Codex Issue Digest
Turns recent openai/codex bugs and enhancement requests into focused, owner-ready digests.
The script primarily reads GitHub issues, comments, and reactions through gh, with a visible data flow; no credential theft, malware, or destructive default is evident. Points are deducted because external network access and local gh authentication are required, while least-privilege guidance, confirmation, sensitive-data handling, isolation, and rollback are not specified.
The skill, collector, and focused tests are broadly consistent around input parsing, label filtering, time windows, pagination, reaction counting, and error reporting. Points are deducted because this static review did not execute the code and provides no end-to-end evidence; search is seeded by updated_at and may miss reaction-only activity, while comment pagination may truncate history.
Audience, label inputs, all-label scanning, time windows, default summary behavior, and details mode are clearly described, with agent configuration and examples. Points are deducted because the output is primarily English Markdown, non-GitHub or unauthenticated environments are not clearly marked as non-fit, and reachability of the GitHub-dependent core from mainland-China networks is unverified.
The documentation is organized into objective, inputs, workflow, reaction handling, attention markers, freshness, examples, and validation; script version 5 and the repository Apache-2.0 license are present. Points are deducted for no skill-specific changelog, explicit maintenance owner, dependency installation guidance, or fuller troubleshooting and update path.
The skill can turn recent bug/enhancement activity into summary inputs and table-ready rows containing interaction, label, comment, and reaction signals, providing clear value over manual searching. Points are deducted because final insights still require model synthesis, static materials do not establish completeness or direct usability on real GitHub data, and reaction-only activity can be missed.
Committed tests cover query construction, parsing, filtering, windows, pagination, interaction deduplication, reactions, and output structure, and the implementation is auditable. Points are deducted because tests were not executed and there is no third-party or end-to-end evidence covering real GitHub APIs, gh authentication, network failures, and the complete collection path.
- Execution requires the gh CLI, valid authentication, and GitHub network access; confirm permission scope and connectivity first.
- Comment pagination is capped by default, and updated_at-based search may miss reaction-only activity; do not treat the digest as a complete audit.
- Output is primarily English Markdown and may require translation or reformatting for Chinese teams.
What it does & when to use it
This skill collects recent GitHub issues from openai/codex and filters them by feature-area labels. It defaults to the previous 24 hours, while supporting custom windows and an all-labels scan. The default output is a headline-first summary, with compact issue details available when requested. It runs from a current Codex repository checkout and requires the accompanying Python collector plus GitHub network access.
Runs collect_issue_digest.py to read issues, recent comments, recent reactions/upvotes, current labels, and current reaction totals; selects issues carrying bug or enhancement plus requested owner labels, or scans all feature labels in all-labels mode; calculates interaction counts and attention markers for the requested window; and produces summary_inputs and digest_rows for summary and optional detail output.
- A Codex maintainer needs the latest tui, exec, or app bugs from the past 24 hours.
- An engineering lead wants to review high-interaction issues across all feature areas over the past week.
- A triage team needs to identify recurring customer concerns from recent comments, reactions, and upvotes.
- A project team wants a summary-first daily digest that can expand into an issue table when needed.
Pros & cons
- Supports owner-label filtering, all-label scans, and configurable time windows.
- Combines new issues, comments, reactions/upvotes, and unique-user interaction counts.
- Uses summary-first reporting with maintainer-oriented attention markers and optional details.
- Reports the script version, git head, and time window for traceability.
- Depends on a current openai/codex checkout and its scripts; no standalone installation flow is documented.
- Requires GitHub network access; a model-only API environment needs an external shell, filesystem, and network runner.
- Issue search is seeded by updated_at, so reaction-only activity may be missed when it does not update an issue.
- The source provides no evidence of direct support for other repositories, platforms, or non-GitHub issue systems.
How to install
The source does not document a standalone installation command for this skill. It is located at .codex/skills/codex-issue-digest/SKILL.md in the openai/codex repository and should be used from a current checkout containing the referenced script. The repository README documents Codex CLI installation through npm, Homebrew, or installation scripts.
How to use
From a current openai/codex checkout, run: python3 .codex/skills/codex-issue-digest/scripts/collect_issue_digest.py --labels tui exec --window-hours 24. Use --window "past week" or --window-hours 168 for a longer window, and --all-labels to scan every current feature label. Example trigger: Use $codex-issue-digest to run the Codex issue digest for labels tui and exec over the previous 24 hours.