fix: preserve task prompts as free text

plugins/codex/agents/codex-rescue.md

@@ -19,7 +19,8 @@ Selection guidance:
 
 Forwarding rules:
 
-- Use exactly one `Bash` call to invoke `node "${CLAUDE_PLUGIN_ROOT}/scripts/codex-companion.mjs" task ...`.
+- Use exactly one `Bash` call to invoke `node "${CLAUDE_PLUGIN_ROOT}/scripts/codex-companion.mjs" task [runtime flags] -- "<raw prompt>"`.
+- Put every runtime flag before `--` and the complete user request after it as one argument. Never re-tokenize prompt text or interpret flag-looking words inside it as runtime controls.
 - 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.
 - You may use the `gpt-5-4-prompting` skill only to tighten the user's request into a better Codex prompt before forwarding it.

plugins/codex/scripts/codex-companion.mjs

@@ -79,7 +79,7 @@ function printUsage() {
       "  node scripts/codex-companion.mjs setup [--enable-review-gate|--disable-review-gate] [--json]",
       "  node scripts/codex-companion.mjs review [--wait|--background] [--base <ref>] [--scope <auto|working-tree|branch>]",
       "  node scripts/codex-companion.mjs adversarial-review [--wait|--background] [--base <ref>] [--scope <auto|working-tree|branch>] [focus text]",
-      "  node scripts/codex-companion.mjs task [--background] [--write] [--resume-last|--resume|--fresh] [--model <model|spark>] [--effort <none|minimal|low|medium|high|xhigh>] [prompt]",
+      "  node scripts/codex-companion.mjs task [--background] [--write] [--resume-last|--resume|--fresh] [--model <model|spark>] [--effort <none|minimal|low|medium|high|xhigh>] [--prompt-file <path>|-- <prompt>]",
       "  node scripts/codex-companion.mjs transfer [--source <claude-jsonl>] [--json]",
       "  node scripts/codex-companion.mjs status [job-id] [--all] [--json]",
       "  node scripts/codex-companion.mjs result [job-id] [--json]",
@@ -127,8 +127,8 @@ function normalizeReasoningEffort(effort) {
   return normalized;
 }
 
-function normalizeArgv(argv) {
-  if (argv.length === 1) {
+function normalizeArgv(argv, options = {}) {
+  if (options.splitSingleArgument !== false && argv.length === 1) {
     const [raw] = argv;
     if (!raw || !raw.trim()) {
       return [];
@@ -139,11 +139,12 @@ function normalizeArgv(argv) {
 }
 
 function parseCommandInput(argv, config = {}) {
-  return parseArgs(normalizeArgv(argv), {
-    ...config,
+  const { splitSingleArgument, ...parseConfig } = config;
+  return parseArgs(normalizeArgv(argv, { splitSingleArgument }), {
+    ...parseConfig,
     aliasMap: {
       C: "cwd",
-      ...(config.aliasMap ?? {})
+      ...(parseConfig.aliasMap ?? {})
     }
   });
 }
@@ -761,6 +762,10 @@ async function handleReview(argv) {
 
 async function handleTask(argv) {
   const { options, positionals } = parseCommandInput(argv, {
+    // A task prompt is free text, not a shell command line. Re-tokenizing a
+    // single prompt argument strips quotes/backslashes and lets prose such as
+    // "--model" or "--write" mutate runtime options.
+    splitSingleArgument: false,
     valueOptions: ["model", "effort", "cwd", "prompt-file"],
     booleanOptions: ["json", "write", "resume-last", "resume", "fresh", "background"],
     aliasMap: {

plugins/codex/skills/codex-cli-runtime/SKILL.md

@@ -9,7 +9,7 @@ user-invocable: false
 Use this skill only inside the `codex:codex-rescue` subagent.
 
 Primary helper:
-- `node "${CLAUDE_PLUGIN_ROOT}/scripts/codex-companion.mjs" task "<raw arguments>"`
+- `node "${CLAUDE_PLUGIN_ROOT}/scripts/codex-companion.mjs" task [runtime flags] -- "<raw prompt>"`
 
 Execution rules:
 - The rescue subagent is a forwarder, not an orchestrator. Its only job is to invoke `task` once and return that stdout unchanged.
@@ -25,6 +25,7 @@ Execution rules:
 
 Command selection:
 - Use exactly one `task` invocation per rescue handoff.
+- Pass runtime flags as separate arguments before `--`, then pass the complete user request as one prompt argument after `--`. This boundary keeps flag-like prose, quotes, backslashes, and newlines inside the prompt.
 - 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`.

