git-issue-create skill
✓ VerifiedCreates standalone GitHub issues with automatic duplicate checking, conventional titles, template awareness, and label resolution.
- GitHub stars
- 6
- Files
- 2
- Last commit
- Nov 6, 2026
- License
- MIT
- Allowed tools
- —
What does the git-issue-create skill do?
git-issue-create creates standalone GitHub issues with automatic duplicate checking, conventional titles, template awareness, and label resolution.
Summary generated from the skill's SKILL.md.
About the git-issue-create skill
Developers who need to document defects or requested features in GitHub can use this skill to file structured, standalone issues. It automates the tedious parts of issue management by performing a duplicate check using the GitHub CLI before any content is drafted. If a strong match is found, the process stops to prevent backlog clutter, whereas adjacent issues are linked in a context section.
The tool relies on the gh command to interact with repositories and fetch existing labels or templates. It automatically detects repository issue templates to maintain consistent formatting and uses a specific grammar for titles to ensure they follow conventional commit styles. It also resolves labels through a candidate chain to match the existing taxonomy of a project without ever creating new labels. This ensures that every issue has exactly one type label and follows the project's established organizational structure.
What an agent can do with it
- 1File a bug reportAn agent can take a user's description of a software defect and create a formatted GitHub issue with a conventional title and appropriate labels.
- 2Prevent duplicate issue creationThe skill searches for existing issues using distinctive keywords to ensure the same problem is not reported multiple times.
- 3Maintain repository template consistencyWhen a repository has predefined issue templates, the skill populates them while preserving the required structure and headings.
- 4Automate label assignmentThe skill maps intended issue types to the specific label names used in a given repository through a predefined candidate chain.
Written from the skill's SKILL.md. Every use case is checked against a quote from it.
How to install the git-issue-create skill
Each command copies the skill folder into the client's skills directory. Restart the client or start a new session to load it.Install the git-issue-create skill in Claude Code
~/.claude/skills/git-issue-create/SKILL.md
$ npx degit pivoshenko/pivoshenko.ai/skills/git-issue-create ~/.claude/skills/git-issue-createInstall the git-issue-create skill in Claude Code (project)
.claude/skills/git-issue-create/SKILL.md
$ npx degit pivoshenko/pivoshenko.ai/skills/git-issue-create .claude/skills/git-issue-createInstall the git-issue-create skill in Codex
~/.agents/skills/git-issue-create/SKILL.md
$ npx degit pivoshenko/pivoshenko.ai/skills/git-issue-create ~/.agents/skills/git-issue-createInstall the git-issue-create skill in Gemini CLI
~/.gemini/skills/git-issue-create/SKILL.md
$ npx degit pivoshenko/pivoshenko.ai/skills/git-issue-create ~/.gemini/skills/git-issue-createInstall the git-issue-create skill in Claude.ai
npx degit pivoshenko/pivoshenko.ai/skills/git-issue-create git-issue-create && zip -r git-issue-create.zip git-issue-create # claude.ai → Customize → Skills → + → Create skill → Upload a skill → git-issue-create.zip
Automated checks
Ran Nov 15, 2026Full report →Passed our automated checks. That is not a security audit: read the scripts before you let an agent run them. How checks work
- Source repository linked and reachablepivoshenko/pivoshenko.ai/skills/git-issue-create is public and reachablegithub.com/pivoshenko/pivoshenko.ai/tree/HEAD/skills/git-iss ↗Pass
- Open source license (OSI)MITspdx.org/licenses/MIT.html ↗Pass
- Commit or release in the last yearLast commit 2026-11-06github.com/pivoshenko/pivoshenko.ai/commits ↗Pass
- No download-and-run or credential theft in scriptsNo scripts in the skill folderPass
- No hidden instructions in SKILL.mdNo override phrases, safeguard switches, credential requests or hidden charactersgithub.com/pivoshenko/pivoshenko.ai/blob/HEAD/skills/git-iss ↗Pass
- Pre-approved tools are scopedNo pre-approved tools; the agent asks before using onegithub.com/pivoshenko/pivoshenko.ai/blob/HEAD/skills/git-iss ↗Pass
- SKILL.md follows the Agent Skills formatValid name and description; says when to use it; 188 linesgithub.com/pivoshenko/pivoshenko.ai/blob/HEAD/skills/git-iss ↗Pass
SKILL.md
View on GitHubGit Issue Create
File one standalone issue. No confirm.
