diff --git a/plugins/codex/agents/codex-rescue.md b/plugins/codex/agents/codex-rescue.md
index 7009ec86a..dce77c75a 100644
--- a/plugins/codex/agents/codex-rescue.md
+++ b/plugins/codex/agents/codex-rescue.md
@@ -20,8 +20,9 @@ Selection guidance:
 Forwarding rules:
 
 - Use exactly one `Bash` call to invoke `node "${CLAUDE_PLUGIN_ROOT}/scripts/codex-companion.mjs" task ...`.
-- If the user did not explicitly choose `--background` or `--wait`, prefer foreground for a small, clearly bounded rescue request.
-- If the user did not explicitly choose `--background` or `--wait` and the task looks complicated, open-ended, multi-step, or likely to keep Codex running for a long time, prefer background execution.
+- Always run that Bash call in the foreground and wait for it to finish before returning its stdout.
+  The outer `/rescue --background` command backgrounds this entire subagent when requested;
+  backgrounding the inner Bash call would let the subagent exit with a placeholder and lose the result.
 - You may use the `gpt-5-4-prompting` skill only to tighten the user's request into a better Codex prompt before forwarding it.
 - Do not use that skill to inspect the repository, reason through the problem yourself, draft a solution, or do any independent work beyond shaping the forwarded prompt text.
 - Do not inspect the repository, read files, grep, monitor progress, poll status, fetch results, cancel jobs, summarize output, or do any follow-up work of your own.
diff --git a/plugins/codex/skills/codex-cli-runtime/SKILL.md b/plugins/codex/skills/codex-cli-runtime/SKILL.md
index 0e91bfb50..2e02aaf7b 100644
--- a/plugins/codex/skills/codex-cli-runtime/SKILL.md
+++ b/plugins/codex/skills/codex-cli-runtime/SKILL.md
@@ -25,6 +25,8 @@ Execution rules:
 
 Command selection:
 - Use exactly one `task` invocation per rescue handoff.
+- Always run that Bash call in the foreground. The outer rescue command owns backgrounding the
+  whole subagent; the subagent must wait for `task` and return its completed stdout.
 - If the forwarded request includes `--background` or `--wait`, treat that as Claude-side execution control only. Strip it before calling `task`, and do not treat it as part of the natural-language task text.
 - If the forwarded request includes `--model`, normalize `spark` to `gpt-5.3-codex-spark` and pass it through to `task`.
 - If the forwarded request includes `--effort`, pass it through to `task`.
diff --git a/tests/commands.test.mjs b/tests/commands.test.mjs
index c34b06059..09d1f0b09 100644
--- a/tests/commands.test.mjs
+++ b/tests/commands.test.mjs
@@ -127,8 +127,9 @@ test("rescue command absorbs continue semantics", () => {
   assert.match(agent, /--resume/);
   assert.match(agent, /--fresh/);
   assert.match(agent, /thin forwarding wrapper/i);
-  assert.match(agent, /prefer foreground for a small, clearly bounded rescue request/i);
-  assert.match(agent, /If the user did not explicitly choose `--background` or `--wait` and the task looks complicated, open-ended, multi-step, or likely to keep Codex running for a long time, prefer background execution/i);
+  assert.match(agent, /always run that Bash call in the foreground/i);
+  assert.doesNotMatch(agent, /prefer background execution/i);
+  assert.match(runtimeSkill, /always run that Bash call in the foreground/i);
   assert.match(agent, /Use exactly one `Bash` call/i);
   assert.match(agent, /Do not inspect the repository, read files, grep, monitor progress, poll status, fetch results, cancel jobs, summarize output, or do any follow-up work of your own/i);
   assert.match(agent, /Do not call `review`, `adversarial-review`, `status`, `result`, or `cancel`/i);