Codex Bug Triage Assistant
Determines whether an openai/codex GitHub bug should be verified, clarified, or explained as expected behavior.
The skill restricts input to an openai/codex GitHub issue and makes the primary data flow visible by preferring read-only GitHub API calls; however, it mandates immediate network access without explaining sensitive log/attachment handling, least-privilege permissions, user confirmation, retention, or rollback. The repository security policy and Apache-2.0 license provide context but do not complete the skill's own safety boundaries.
Input validation, API endpoints, extraction fields, and the three decision paths are reasonably coherent, and failure is supposed to produce an offline explanation. Deductions apply because timeout/retry behavior, API errors, attachments, unavailable thread logs, and missing local source are underspecified; static review provides no executable reproduction evidence, so the score remains below the static ceiling of 10.
The target scenario is clear: diagnose an openai/codex GitHub issue and choose the next action. However, it defines no boundaries for non-bugs, duplicates, permission-restricted issues, or other failure cases, and gives no Chinese-language output guidance or mainland-China network fallback despite mandatory overseas network access.
Front matter, overview, and workflow are readable, and the name broadly matches the capability. Repository context supplies Apache-2.0 licensing, verified OpenAI provenance, security policy, and installation documentation, but the skill itself lacks versioning, changelog, maintenance ownership, install/dependency notes, example outputs, and FAQs.
With issue, comment, and repository access, the workflow covers summarization, evidence extraction, verification, information requests, not-a-bug explanations, and a concise report, so it plausibly completes the core task. The output format lacks a fixed template or verified representative result, and network dependence limits direct usability; under static calibration the score cannot exceed 7.
The skill names concrete API endpoints and extraction fields, providing limited auditability. It contains no tests, fixed fixtures, third-party execution evidence, or independent corroboration, and this review did not execute network or code checks, so evidence coverage is thin.
- The workflow mandates immediate GitHub API access, but fallback behavior is underspecified for network, permission, or mainland-China reachability failures.
- Issue bodies, comments, logs, and attachments may contain credentials or personal data; no redaction, data minimization, or output-boundary guidance is provided.
- There is no fixed report template, version/changelog policy, or skill-specific test evidence, making compatibility after updates difficult to assess.
What it does & when to use it
This skill diagnoses bug reports filed in the openai/codex repository. It requires a GitHub issue URL for that repository and reads the issue and its comments through the GitHub API. It extracts reproduction steps, expected and actual behavior, environment details, logs, attachments, and thread IDs before choosing a next action. Its scope is limited to openai/codex issues, rather than general code review or debugging across arbitrary repositories.
Validates that an issue URL belongs to github.com/openai/codex/issues/...; accesses the corresponding GitHub API endpoints for the issue and comments; extracts the title, body, reproduction steps, expected behavior, actual behavior, environment, logs, attachments, and thread IDs; summarizes the report before deeper investigation; chooses whether to verify the report, request more information, or explain why it is not a bug; and, when appropriate, inspects relevant Codex files, documentation, or logs before producing a concise finding and next-step recommendation.
- A Codex maintainer receives a detailed issue with reproducible steps and needs to compare it with the repository's current behavior.
- A contributor submits a report without sufficient logs or environment details and needs a precise request for follow-up information.
- A support engineer needs to determine whether a report conflicts with documented Codex constraints or current repository behavior.
- An engineering team needs to extract thread IDs, logs, or session details from an issue's comment history before investigating.
Pros & cons
- Has a clearly defined scope centered on openai/codex issues.
- Uses the GitHub API to read both issue details and comments.
- Separates verification, information gathering, and expected-behavior explanations.
- Explicitly checks for evidence such as logs, environment details, attachments, and thread IDs.
- Requires a valid openai/codex issue URL.
- Depends on network access to GitHub.
- Provides no documented automated test suite or dedicated verification scripts.
- Does not cover issues from other GitHub repositories.
How to install
The README does not document a separate installation procedure for this skill. It is located at .codex/skills/codex-bug/SKILL.md and is distributed as part of the repository's skill collection. The README documents Codex CLI installation; for example, macOS or Linux can use curl -fsSL https://chatgpt.com/codex/install.sh | sh, while Windows can use powershell -ExecutionPolicy ByPass -c "irm https://chatgpt.com/codex/install.ps1 | iex".
How to use
Provide a GitHub issue URL under github.com/openai/codex/issues/... and ask for triage. Example: Diagnose this openai/codex bug and decide whether to verify it, request more information, or explain why it is not a bug: https://github.com/openai/codex/issues/123. If the URL is missing or points elsewhere, the skill asks for the correct link.