Gate Codex skills flag on CLI support by YusukeShimizu · Pull Request #6520 · block/goose

YusukeShimizu · 2026-01-15T21:51:14Z

Summary

Addresses compatibility issues with newer versions of the Codex CLI where the skills feature flag has been removed (referenced in openai/codex#8850 ).

Previously, goose configure would fail with an exit code 1 on newer CLIs because it unconditionally passed the --enable/--disable skills flags.

While simply removing the flag checks was an option, this implementation is explicitly designed to maintain backward compatibility. This ensures stability for automation workflows and environments where CLI versions may be pinned.

Type of Change

AI Assistance

This PR was created or reviewed with AI assistance

Testing

This change has been tested manually with both new and legacy versions of the Codex CLI to ensure robustness.

Screenshots/Demos (for UX changes)

Before (Crash on Codex CLI 0.85.0):

◇  Enter a model from that provider:
│  gpt-5.2-codex
│
◇  Request failed: Codex command failed with exit code: Some(1)
│
└  Failed to configure provider: init chat completion request with tool did not succeed.

Before (Success on Codex CLI 0.77.0):

/opt/homebrew/bin/codex --version
codex-cli 0.77.0

goose configure
...
◇  Enter a model from that provider:
│  gpt-5.2-codex
│
◐  Checking your configuration...                                                                                                                                                                                                └  Configuration saved successfully to /Users/bruwbird/.config/goose/config.yaml

After (Backward Compatibility on Codex CLI 0.77.0):

codex --version
codex-cli 0.77.0

goose configure
...
◐  Checking your configuration...                                                                                                                                                                                                └  Configuration saved successfully to /Users/bruwbird/.config/goose/config.yaml

The Codex CLI removed the `skills` feature flag in newer versions , causing `goose configure` to fail with exit code 1 when it unconditionally passes `--enable/--disable skills`. This commit updates the logic to check `codex features list` before applying these flags. This ensures the flags are only passed when the installed CLI explicitly advertises support for them. While simply removing the flag usage was an option, this approach is designed to maintain backward compatibility for environments where the CLI version is pinned (e.g., in automation workflows). This also updates the documentation to clarify that the `skills` flag logic is legacy behavior for older CLIs. Signed-off-by: Yusuke Shimizu <stm1051212@gmail.com>

Copilot

Pull request overview

This PR addresses a breaking change in newer versions of the Codex CLI where the skills feature flag was removed (openai/codex#8850). The implementation adds backward compatibility by detecting whether the installed Codex CLI supports the skills feature flag via codex features list and only passing --enable/--disable skills flags when supported.

Changes:

Added runtime feature detection to check if the Codex CLI supports the skills feature flag
Updated documentation to clarify that CODEX_ENABLE_SKILLS is a legacy toggle only respected on older CLIs
Added test coverage for the feature detection logic

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 3 comments.

File	Description
crates/goose/src/providers/codex.rs	Added `skills_feature_flag_supported` field, feature detection logic via `codex features list`, and conditional flag passing
documentation/docs/guides/cli-providers.md	Updated documentation to clarify skills support is native in newer CLIs and the env var is legacy-only

crates/goose/src/providers/codex.rs

Copilot · 2026-01-16T00:36:38Z

crates/goose/src/providers/codex.rs

+    fn feature_list_contains_feature(stdout: &str, feature: &str) -> bool {
+        // `codex features list` output is a whitespace-separated sequence of:
+        // `<name> <stability> <enabled>` repeated, e.g.:
+        // `undo stable false shell_tool stable true ...`
+        //
+        // We match by token to be robust to both newline- and space-delimited output.
+        stdout
+            .split_whitespace()
+            .collect::<Vec<_>>()
+            .windows(3)
+            .any(|w| {
+                w[0] == feature
+                    && matches!(w[1], "stable" | "beta" | "experimental")
+                    && matches!(w[2], "true" | "false")
+            })
+    }


Add test cases for edge scenarios: empty stdout, feature name appearing but with invalid stability/enabled values, feature name as substring of another word, and stdout with fewer than 3 tokens. The windows(3) call could panic on short input if the slice has fewer than 3 elements.

Copilot · 2026-01-16T00:36:38Z

crates/goose/src/providers/codex.rs

+        stdout
+            .split_whitespace()
+            .collect::<Vec<_>>()
+            .windows(3)
+            .any(|w| {
+                w[0] == feature
+                    && matches!(w[1], "stable" | "beta" | "experimental")
+                    && matches!(w[2], "true" | "false")
+            })


If stdout has fewer than 3 tokens after split_whitespace(), the windows(3) iterator will be empty (not panic), but this edge case should be tested. An empty or malformed output from codex features list would silently return false, which is correct behavior but should be explicitly verified in tests.

