diff --git a/packages/server/src/server/agent/providers/codex-agent.ts b/packages/server/src/server/agent/providers/codex-agent.ts
index 3e28d1d3f..1c362054b 100644
--- a/packages/server/src/server/agent/providers/codex-agent.ts
+++ b/packages/server/src/server/agent/providers/codex-agent.ts
@@ -1660,23 +1660,39 @@ function finalizeRolloutFunctionCall(
     return;
   }
   const result = parseJsonOrObject<CommandExecutionResult>(payload.output);
-  const exitCode = result?.metadata?.exit_code;
+  const parsedOutput =
+    typeof payload.output === "string"
+      ? parseCommandOutputText(payload.output)
+      : null;
+  const exitCode = result?.metadata?.exit_code ?? parsedOutput?.exitCode;
   const status =
     exitCode === undefined || exitCode === 0 ? "completed" : "failed";
+  const stdout =
+    result?.stdout ??
+    (typeof result?.output === "string" ? result.output : undefined) ??
+    parsedOutput?.stdout;
+  const metadata =
+    result?.metadata ??
+    (typeof exitCode === "number" ? { exit_code: exitCode } : undefined);
 
   // Build structured command output
-  const output = result?.stdout
-    ? {
-        type: "command" as const,
-        command: command.command,
-        output: result.stdout,
-        exitCode,
-        cwd: command.cwd,
-        metadata:
-          result?.metadata ??
-          (typeof exitCode === "number" ? { exit_code: exitCode } : undefined),
-      }
-    : result;
+  let output: unknown;
+  if (stdout !== undefined) {
+    output = {
+      type: "command" as const,
+      command: command.command,
+      output: stdout,
+      exitCode,
+      cwd: command.cwd,
+      metadata,
+    };
+  } else if (result && typeof result === "object") {
+    output = { ...result, metadata: metadata ?? result.metadata };
+  } else if (metadata) {
+    output = { metadata };
+  } else {
+    output = result;
+  }
 
   events.push({
     type: "timeline",
@@ -2116,6 +2132,32 @@ function parseJsonOrObject<T = unknown>(value: unknown): T | null {
   return null;
 }
 
+function parseCommandOutputText(
+  output: string
+): { exitCode?: number; stdout?: string } | null {
+  if (!output.trim()) {
+    return null;
+  }
+
+  const exitCodeMatch = output.match(/Exit code:\s*(\d+)/i);
+  const exitCode =
+    exitCodeMatch && exitCodeMatch[1]
+      ? Number.parseInt(exitCodeMatch[1], 10)
+      : undefined;
+
+  const outputMatch = output.match(/\nOutput:\s*\n([\s\S]*)$/);
+  const stdout = outputMatch ? outputMatch[1].replace(/\s+$/, "") : undefined;
+
+  if (exitCode === undefined && stdout === undefined) {
+    return null;
+  }
+
+  return {
+    exitCode: Number.isNaN(exitCode as number) ? undefined : exitCode,
+    stdout,
+  };
+}
+
 function formatCommand(args: unknown): string | null {
   if (!args || typeof args !== "object") {
     return null;
diff --git a/plan.md b/plan.md
index 9db0572fd..ca07ea08c 100644
--- a/plan.md
+++ b/plan.md
@@ -175,29 +175,29 @@ Build a new Codex MCP provider side‑by‑side with the existing Codex SDK prov
   - Decide: keep MCP for tools, or drop it and use SDK only?
   - **Done (2025-12-24 20:04)**: Concluded MCP provider is only necessary for external MCP tool integration; for permissions and core workflows, SDK provider is simpler and already supports approvals. Recommend dropping/parking MCP unless external MCP tool usage is a requirement.
 
-- [ ] **CRITICAL FINDING**: `codex exec` (SDK) is non-interactive - cannot prompt!
-
-  - Ran `scripts/codex-sdk-permission-test.mjs` with workspace-write + untrusted
-  - NO approval events fire - commands just run or refuse
-  - `codex exec --experimental-json` mode is NON-INTERACTIVE by design
-  - It either refuses ("approvals are disabled") or runs in sandbox
-  - The interactive `codex` CLI is what prompts - but SDK uses `codex exec`
-  - This explains why SDK never emits `exec_approval_request` events
-  - MCP server (`codex mcp-server`) may work differently via MCP elicitation
-  - The SDK provider CANNOT support interactive approvals by design
-
 - [x] **Plan**: Re-audit based on test results.
 
   - **Done (2025-12-24 20:06)**: Reviewed latest E2E failures/hang; added fix tasks for Codex SDK hydration and Codex MCP abort hang, plus a retest task.
 
-- [ ] **Fix**: Codex SDK persistence hydration should emit completed shell_command tool entries.
+- [x] **Fix**: Codex SDK persistence hydration should emit completed shell_command tool entries.
 
   - Capture the failing rollout entry and ensure hydrated tool calls include completed status + exit code metadata.
+  - **Done (2025-12-24 20:12)**: Parsed rollout command output strings to extract exit codes/stdout and attach metadata to hydrated shell tool results.
 
 - [ ] **Fix**: Codex MCP E2E hang in long-running command abort test.
 
   - Add deterministic abort/timeout handling and ensure the session closes even if the sleep tool call is never surfaced.
 
+- [ ] **CRITICAL FINDING (VERIFIED IN SOURCE)**: `codex exec` IGNORES approval events!
+
+  - In `codex-rs/exec/src/event_processor_with_human_output.rs:568`:
+    - `ExecApprovalRequest` and `ApplyPatchApprovalRequest` are in ignore match arm `=> {}`
+  - `codex exec` receives approval events but DOESN'T emit them as JSON
+  - SDK uses `codex exec` → CANNOT support approvals BY DESIGN
+  - MCP server DOES handle them → sends `ElicitRequest` (see `exec_approval.rs:107`)
+  - CONCLUSION: MCP provider is the ONLY path to real permissions, not SDK
+  - The agent's earlier conclusion to "park MCP" was WRONG
+
 - [ ] **Test (E2E)**: Rerun server vitest after fixes.
 
   - If failures: add follow-up fix tasks immediately after this item.