| name | github-issue |
| description | Unified GitHub issue workflow via the gh CLI — fetch an issue (body, labels, comments, image/video attachments) into the session, create a new issue from a template, or update an existing one (body, labels, state, comments). Handles the 404 that plain curl/WebFetch hits on GitHub attachment URLs by downloading with an auth token, and extracts viewable frames from video attachments. Use whenever the user mentions GitHub issues — 이슈 가져와, 이슈 조회, 이슈 등록, 이슈 생성, 이슈 만들어줘, 이슈 업데이트, 이슈 수정, 이슈 코멘트, 이슈 작업, fetch issue N, work on issue N, create an issue, update issue N, file a bug, /github-issue. |
GitHub Issue
Operate on GitHub issues with the gh CLI. One skill, three operations — dispatch on the user's words or the argument:
| User intent | Operation |
|---|
"이슈 N 가져와 / 조회 / 작업", "fetch issue N", bare /github-issue 94 | Fetch |
| "이슈 등록 / 생성 / 만들어줘", "create issue", "file a bug" | Create |
| "이슈 N 수정 / 업데이트 / 코멘트 / 닫아줘", "update issue N" | Update |
If the intent is genuinely ambiguous, ask which operation — don't guess.
Common setup
- Confirm
gh auth status works once; if not, stop and tell the user to run gh auth login.
- Default to the current repo. If the user names another repo, pass
-R <owner>/<repo> to every gh call.
Fetch
- If no issue number was given, run
gh issue list, show the open issues, and let the user pick.
gh issue view <N> --json number,title,state,labels,author,url,body,comments- Download every attachment found in the body and comments — plain curl/WebFetch returns 404 on GitHub attachment URLs. Follow fetching.md exactly (auth-token download, image Read, video → ffmpeg frames).
- Summarize in the conversation: issue number/title/labels, the problem or request in one or two lines, and what the attachments show.
- If the user is picking the issue up to work on it: locate the relevant code (grep for the symbols/strings the issue points at), state an initial hypothesis and candidate approach in a sentence or two — then stop. Do not start editing code; intake ends at understanding.
Create
-
Pick a template, in this order:
-
Fill the template from the conversation and any context the user gave. Write the issue in the language of the repo's existing issues (check a recent one if unsure). Ask only for fields you genuinely can't fill — don't interrogate.
-
Show the complete draft (title, body, labels) and wait for the user's OK before creating. Filing an issue is outward-facing and visible to others; never skip this gate.
-
Create with a body file so markdown survives quoting:
gh issue create --title "<title>" --body-file <draft.md> --label "<labels>"
Only pass labels that already exist in the repo (gh label list) — gh errors on unknown labels. Same rule for --add-label on update.
-
Report the created issue URL.
Update
- Fetch the current state first (
gh issue view <N> --json title,body,labels,state) — never edit blind.
- Show what will change as before → after (title/body diff, labels added/removed, state change) and get confirmation before applying.
- Apply with the matching command:
- body/title/labels:
gh issue edit <N> --title/--body-file/--add-label/--remove-label
- comment:
gh issue comment <N> --body-file <comment.md>
- state:
gh issue close <N> [--comment] / gh issue reopen <N>
- Report the issue URL and what changed.
Anti-patterns
- WRONG:
curl <attachment-url> or WebFetch on github.com/user-attachments/assets/... → 404. RIGHT: download with Authorization: token $(gh auth token) per fetching.md.
- WRONG: Read a downloaded
.mp4 directly (wastes a turn, shows nothing). RIGHT: extract frames with ffmpeg and Read those.
- WRONG:
gh issue create/edit straight away with a body you never showed. RIGHT: full draft → user confirmation → apply.
- WRONG: free-form issue body because the template "doesn't quite fit". RIGHT: pick the closest template and drop sections that are truly N/A.