tests/commands.test.mjs

@@ -130,6 +130,7 @@ test("rescue command absorbs continue semantics", () => {
   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, /Use exactly one `Bash` call/i);
+  assert.match(agent, /every runtime flag before `--` and the complete user request after it as one argument/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);
   assert.match(agent, /Leave `--effort` unset unless the user explicitly requests a specific reasoning effort/i);
@@ -142,6 +143,7 @@ test("rescue command absorbs continue semantics", () => {
   assert.match(agent, /only to tighten the user's request into a better Codex prompt/i);
   assert.match(agent, /Do not use that skill to inspect the repository, reason through the problem yourself, draft a solution, or do any independent work/i);
   assert.match(runtimeSkill, /only job is to invoke `task` once and return that stdout unchanged/i);
+  assert.match(runtimeSkill, /Pass runtime flags as separate arguments before `--`/i);
   assert.match(runtimeSkill, /Do not call `setup`, `review`, `adversarial-review`, `status`, `result`, or `cancel`/i);
   assert.match(runtimeSkill, /use the `gpt-5-4-prompting` skill to rewrite the user's request into a tighter Codex prompt/i);
   assert.match(runtimeSkill, /That prompt drafting is the only Claude-side work allowed/i);
@@ -192,7 +194,7 @@ test("internal docs use task terminology for rescue runs", () => {
   const promptingSkill = read("skills/gpt-5-4-prompting/SKILL.md");
   const promptRecipes = read("skills/gpt-5-4-prompting/references/codex-prompt-recipes.md");
 
-  assert.match(runtimeSkill, /codex-companion\.mjs" task "<raw arguments>"/);
+  assert.match(runtimeSkill, /codex-companion\.mjs" task \[runtime flags\] -- "<raw prompt>"/);
   assert.match(runtimeSkill, /Use `task` for every rescue request/i);
   assert.match(runtimeSkill, /task --resume-last/i);
   assert.match(promptingSkill, /Use `task` when the task is diagnosis/i);

tests/runtime.test.mjs

@@ -193,6 +193,56 @@ test("task runs without auth preflight so Codex can refresh an expired session",
   assert.match(result.stdout, /Handled the requested task/);
 });
 
+test("task preserves a single free-text prompt without parsing its contents as options", () => {
+  const repo = makeTempDir();
+  const binDir = makeTempDir();
+  const statePath = path.join(binDir, "fake-codex-state.json");
+  installFakeCodex(binDir);
+  initGitRepo(repo);
+  fs.writeFileSync(path.join(repo, "README.md"), "hello\n");
+  run("git", ["add", "README.md"], { cwd: repo });
+  run("git", ["commit", "-m", "init"], { cwd: repo });
+
+  const prompt = [
+    "Review the plan: swap to claude -p --model claude-haiku --output-format json.",
+    String.raw`Config lives at C:\Users\me\Projects\my-app\config\app-config.json.`,
+    'The log line was "RUN ERROR: exit 1: Not logged in" and then --write the summary.'
+  ].join("\n");
+  const result = run(process.execPath, [SCRIPT, "task", prompt], {
+    cwd: repo,
+    env: buildEnv(binDir),
+    shell: false
+  });
+
+  assert.equal(result.status, 0, result.stderr);
+  const fakeState = JSON.parse(fs.readFileSync(statePath, "utf8"));
+  assert.equal(fakeState.lastTurnStart.prompt, prompt);
+  assert.equal(fakeState.lastTurnStart.model, null);
+});
+
+test("task keeps runtime options before the explicit prompt boundary", () => {
+  const repo = makeTempDir();
+  const binDir = makeTempDir();
+  const statePath = path.join(binDir, "fake-codex-state.json");
+  installFakeCodex(binDir);
+  initGitRepo(repo);
+  fs.writeFileSync(path.join(repo, "README.md"), "hello\n");
+  run("git", ["add", "README.md"], { cwd: repo });
+  run("git", ["commit", "-m", "init"], { cwd: repo });
+
+  const prompt = "Explain why --model claude-haiku must stay in the prompt.";
+  const result = run(process.execPath, [SCRIPT, "task", "--model", "spark", "--", prompt], {
+    cwd: repo,
+    env: buildEnv(binDir),
+    shell: false
+  });
+
+  assert.equal(result.status, 0, result.stderr);
+  const fakeState = JSON.parse(fs.readFileSync(statePath, "utf8"));
+  assert.equal(fakeState.lastTurnStart.prompt, prompt);
+  assert.equal(fakeState.lastTurnStart.model, "gpt-5.3-codex-spark");
+});
+
 test("transfer delegates the current Claude session directly to native import", () => {
   const home = makeTempDir();
   const repo = path.join(home, "repo");