gh-issue-sync: sync GitHub issues locally and back

gh-issue-sync is a command line tool that syncs GitHub issues to local Markdown files for offline editing, batch updates, and integration with coding agents.

Pull issues locally, refine them until you are satisfied, and sync changes back. Also useful for offline access to your issues.

When refining many issues at once, editing them on GitHub can be tedious. It is easier to make changes locally and push them all at once. This is particularly useful when using Claude Code or similar tools to refine issues iteratively.

Agents can work with issues locally until you are ready to push. The tool also supports creating new issues locally with temporary IDs that get replaced with real issue numbers after pushing.

Watch the demo video

gh-issue-sync mirrors GitHub issues into a local .issues/ directory as Markdown files with YAML front matter. Edit issues in your favorite editor, create new issues locally, and push changes back to GitHub when ready.

Prerequisites:

• GitHub CLI (gh) installed and authenticated (gh auth login)

curl -sSfL https://github.com/mitsuhiko/gh-issue-sync/releases/latest/download/install.sh | sh

go install github.com/mitsuhiko/gh-issue-sync/cmd/gh-issue-sync@latest

Download the latest binary from GitHub Releases and place it in your PATH.

git clone https://github.com/mitsuhiko/gh-issue-sync.git
cd gh-issue-sync
go build -o gh-issue-sync ./cmd/gh-issue-sync
mv gh-issue-sync ~/.local/bin/

# Navigate to your project
cd my-project

# Initialize issue sync (auto-detects repo from git remote)
gh-issue-sync init

# Pull all open issues from GitHub
gh-issue-sync pull

# View your local issues
ls .issues/open/

# Edit an issue
$EDITOR .issues/open/123-fix-login-bug.md
gh-issue-sync edit 123

# Push your changes
gh-issue-sync push

# Or sync both ways (push then pull)
gh-issue-sync sync

When you run gh-issue-sync init, the .issues directory is created next to the .git directory (at the repository root), regardless of your current working directory.

For other commands, gh-issue-sync searches for .issues by walking upward from the current directory until it finds one or reaches a .git root. This means you can run commands from any subdirectory within your project.

Set GH_ISSUE_SYNC_DIR to explicitly specify the .issues directory location:

# Use a custom location
export GH_ISSUE_SYNC_DIR=/path/to/my-project/.issues
gh-issue-sync list

# Or inline
GH_ISSUE_SYNC_DIR=~/.issues/work-project gh-issue-sync pull

This is useful when:

• Working with multiple repositories
• Storing issues outside the repository
• Using a shared issues directory across projects

This tool is designed to work with coding agents. Install the skill file so your agent knows how to use gh-issue-sync:

gh-issue-sync write-skill --agent codex     # Codex
gh-issue-sync write-skill --agent pi        # For Pi
gh-issue-sync write-skill --agent claude    # Claude Code
gh-issue-sync write-skill --agent opencode  # OpenCode
gh-issue-sync write-skill --agent generic   # Amp and others

Use --scope to choose between user-level (default) or project-level installation:

# Install to user home directory (default)
gh-issue-sync write-skill --agent codex --scope user

# Install to current project directory
gh-issue-sync write-skill --agent codex --scope project

To install to a custom location:

gh-issue-sync write-skill --output /path/to/skills/gh-issue-sync/

You can also read or copy the skill file directly: skill/SKILL.md

Since issue numbers come from GitHub, you can use temporary issue numbers until then. T42 or TABC are valid temporary issue IDs. They must start with "T" to mark them as temporary. After syncing, they receive real numbers and all references are updated.

Push and pull in a single command:

# Push local changes, then pull remote updates
gh-issue-sync sync

# Include closed issues
gh-issue-sync sync --all

# Filter by label
gh-issue-sync sync --label bug

The tool uses three-way comparison (local, original, remote) to detect conflicts. Original versions are stored in .issues/.sync/originals/.

On pull: New issues are saved, unchanged local files are updated, conflicts are skipped (use --force to overwrite). Deleted local files are restored.

On push: Local issues (T1, T2, etc.) are created and renamed with real numbers. References like #T1 are updated automatically. Missing labels and milestones are created. Conflicts with remote changes are skipped.

List and filter local issues:

# List open issues
gh-issue-sync list

# Include closed issues
gh-issue-sync list --all

# Filter by label, assignee, author, milestone
gh-issue-sync list --label bug --assignee alice

# GitHub-style search query
gh-issue-sync list --search "error no:assignee sort:created-asc"

The --search flag supports GitHub issue search syntax:

• is:open, is:closed - Filter by state
• label:NAME - Filter by label
• no:label, no:assignee, no:milestone - Filter by missing field
• assignee:USER, author:USER, milestone:NAME - Filter by field
• sort:created-asc, sort:created-desc - Sort results
• Free text - Search in title and body (case-insensitive)

See what's changed locally:

gh-issue-sync status

Create issues locally before pushing to GitHub:

# Create with a title
gh-issue-sync new "My new feature idea"

# Create and open in editor
gh-issue-sync new "Fix login bug" --edit

# Create with labels
gh-issue-sync new "Critical bug" --label bug --label urgent

# Create with just the editor (no title required)
gh-issue-sync new --edit

Local issues get temporary IDs like T1, T2. When pushed, they become real GitHub issues and files are renamed automatically.

# Close an issue (marks for closing on next push)
gh-issue-sync close 123

# Close with a reason
gh-issue-sync close 123 --reason not_planned

# Reopen a closed issue
gh-issue-sync reopen 456

Alternatively, move files manually:

• Move from open/ to closed/ to close
• Move from closed/ to open/ to reopen

See Issue Format for details on file structure, front matter fields, and pending comments.

• Skill File
• Issue Tracker
• License: Apache-2.0

This code is entirely LLM generated. It is unclear if LLM generated code can be copyrighted.