26% context use from first message

Hmm, this is definitely unexpected.

When I try to replicate your usage (gpt-5.1-codex-mini medium with no AGENTS.md), I see 95% left after the first message.

Do you have any non-default settings configured?

I see that you have several MCPs configured. Depending on how an MCP is written, it can consume significant context. But if you haven't changed or updated your MCPs in a long time, that doesn't seem like it would be the problem.

You mentioned that you recently observed a behavior change. Do you suspect that it might be a client-side regression in 0.93? If so, does downgrading to 0.92 change the behavior?

019c17eb-b715-7912-85ee-afd07486bbfc i temporarily deleted all MCPs tried in a completely different folder, still exactly the same:

│ >_ OpenAI Codex (v0.93.0)                   │
│                                             │
│ model:     gpt-5.2-codex   /model to change │
│ directory: /mnt/e/cox-t                     │
╰─────────────────────────────────────────────╯

  Tip: Press Tab to queue a message instead of sending it immediately; Enter always sends immediately.

/mcp

🔌  MCP Tools

  • No MCP servers configured.
    See the MCP docs to configure them.

• Model changed to gpt-5.1-codex-mini medium


› Hi


• Quick hello

  - Hi there! Let me know what you’d like to work on—code change, review, or something else.

• Feedback uploaded. Please open an issue using the following URL:

  https://github.com/openai/codex/issues/new?template=2-bug-report.yml&steps=Uploaded%20thread:%20019c17eb-b715-7912-85ee-afd07486bbfc

  Or mention your thread ID 019c17eb-b715-7912-85ee-afd07486bbfc in an existing issue.

• Model changed to gpt-5.2-codex medium


› Improve documentation in @filename

  ? for shortcuts                                                                                                                                                                                                                                                 

74% context left

I tried reverting then to 0.92 (019c180a-cf05-7d82-97e9-97e7517182b2 ) and it doesnt happen:

╭─────────────────────────────────────────────╮
│ >_ OpenAI Codex (v0.92.0)                   │
│                                             │
│ model:     gpt-5.2-codex   /model to change │
│ directory: /mnt/e/empty                     │
╰─────────────────────────────────────────────╯

  Tip: Start a fresh idea with /new; the previous session stays in history.


› hi


• Hi! What can I help with?


› Improve documentation in @filename

  
100% context left · ? for shortcuts

But when using
_tmp="/tmp/codex-0930-$$"; WORKDIR="/mnt/e/empty"; mkdir -p "$tmp/codex-home" "$tmp/npm-cache" "$tmp/npm-prefix"; cd "$WORKDIR" && CODEX_HOME="$tmp/codex-home" NPM_CONFIG_CACHE="$tmp/npm-cache" NPM_CONFIG_PREFIX="$tmp/npm-prefix" npm exec --yes --package @openai/codex@0.93.0 --
codex --config 'cli_auth_credentials_store="file"'; rm -rf "$tmp"_

to create a completely isolated fresh install of 0.93 it again works, so the problem must be on my end, but for the life of me I cannot understand why.
Should I just nuke everything, reinstall and just copy over my current config.toml will that resolve it?

Looking at your session details and logs, I don't have any good working theories as to the cause.

Yeah, I think I'd try reinstalling to see if that fixes it. Make sure you don't have CODEX_HOME pointing to an unintended location. Make sure you don't have a project-local .codex/config.toml that's overriding values in your user-level config.