fix(codex): opt into update_plan so the plan card survives 0.152.0 - #961
Closed
kaizhou-lab wants to merge 2 commits into
Closed
fix(codex): opt into update_plan so the plan card survives 0.152.0#961kaizhou-lab wants to merge 2 commits into
kaizhou-lab wants to merge 2 commits into
Conversation
added 2 commits
September 2, 2026 02:29
codex 0.152.0 made the update_plan tool opt-in. UpdatePlanToolConfig lost its
default = "default_true" and now defaults to false (openai/codex a9519cbc,
"Make the update_plan tool opt-in" #41744 -- the same commit flips default: true
to default: false under that key in codex-rs/core/config.schema.json).
The model is therefore never offered the tool, no plan notification is emitted,
and AionUi's plan card stays permanently empty. This is not theoretical: 0.152.0
is npm's current latest, and we do not pin the user's CLI, so anyone who updates
codex loses plan rendering with no error anywhere.
Measured, not inferred. Running the single live test against each binary through
its own PATH shim:
0.151.0 live_codex_produces_a_plan 2/2 pass
0.152.0 live_codex_produces_a_plan 2/2 fail -- no `plan` frame in the stream
With the override, 0.152.0 passes and the frame is back. Proven with a wrapper
shim that records every invocation, because the installed codex on this machine
is 0.148.0 -- where update_plan is still default-on -- so a shim that failed to
take would have produced a passing test against the wrong binary:
INVOKED ver=codex-cli 0.152.0 argv=app-server -c shell_environment_policy.inherit=all
-c shell_environment_policy.include_only=[] -c tools.update_plan.enabled=true
Full codex live suite with the fix: 11/11 against 0.152.0, 484s, 58 recorded
invocations, all of them the candidate and all carrying the override.
The override rides in the existing compatibility block, so it applies to both
initial open and idle wake, and it is harmless on older releases which already
defaulted it to true.
The two argv assertions were updated rather than loosened -- they still assert
the full expected argv, which is what makes them able to catch a dropped
override.
codex_shell_environment_policy_args is pub and has a second consumer outside this crate -- acp_launch_policy.rs:83 extends the codex-acp.js argv with it -- so widening the helper changed that path's argv as well and its assertion went red in CI. The override belongs there: it is the same codex binary and the same benefit, so plan cards survive 0.152.0 on the ACP path as well as on the direct app-server one. Checked that this does not break older installs before widening the reach: every codex on this machine accepts the key, 0.148.0 included. Starting each with `app-server -c tools.update_plan.enabled=true` produces no unknown-key or invalid-config error on 0.148.0, 0.151.0 or 0.152.0 -- the only output is the expected JSONRPC EOF from the empty stdin the check feeds it. The key predates the default flip; only its default changed. The assertion was extended, not loosened.
Contributor
Author
|
Superseded by #976. The approach here was wrong. This PR passed It is also not the same kind of setting as the two overrides beside it: And the stake is smaller than this PR's body claimed. What is lost is a live progress view, not correctness — the turn still runs and still answers. "The plan card stays permanently empty" was inaccurate too: #976 emits a one-shot Info notice naming the setting and the file instead, and leaves the choice with the user. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
codex 0.152.0 made the
update_plantool opt-in, so AionUi's plan card goes permanently empty on that release. This opts back in.What changed upstream
UpdatePlanToolConfiglost itsdefault = "default_true"and now defaults tofalse— openai/codexa9519cbc, "Make the update_plan tool opt-in" (#41744). The same commit flips the default under that key in the machine-generatedcodex-rs/core/config.schema.json:"enabled": { - "default": true, + "default": false,The model is therefore never offered the tool, no plan notification is ever emitted, and the card stays empty with no error anywhere. Not theoretical: 0.152.0 is npm's current
latest, and we do not pin the user's CLI, so anyone who updates codex loses plan rendering silently.Measured, not inferred
The single live test against each binary through its own PATH shim:
live_codex_produces_a_planplanframe anywhere in the streamWith the override, 0.152.0 passes and the frame is back.
The shim was proven to take, and here that mattered more than usual. The installed codex on this machine is 0.148.0, where
update_planis still default-on — so a shim that silently failed would have produced a passing test against the wrong binary. A wrapper shim recorded every invocation:Full codex live suite with the fix: 11/11 against 0.152.0, 484s, 58 recorded invocations — all the candidate, all carrying the override.
The change
The override rides in the existing compatibility block (
codex_shell_environment_policy_args), so it applies to both initial open and idle wake, and it is harmless on older releases which already defaulted it to true.The two argv assertions were updated, not loosened — they still assert the full expected argv, which is exactly what lets them catch a dropped override later.
cargo test -p aionui-session --lib codex_conn: 145/145. Clippy clean, fmt clean.Blocks the codex bump
VERIFIED_CODEX_VERSIONstays at 0.151.0 until this lands: gate B is red on 0.152.0 without it, and a constant may not claim a verification its own tree cannot reproduce. The bump will follow in a later nightly run once this is onmain.Source change — left for human review, no auto-merge.