Flow
- Boundary check: is the work starting now? Yes -> hand off to
git-branch-create(new work) orgit-issue-start(issue already exists) and stop. Why -> an issue is a record of work deferred; work in flight belongs on a branch, and a second tracker entry for it just rots - Repo:
gh repo view --json nameWithOwner -q .nameWithOwner. Fails -> stop, surface the error (no remote, no auth, not a repo). Never guessowner/name - Search for duplicates before drafting anything:
gh issue list --search "<2-4 distinctive keywords>" --state all --limit 10 \ --json number,title,state,url -q '.[] | "\(.number)\t\(.state)\t\(.title)"'--state all-> a closed issue is the answer often enough that skipping it files the same bug twice. Keywords are nouns from the user's description, not the whole sentence. 10 hits returned -> the search was too broad and real duplicates may sit past the cap; narrow the keywords and search again before proceeding - Triage the hits:
- Strong match (same defect, same capability) -> print its number, title, state, URL. Stop. Offer to comment on it instead (<= 4 prose lines, see Length); reopening is the user's call, not this skill's. Why -> a duplicate backlog is worse than a missing issue: every triage pass afterwards pays for it
- Adjacent match (related, not the same) -> proceed, link it in
## Context - No hits -> proceed
- Parallel:
- Repo issue template:
.github/ISSUE_TEMPLATE/*.md,.github/ISSUE_TEMPLATE/*.yml,.github/issue_template.md, rootissue_template.md(first match wins) gh label list --limit 200 --json name -q '.[].name'
- Repo issue template:
- Draft title per Title. Draft body:
- Template found -> fill that template's structure (preserve headings, order, checklists, HTML comments). A
*.ymlform -> render itslabel/attributesas markdown headings in the same order - None found (the normal case) -> use Fallback Body below
- Template found -> fill that template's structure (preserve headings, order, checklists, HTML comments). A
- Resolve labels per Labels. Exactly one
type: * - Heredoc body so markdown survives the shell:
gh issue create --title "fix(auth): session drops on api deploy" --label "type: bug" --body "$(cat <<'EOF' ## Problem Sessions drop on every API deploy because the cookie is signed with a per-instance key. ## Done when - a session survives a rolling restart of all API pods - no change required to the `users` table EOF )"'EOF'quoted -> no shell interpolation, so$, backticks and#in the body survive intact - Print the issue URL. Name any label dropped in step 7 and why
- User wants to start on it now -> hand off to
git-issue-start
Title
Same grammar as git-commit and git-pr-create:
<type>(<scope>): <short summary>
- Type:
build|chore|ci|docs|feat|fix|perf|refactor|test. Type picks and tiebreakers live ingit-commit- use that table, do not re-derive one - Scope: optional, lowercase, one only
- Summary: imperative present, lowercase, no trailing
. - Whole title <= 72 chars
The title is the type signal: fix( is what makes type: bug the obvious label in step 7.
Examples:
fix(api): handle timeout on retryfeat(catalog): filter skills by tagdocs(readme): document the sync contract
Labels
Never hardcode the namespaced taxonomy - detect it. These rules run on forks and third-party repos too.
Fetch once per run:
gh label list --limit 200 --json name -q '.[].name'
Resolve each intended label through its candidate chain, first existing name wins:
| Intent | Candidate chain |
|---|---|
feat |
type: enhancement -> enhancement -> feature |
fix |
type: bug -> bug |
docs |
type: documentation -> documentation -> docs |
perf |
performance -> type: enhancement -> enhancement |
refactor / chore / build / ci |
type: maintenance -> maintenance -> chore |
test |
type: maintenance -> tests -> test |
| breaking change | type: breaking -> breaking-change -> breaking |
| dependency bump | type: dependencies -> dependencies -> deps |
| security | type: security -> security |
| priority | priority: <level> -> <level> priority -> <level> |
| blocked | status: blocked -> blocked |
| duplicate | status: duplicate -> duplicate |
- Whole chain missing -> drop that label and say so in the output. Never
gh label create - Exactly one
type: *per issue. Pick the dominant type, never two priority: *only when the user signalled urgency. Never guess a priority- Multi-word names must be quoted:
--label "type: bug"
Body - Repo Template Present
- Template verbatim as the skeleton: headings, order, checklist items, HTML comments
- Fill only the fields the user actually gave you. A required field you cannot answer -> ask, do not invent
- Tick checklist items that apply; leave the rest unchecked
*.ymlform with alabels:key -> those labels are additive to step 7, and still go through the candidate chains
Fallback Body
No template found - the normal case. Fill references/issue-body.md.