Suggested change

stdout

.split_whitespace()

.collect::<Vec<_>>()

.windows(3)

.any(|w| {

w[0] == feature

&& matches!(w[1], "stable" | "beta" | "experimental")

&& matches!(w[2], "true" | "false")

})

let mut tokens = stdout.split_whitespace();

loop {

let name = match tokens.next() {

Some(t) => t,

None => break,

};

let stability = match tokens.next() {

Some(t) => t,

None => break,

};

let enabled = match tokens.next() {

Some(t) => t,

None => break,

};

if name == feature

&& matches!(stability, "stable" | "beta" | "experimental")

&& matches!(enabled, "true" | "false")

{

return true;

}

}

false

- Verify behavior when skills feature is not supported. - Ensure correct arguments are returned for skills flag. Signed-off-by: Yusuke Shimizu <stm1051212@gmail.com>

YusukeShimizu · 2026-01-20T02:14:43Z

@michaelneale I would appreciate your review on this.

michaelneale

nice - nice catch BTW.

Also - goose now can use codex via oauth - so doens't have to use cli if you don't wnat it

DOsinga · 2026-01-28T17:11:07Z

I think we can just drop the backwards compatibility, so cloned this, merged in main now at: #6775

thanks for doing this though!

YusukeShimizu force-pushed the codex-skills-feature-flag branch from 4e2b015 to 468d8f8 Compare January 16, 2026 00:23

YusukeShimizu marked this pull request as ready for review January 16, 2026 00:34

YusukeShimizu requested a review from a team as a code owner January 16, 2026 00:34

Copilot AI review requested due to automatic review settings January 16, 2026 00:34

Copilot started reviewing on behalf of YusukeShimizu January 16, 2026 00:34 View session

Copilot AI reviewed Jan 16, 2026

View reviewed changes

codex: add tests for skills feature flag handling

feb0a1c

- Verify behavior when skills feature is not supported. - Ensure correct arguments are returned for skills flag. Signed-off-by: Yusuke Shimizu <stm1051212@gmail.com>

zanesq assigned michaelneale Jan 27, 2026

michaelneale approved these changes Jan 27, 2026

View reviewed changes

michaelneale removed their assignment Jan 28, 2026

DOsinga closed this Jan 28, 2026

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Comments

Gate Codex skills flag on CLI support#6520

Gate Codex skills flag on CLI support#6520
YusukeShimizu wants to merge 2 commits intoblock:mainfrom
YusukeShimizu:codex-skills-feature-flag

YusukeShimizu commented Jan 15, 2026 •

edited

Loading

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

Copilot AI Jan 16, 2026

Uh oh!

Copilot AI Jan 16, 2026

Uh oh!

YusukeShimizu commented Jan 20, 2026

Uh oh!

michaelneale left a comment

Uh oh!

DOsinga commented Jan 28, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

-        stdout
-            .split_whitespace()
-            .collect::<Vec<_>>()
-            .windows(3)
-            .any(|w| {
-                w[0] == feature
-                    && matches!(w[1], "stable" | "beta" | "experimental")
-                    && matches!(w[2], "true" | "false")
-            })
+        let mut tokens = stdout.split_whitespace();
+        loop {
+            let name = match tokens.next() {
+                Some(t) => t,
+                None => break,
+            };
+            let stability = match tokens.next() {
+                Some(t) => t,
+                None => break,
+            };
+            let enabled = match tokens.next() {
+                Some(t) => t,
+                None => break,
+            };
+            if name == feature
+                && matches!(stability, "stable" | "beta" | "experimental")
+                && matches!(enabled, "true" | "false")
+            {
+                return true;
+            }
+        }
+        false

Comments

Conversation

YusukeShimizu commented Jan 15, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Type of Change

AI Assistance

Testing

Screenshots/Demos (for UX changes)

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Copilot AI Jan 16, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Jan 16, 2026

Choose a reason for hiding this comment

Uh oh!

YusukeShimizu commented Jan 20, 2026

Uh oh!

michaelneale left a comment

Choose a reason for hiding this comment

Uh oh!

DOsinga commented Jan 28, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

YusukeShimizu commented Jan 15, 2026 •

edited

Loading