codex - 💡(How to fix) Fix In planning mode Codex seems to fail to call the question asking tool despite being reminded it exists, causing it to loop because its waiting for answers. 2x occurances, there seems to be no failsafe so wiping out usage quota.

What version of Codex CLI is running?

codex-cli 0.130.0

What subscription do you have?

Business Plan

Which model were you using?

5.5 high

What platform is your computer?

Linux 6.17.9-76061709-generic x86_64 x86_64

What terminal emulator and version are you using (if applicable)?

VS Code Terminal Shell integration: Rich

Codex doctor report

not available

What issue are you seeing?

Terminal status shows "Ready | Tasks 4/5". Planning mode essentially failed to asked to use the questions tool in planning mode, and looped. The initial prompt specifically said " howevr you can only ask 3 questions at a time with the planning tool. I'd much prefer 3qs + update, 3 qs + uipdate etc format, as it means we dont lose track." [sic]

Note the below ******* out sections are sanitation I've added for issue submission. It is in chronological order, there is obviously much more BEFORE the quote starts, but its just a standard session.

Initial error appearance:

No files were edited and no tests were run; this was a read-only plan/code/docs verification pass. ─────────────────────────────────────────────────────────────────────────────────────────────────────── • At this point the remaining work is not more file discovery; it is decision locking. I’m doing the requested completion audit now so the blocker is explicit against the original objective, not just a loose “waiting for answers.” • Updated Plan └ Repo and Supabase-doc verification is done enough for read-only review. The only blocker to decision-complete implementation planning is the unresolved user intent batch around****************** ✔ Inspect plan and current a****************** ✔ Ask fi****************** ✔ Verify current ****************** ✔ Inspect********** □ Produce concise decision-complete review and implementation-ready plan outline

Completion Audit

Objective deliverables:

• Review ***
• Verify against current repo code and current Supabase docs for **********.
• Inspect ***
• Produce confirmed-correct decisions, needed corrections, open decisions, and implementation-ready outline.
• Ask only unresolved intent questions, max 3 at a time.

Covered Evidence

• Plan inspected: yes.
• Supabase docs checked: yes, via MCP where available and official .md docs/web fallback.
• Repo ********** inspected: ***
• ********** inspected: ***
• RLS/******************
• ********** inspected: ***
• AGENTS drafts inspected as intent references only.
• Live Supabase advisor refresh: blocked. MCP tool discovery exposed advisors, but invocation failed as Unknown tool: supabase_get_advisors.

Status

The review is not decision-complete yet. The implementation outline depends on 3 unresolved decisions:

1. **************** shape: ***
2. Your “****************” rule: ***
3. *************fallback

Until those are answered, the secure plan cannot be finalized without guessing. I’m not marking the goal complete. • I’m blocked on the same three intent decisions. The repo and Supabase-doc evidence is already enough to say what must change, but not enough to choose the implementation-safe migration shape without guessing.

Please answer these three:

1.**************** shape: - ****************** - ****************** - ****************** 2. “**************” means: - n**************** -****************s - ****************** 3. fallback: - ****************** - ****************** - f

Once those are answered, I can finalize the confirmed decisions, corrections, and implementation- ready plan outline. • Still blocked on the same three decisions. I can’t finalize the plan safely without choosing those migration semantics, and the repo/Supabase docs cannot answer them for you.

Needed answers:

1.****************** 2. ****************** 3. ****************** • Blocked on user input. I need your answers to the 3 open decisions before the plan can become decision-complete:

1.****************** 2. ****************** 3. ****************** • I’m stopped at the required decision point. The review cannot become implementation-safe until you answer:

1.****************** 2. ****************** 3. ****************** • Blocked on the same 3 required user decisions:

1.****************** 2. ****************** 3. ****************** • I can’t progress without user input. The exact missing decisions are:

1.****************** 2. ****************** 3. ****************** • I’m blocked on user input. The plan cannot be finalized until these are answered:

The above "1 2 3 I'm blocked until" sections happened approx 50 times before:

Context compacted • I’m blocked on user intent, not repo evidence. The Supabase/current-code review is far enough to say the plan is directionally right, but it is not decision-complete until these three choices are fixed:

1. S**********

   ---

   O******************
2. “H********** ******************

---

1. ---

---

---

Once those are answered, I can give the final decision-complete review: confirmed-correct decisions, required corrections to the plan, remaining evidence gates, and the implementation-ready phase outline. I’m not marking the goal complete yet because these are still open design decisions. • Completion audit says the goal is still active, not complete.

I have the repo/code/Supabase evidence needed for the review, but the plan cannot be made decision- complete until these unresolved user-intent decisions are answered:

1. ********** shape: u******************
2. ********** ban:******************
3. ********** fallback:******************

I’m stopping here because those are not inferable from Supabase docs or current repo code, and choosing them myself would bake policy into the plan.

This was the point it ran out of usage. I am confident anyone on a high usage plan that hadn't noticed this running, would have had their entire usage wiped out. As mentioned before this is the second time this has happened to me.

What steps can reproduce the bug?

Uploaded thread: 019e3666-f4c5-7af1-9edb-5ed309e25abf

What is the expected behavior?

The planning mode to use the question tool correctly, clearly some kind of failsafe to stop this kind of loop happening.

Additional information

This has happened both inside the new "goal" mode, and happened yesterday (no log sorry) without goal mode usage.