Repository navigation
Conversation
Castiron custom codeEvaluated main: ✅ No new custom-code files detected. 2 mixed files remain; 0 existing customizations changed. Compared 2 existing customizations unchanged
A changed generated baseline means this report cannot reliably identify which handwritten lines changed. Inspect the custom-code diffDownload the exact patch produced by this run (requires repository access): gh run download 37896656145 --repo openai/openai-cli \
--name castiron-custom-code-37896656145-1 --dir /tmp/castiron-custom-code-37896656145-1
git apply --stat /tmp/castiron-custom-code-37896656145-1/custom-code.patch
cat /tmp/castiron-custom-code-37896656145-1/custom-code.patchOr reproduce it from an SDK checkout containing the vendored reporter: git fetch --no-tags origin 390a9c3d68e93f7348509ffec169bb5ab38d68cc 8487ddf7f92c8a49ac06fc3e86f5e2490d79d0d1
python3 scripts/castiron/custom_code_report.py report \
--base 390a9c3d68e93f7348509ffec169bb5ab38d68cc \
--head 8487ddf7f92c8a49ac06fc3e86f5e2490d79d0d1 --fetch --require-head-hash --public \
--out /tmp/castiron-custom-code-8487ddf7f92c
cat /tmp/castiron-custom-code-8487ddf7f92c/custom-code.patchThis is the current full custom patch for mixed files, not an attribution of only the handwritten lines changed by this PR. |
|
@codex review Please review commit |
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: ce129ba366
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
🛡️ Codex Security Review · Automatically triggeredSecurity review completed. No security issues were found in this pull request. Reviewed commit: Only the user who started this review can view the report in Codex. ℹ️ About Codex security reviews in GitHubThis is an experimental Codex feature. Security reviews are triggered when:
Once complete, Codex will leave suggestions, or a comment if no findings are found. |
|
@codex review Please review the functional correction in It addresses the three findings from |
…min-setup-public-20261008
🛡️ Codex Security Review · Automatically triggeredSecurity review completed. No security issues were found in this pull request. Reviewed commit: Only the user who started this review can view the report in Codex. ℹ️ About Codex security reviews in GitHubThis is an experimental Codex feature. Security reviews are triggered when:
Once complete, Codex will leave suggestions, or a comment if no findings are found. |
|
Codex Review: Didn't find any major issues. 🚀 Reviewed commit: ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
If Codex has suggestions, it will comment; otherwise it will react with 👍. Codex can also answer questions or update the PR. Try commenting "@codex address that feedback". |
…min-setup-public-20261008 # Conflicts: # pkg/custom/help.go
|
@codex review Please review code and security for Help-aligned head This local merge preserves the reviewed Admin correction from |
🛡️ Codex Security Review · Automatically triggeredSecurity review completed. No security issues were found in this pull request. Reviewed commit: Only the user who started this review can view the report in Codex. ℹ️ About Codex security reviews in GitHubThis is an experimental Codex feature. Security reviews are triggered when:
Once complete, Codex will leave suggestions, or a comment if no findings are found. |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 3c60f6970a
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| if client, ok := command.Root().Metadata[mtlsHTTPClientMetadata].(*http.Client); ok && client != nil { | ||
| return nil |
There was a problem hiding this comment.
When an admin command targets the standard API with mTLS files configured but no OPENAI_ADMIN_KEY, configureMTLS stores a non-nil client and this branch treats that client as sufficient authentication. mTLS only authenticates API-key requests (README.md:211-223) and does not add an Authorization header, so the request proceeds without an admin key and returns a remote authentication error instead of the new setup guidance. Keep the key check active for mTLS clients, including the documented mtls.api.openai.com endpoint that usesStandardAdminService currently excludes, and cover that configuration with a focused regression test.
AGENTS.md reference: AGENTS.md:L76-L81
Useful? React with 👍 / 👎.
| } | ||
| ctx, stop := adminSetupSignalContext(ctx) | ||
| defer stop() | ||
| if _, err := fmt.Fprintf(os.Stderr, "Admin key verification\n\nNeed a key? Open:\n https://platform.openai.com/settings/organization/admin-keys\nSelect your organization.\nCreate an admin key.\nCopy the key into the hidden prompt below.\nOnly organization owners can create admin keys.\nIf you are not an owner, ask an owner to run the operation without sharing their key.\n\nDestination: %s\nThe key stays in this process. Ctrl+C cancels.\n\n", destination); err != nil { |
There was a problem hiding this comment.
When --base-url or OPENAI_BASE_URL contains a valid IDN hostname, such as one using a Cyrillic а in аpi.openai.com, url.Parse preserves the Unicode spelling while net/http converts it to punycode for the request. This prints the visually confusable spelling immediately before asking for the admin key, so a malicious inherited configuration can make an attacker-controlled endpoint look like the trusted API and receive the pasted key. Display a canonical ASCII/punycode host and terminal-escape it before prompting.
