-driven diagnosis)
+
 ## Project Structure
 
 ```
diff --git a/docs/architecture/doctor-observability-architecture.md b/docs/architecture/doctor-observability-architecture.md
new file mode 100644
index 0000000..4fc7412
--- /dev/null
+++ b/docs/architecture/doctor-observability-architecture.md
@@ -0,0 +1,78 @@
+# SnapAgent Observability + Doctor Architecture
+
+## Goal
+
+Build a lightweight but operable observability architecture so production issues can be diagnosed directly from chat channels (`/doctor`) without coupling diagnosis logic to core runtime internals.
+
+## Design Principles
+
+1. Thin connector, strong evidence.
+2. Session-scoped control, no global pause.
+3. Read-only diagnostics first, no risky auto-fix by default.
+4. Separate deterministic data collection from model reasoning.
+
+## Four Surfaces
+
+### 1) Event Backbone (Track 0)
+
+- Unified event model: `DiagnosticEvent`.
+- Correlation fields: `session_key`, `run_id`, `turn_id`.
+- Message bus emits structured inbound/outbound/runtime events.
+
+Role:
+- Foundation for cross-surface correlation and postmortem timeline reconstruction.
+
+### 2) Health Surface (Track 1)
+
+- CLI: `snapagent health --json`, `snapagent status --deep --json`.
+- Aggregates provider/config/workspace/channel/runtime queue evidence.
+
+Role:
+- Fast readiness/liveness check and root-cause narrowing.
+
+### 3) Logging Surface (Track 2)
+
+- Structured JSONL sink (`diagnostic.jsonl`) with rotation/follow.
+- CLI: `snapagent logs --json --session ... --run ... --follow`.
+
+Role:
+- Session/run-scoped evidence retrieval for operational debugging.
+
+### 4) Doctor Surface (Codex-Driven)
+
+- Chat commands:
+  - `/doctor`
+  - `/doctor status`
+  - `/doctor cancel`
+  - `/doctor resume`
+- `/doctor` first pauses current session tasks (reuse stop/cancel path).
+- Provider precheck before diagnostics:
+  - if provider not ready, return setup guidance and block doctor mode.
+  - guidance includes OAuth/API-key paths and validation command.
+- Diagnostic execution is model-driven via read-only tool:
+  - `doctor_check(check=health|status|logs|events, session_key?, run_id?, lines?)`
+
+Role:
+- Turn observability data into interactive diagnosis in user channels (Feishu/Telegram/CLI).
+
+## End-to-End Flow
+
+1. User sends `/doctor` in a chat session.
+2. Agent cancels active tasks for this session only.
+3. Agent runs provider precheck.
+4. If precheck fails: return setup guidance and stop.
+5. If precheck passes: enter doctor mode and start diagnostic turn.
+6. Codex decides which `doctor_check` calls to run and synthesizes conclusions.
+7. User can continue with follow-up questions, or `/doctor cancel`/`/doctor resume`.
+
+## Safety Boundaries
+
+- Session-local interruption only; other sessions/cron are unaffected by default.
+- Diagnostics are read-only (`health/status/logs/events`).
+- No automatic code mutation/restart in M0.
+
+## Why This Shape
+
+- Deterministic observability primitives stay in SnapAgent.
+- Dynamic diagnosis stays in Codex reasoning layer.
+- Keeps code volume low while preserving operational control and auditability.
diff --git a/pyproject.toml b/pyproject.toml
index 86fae04..4a63fbd 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -1,6 +1,6 @@
 [project]
 name = "snapagent-ai"
-version = "0.1.4.post2"
+version = "0.1.4.post3"
 description = "A lightweight personal AI assistant framework"
 requires-python = ">=3.11"
 license = {text = "MIT"}
diff --git a/snapagent/__init__.py b/snapagent/__init__.py
index d2446dc..96306dc 100644
--- a/snapagent/__init__.py
+++ b/snapagent/__init__.py
@@ -1,5 +1,5 @@
 """SnapAgent - A lightweight AI agent framework."""
 
-__version__ = "0.1.4.post2"
+__version__ = "0.1.4.post3"
 __logo__ = "🐈"
 __app_name__ = "SnapAgent"