Codex 5.5: pin the model before you swap it

Yes, you can run GPT-5.5 in Codex without breaking your review loop, but write down the swap before you make it. Codex CLI is OpenAI's coding agent, and a model pin note is the two-line AGENTS.md entry that fixes your team's default model and the rule for when a different one may step in. That note is the whole difference between a swap your leads can reason about and a surprise they discover at merge time.

The question almost always shows up the same way. Someone proposes moving Codex to GPT-5.5 the week of a release cut, and review expectations start to wobble before the first diff lands. The fix is small. Put the decision in the file the whole team already reads.

Pin the model in AGENTS.md first

Different models carry different risk appetite, so a casual swap quietly changes what review is catching. The repair is two lines, written before the swap branch exists.

# AGENTS.md: model governance

- Default model: GPT-5.5 (set 2026-05-02; review for code-review depth).
- Escalation rule: a smaller/faster model may handle docs + scaffolding,
  but security-touching diffs run on the default and need a human ack.

Once that lives in AGENTS.md, a lead can think about blast radius before the swap instead of meeting an unfamiliar diff cold. Casual swaps that always end fine become the new normal, and the day your risk appetite actually changes, nobody has it written down. The pin is how you keep that from happening.

Keep every run ending in a replayable receipt

The swap-week temptation is to skip tests because everyone wants to watch the new model move. Regressions come back quietly when that happens. So make the proof part of the ritual: every Codex CLI run ends with a transcript snippet a reviewer can replay.

# AGENTS.md: verification

- Every Codex CLI run ends with the transcript snippet reviewers can replay.
- Pair browser evidence with the project's normal CLI checks before merge.
- If MCP servers are enabled, list allowed actions beside each connector name.

This is also how you answer the hardest question a branch can face: which instruction owned the final result? A branch that looks complete but cannot say which AGENTS.md or SKILL.md rule governed its behavior is the failure mode worth designing against. The transcript and the rule reference together carry that answer.

Bridge browser checks to CLI receipts

Under a new model, demos are the first thing people reach for, so Chrome workflows drift from CLI habits and reviewers end up looking at two different truths. The browser bridge note keeps them paired. Document the staging URLs and credential boundaries right beside the browser tasks, so a demo cannot contradict a CI artifact.

Slash commands wander too, since operators improvise during swaps. Keep a docs/codex-commands.md linked from AGENTS.md, and point it at the slash commands reference for the built-ins so the team converges on one set of moves instead of private folklore.

Run the merge gates before a model graduates

Before any model change ships, walk these four questions. They are the same gates whether the change is a model swap or a version bump.

A quick checklist you can paste into a PR template:

• Forked agent work lists parent and child responsibilities.
• Red-folder paths received explicit human acknowledgement.
• Scopes in the PR body match the folders in the diff.
• Primary-doc links were smoke-checked after publishing edits.

The Codex CLI docs and features page describe what each build supports, openai/codex and openai/skills show where it is heading, and the MCP specification anchors the connector language. None of it replaces your own architecture judgement. Agents speed up execution, not ownership.

Common questions

• What does codex 5.5 mean for a team workflow?

  It means running GPT-5.5 inside Codex, and the workflow answer is model governance. Pin the default model and an escalation rule inside AGENTS.md before anyone swaps. Casual swaps make review expectations wobble, because different models imply different risk appetite, and without the pin your leads lose track of blast radius right when they need it most.

• How do we adopt a new model without breaking review?

  Use the model pin note. AGENTS.md fixes the default model, and the escalation rule says when a different model may step in. After the pin, a lead reasons about blast radius up front instead of discovering the swap while reading an unfamiliar diff at merge time. The decision lives in the file, not in someone's memory.

• What are browser checks in this memo?

  Browser checks are the Chrome-side verifications that stay paired with the project's normal CLI checks before merge. The browser bridge note documents staging URLs and credential boundaries beside the browser tasks. That pairing is how a demo stops contradicting a CI artifact when work runs on both rails at once during a swap.

• How do teams prove which instruction owned a diff?

  The workspace file carries the proof. Every Codex CLI run ends with a transcript snippet a reviewer can replay, and the rules-precedence gate asks which AGENTS.md or SKILL.md rule governed the behavior. A branch that looks complete but cannot answer that question is exactly the failure this memo is built to catch.

Where to go next

Model governance is one afternoon: pin, escalation rule, and browser bridge installed on your own repo, with the broader practice under Codex CLI workflows. Bring the repo you are about to swap to our training and we will set it up together.