AGENTS.md reference: AGENTS.md:L51-L55
Useful? React with 👍 / 👎.
| reader.Cancel() | ||
| // Join the reader before restoring echo, clearing any unconsumed input. | ||
| for chunk := range chunks { | ||
| clear(chunk) |
There was a problem hiding this comment.
On a terminal that does not emit bracketed-paste markers, or when a pasted suffix remains beyond the current 4096-byte read, the first CR/LF makes collectAdminSetupKey return while later bytes can remain in the terminal's OS input queue. This cleanup only drains chunks already forwarded by the goroutine; it does not flush that queue before restoring canonical mode, so remaining pasted lines are handed to the parent shell after the CLI exits and can execute as commands. Flush pending terminal input before restoring and add a PTY regression using an unmarked multiline paste split across reads.
AGENTS.md reference: AGENTS.md:L51-L55
Useful? React with 👍 / 👎.
| go readAdminSetupKeyBytes(readCtx, reader, chunks) | ||
| defer func() { | ||
| cancel() | ||
| reader.Cancel() |
There was a problem hiding this comment.
On Windows Terminal, the ultraviolet cancel reader explicitly permits Cancel() to return false when its console read cannot be interrupted. Ignoring that result and then ranging until chunks closes waits forever for the still-blocked reader goroutine, so parent-context cancellation or a signal can leave setup hung in raw/no-echo mode instead of restoring the terminal. Handle the failed-cancel path without an unbounded join and cover cancellation in a native Windows console test.
AGENTS.md reference: AGENTS.md:L51-L55
Useful? React with 👍 / 👎.
…min-setup-public-20261008 # Conflicts: # pkg/custom/local_errors.go
…min-setup-public-20261008
…min-setup-public-20261008
…min-setup-public-20261008
jbeckwith-oai
left a comment
jbeckwith-oai
left a comment
There was a problem hiding this comment.
Reviewed credential precedence and validation order, cloned admin routes, endpoint/mTLS preservation, hidden input, cancellation, verification, and error presentation. No blocking findings; the previously reported issues appear addressed.
On the exact head, focused TestAdmin and public TestMainAdmin tests, go vet ./..., and changed-file gofmt checks pass. A temporary credential-acceptance mutation was caught by the tests, and the restored tests passed. Hosted checks pass. Native Windows behavior was reviewed statically, not executed; no live API calls were made.
This branch has not been deployed
2 participants
Admin commands currently fail without explaining that they need an admin key.
This adds clear guidance and
openai setup admin, which hides key entry and verifies access automatically.What changes
Commands
Interactive setup requires terminal input and diagnostic output in text mode.
Debug logging and structured output stop before prompting.
The destination follows
--base-urlorOPENAI_BASE_URL.HTTPS is required except for loopback servers.
The entered key replaces Authorization; other configured headers and mTLS remain active.
An empty project list also verifies access successfully.
Code
pkg/customowns credential checks, setup orchestration, hidden input, and safe errors.The Windows factory uses a console-record reader with cancellable waits and joined cleanup.
macOS and Linux retain the existing reader with platform input cleanup.
The parser, signal handling, and verification request remain shared.
Canonical metadata survives command cloning.
Request setup checks credentials after required-input validation and before file expansion.
No new package, dependency, generated-command change, main.go change, or budget increase is included.
Tested
Candidate 8487ddf includes main 390a9c3 and preserves Models, Images, and Shell behavior.
After main alignment, all 458 affected product-test entries passed without failures or skips.
Separate presenter checks preserved Admin, download, and save-receipt diagnostics around context errors.
The 21 native macOS key-entry cases passed on 15e8c66; their exact reader source remains unchanged.
All 14 public-command cases passed on the aligned binary in the serialized macOS run.
They cover typed/pasted input, cancellation, invalid input, redirected output, rejected modes, and terminal restoration.
Successful verification issued one loopback GET. Tests used synthetic keys and left home files unchanged.
Both Windows test packages compiled, and Windows vet passed.
Local validation used Go 1.27.0 on macOS arm64; compilation does not establish native Windows behavior.
On head 8487ddf, Windows CI passed 35 production sessions and 37 distinct input/error checks.
On head 8487ddf, Linux CI passed the full test suite, including the selected terminal test.
Its quiet package logs do not print individual case names.
Independent exact-source review verified all 950 files in the aligned snapshot.
Head 8487ddf passed CI, CodeQL, build artifacts, all 9 vulnerability targets, and the 21/1000 budget gate.
CI tested the same source tree as the reviewed candidate. Help checks passed on macOS, Linux, and Windows.
No workflow, scanner, or budget gate changed.
Windows Terminal application behavior and live API verification remain untested.
Demo
Actual binaries on macOS arm64, comparing main d32b3ae with ce129ba.
The recording uses a synthetic key and loopback API. No live key is created.
The follow-up preserves this setup output; its validation and cancellation corrections have separate regression coverage.
Before:
After:
Recording recipe · Synthetic fixture