Optional, only when it carries a fact:
## Context- <= 2 lines, links to prior issues, PRs, ADRs## Out of scope- <= 2 lines, the adjacent thing a reader would otherwise assume is included
Never emit an empty placeholder section. Drop it.
Acceptance criteria you were not told -> ask, or omit ## Done when. Why -> an invented criterion is worse than a missing one: it closes the issue on the wrong condition.
Length
Hard cap: 8 lines of prose for an issue body, 4 for a comment on an existing one. Headings and code blocks don't count - only text you write.
## Problem: 1-3 lines. What is wrong and what it costs. Not the fix## Done when: one bullet per condition, one line each, 100 chars or fewer. No sub-bullets, no paragraph bullets- Any other section: <= 2 lines, or drop it
- Nothing to say -> drop the section. An empty section is information, filler is not
Prohibitions:
- Why it exists and what "done" means, not how - the code shows how
- No restating the title. The reader just read it
- No play-by-play ("first X, then Y"). No file inventories, no code dumps
- Never invent acceptance criteria, paths, or repro steps. Unknown -> ask or omit
Good:
## Problem
`kst sync` silently skips new MCP files because `kasetto.yaml` enumerates them by name.
## Done when
- adding `mcps/<name>.json` syncs without editing `kasetto.yaml`
- the site and the sync agree on which MCPs exist
Bad - filler opener, restates the title, narrates the work, invents criteria:
## Problem
This issue tracks the comprehensive implementation of a more robust MCP sync
mechanism. Currently the existing configuration leverages an enumeration-based
approach which presents certain challenges.
## Done when
- First we will investigate how kasetto.yaml is parsed
- Then we will design a holistic solution
- Finally we will ensure that all 6 MCP servers sync seamlessly
Rules
- Duplicate search runs before the body is drafted. Why -> drafting first makes you argue for filing it
- Strong duplicate -> stop and surface it. Never file a near-duplicate "to be safe"
- Never
gh label create. Chain missing -> drop the label, report it - Exactly one
type: *label. Two type labels break every triage query - Never invent acceptance criteria, file paths, numbers, or repro steps. Unknown -> ask or omit
- Never
--assignee/--milestone/--projectunless asked. Self-assigning implies you are starting, which isgit-issue-start - Body over 8 prose lines -> cut before creating, not after
- One issue per invocation. A list of separate problems -> one issue each, filed one at a time
- Never close or edit an existing issue from this skill
- No "Generated with Claude Code" or co-author trailers
Files in this skill
2git-issue-create skill questions
How do I install the git-issue-create skill in Claude Code?
Run npx degit pivoshenko/pivoshenko.ai/skills/git-issue-create ~/.claude/skills/git-issue-create. Claude Code loads skills from ~/.claude/skills (all projects) or .claude/skills (one project) and uses a skill when its description matches the task.
Which clients can use the git-issue-create skill?
Any client that reads the Agent Skills format: Claude Code, Codex, Gemini CLI and claude.ai (upload as a ZIP). Scripts in the skill run only where the client allows code execution.
What happens if a duplicate issue is found?
The skill stops and prints the number, title, state, and URL of the existing issue, then offers to comment on it instead.
Does this skill create new labels?
No, it only resolves intended labels through existing candidate chains and drops them if the chain is missing.
How does the skill handle repository templates?
It searches for various GitHub issue template files and uses them as a skeleton, filling in only the fields provided by the user.
Can I use this to start working on an issue immediately?
No, if the work is starting now, you should use git-branch-create or git-issue-start instead.