gh（GH 搜索）

GitHub CLI (gh)

Overview

GitHub CLI (gh) is the official command-line tool for GitHub that brings pull requests, issues, releases, and other GitHub concepts to the terminal. This skill provides comprehensive guidance for using gh to streamline GitHub workflows, including PR management, issue tracking, repository operations, and API access.

When to Use This Skill

Invoke this skill when users:

• Mention GitHub CLI (gh) commands or workflows
• Ask about creating, viewing, or managing pull requests from the command line
• Need help with GitHub issue management via CLI
• Want to understand gh's mental model or command structure
• Request guidance on GitHub automation or scripting
• Ask about GitHub API access through gh
• Need help with gh authentication or configuration
• Request integration patterns between gh, git, and other tools (like erk)

Core Concepts

Before providing guidance, understand these key concepts:

Three-Layer Architecture:

1. High-Level Commands (porcelain): gh pr, gh issue, gh repo - user-friendly workflows
2. API Access: gh api - direct REST/GraphQL access with built-in auth
3. Git Integration: Smart repo detection from git remotes and current branch

Context Resolution:

• gh automatically detects repository from git remotes
• Uses current branch to infer PR context
• Falls back to interactive selection when ambiguous

Mental Model: Think of gh as "GitHub workflows made CLI-native" - each command maps to a common GitHub workflow optimized for terminal use.

Using the Reference Documentation

When providing gh guidance, load the comprehensive reference documentation:

references/gh.md

This reference contains:

• Complete mental model and terminology
• Full command reference for PRs, issues, repos, releases, and more
• Authentication and configuration patterns
• Workflow patterns for common scenarios
• Integration details (erk, git, CI/CD)
• GitHub API access patterns
• Practical examples for daily development

Loading Strategy:

• Always load references/gh.md when user asks gh-related questions
• The reference is comprehensive (~1480 lines) but optimized for progressive reading
• Use grep patterns to find specific sections when needed:

- gh pr - Pull request commands - gh issue - Issue management commands - gh repo - Repository operations - gh api - API access patterns - Pattern [0-9]: - Workflow patterns - Authentication - Auth setup - erk Integration - Integration with erk

Common Operations

When users ask for help with gh, guide them using these patterns:

First-Time Setup

1. Check if gh is installed: gh --version
2. Authenticate: gh auth login (interactive) or gh auth status (check current)
3. Configure defaults: gh config set <key> <value>
4. Test connection: gh repo view (in a git repo)

Pull Request Operations

Load references/gh.md and search for "gh pr" section to provide:

• Creating PRs: gh pr create (interactive or with flags)
• Viewing PRs: gh pr view, gh pr list
• Checking out PRs: gh pr checkout <number>
• Reviewing PRs: gh pr review, gh pr diff
• Merging PRs: gh pr merge
• PR status checks: gh pr checks

Issue Management

Load references/gh.md and search for "gh issue" section to provide:

• Creating issues: gh issue create
• Listing issues: gh issue list (with filters)
• Viewing issues: gh issue view <number>
• Updating issues: gh issue edit, gh issue close
• Issue assignment and labels

Repository Operations

Load references/gh.md and search for "gh repo" section to provide:

• Viewing repos: gh repo view
• Cloning repos: gh repo clone
• Creating repos: gh repo create
• Forking repos: gh repo fork
• Repository settings and visibility

API Access

Load references/gh.md and search for "gh api" section to provide:

• REST API calls: gh api <endpoint>
• GraphQL queries: gh api graphql -f query='...'
• Pagination handling
• JSON processing with --jq
• Authentication details

For advanced GraphQL use cases, load references/graphql.md:

• Projects V2 management (no REST API exists)
• Discussion API operations
• Batch queries across multiple repos
• Complex nested data retrieval
• Advanced issue search
• See GraphQL API Reference section below

Release Management

Load references/gh.md and search for "gh release" section to provide:

• Creating releases: gh release create
• Listing releases: gh release list
• Downloading assets: gh release download
• Uploading assets: gh release upload

Workflow Guidance

When users describe their workflow needs, map them to patterns in the reference:

Pattern 1: Daily PR Workflow - Create, review, merge PRs Pattern 2: Issue Triage - List, filter, and manage issues Pattern 3: Release Automation - Script release creation and asset management Pattern 4: PR Checks and Status - Monitor CI/CD and review status Pattern 5: Multi-Repo Operations - Work across multiple repositories Pattern 6: GitHub API Scripting - Automate complex GitHub workflows Pattern 7: Code Review Workflow - Review PRs from command line

Load the appropriate pattern sections from references/gh.md based on user needs.

Output Formatting

When users need to parse or format gh output:

1. Load the Output Formatting section from references/gh.md
2. Explain format options: --json, --jq, --template
3. Show JSON field selection: gh pr list --json number,title,state
4. Demonstrate jq filtering: gh api /repos/{owner}/{repo}/pulls | jq '.[] | select(.draft==false)'
5. Provide template examples for custom output

Authentication and Configuration

When users need auth or config help:

1. Load the Authentication & Configuration section from references/gh.md
2. Distinguish between:

- Authentication: gh auth login, token management - Configuration: gh config set, per-command defaults

1. Explain authentication methods: browser, token, SSH
2. Show common config settings: git_protocol, editor, pager
3. Handle multiple accounts and enterprise instances

Integration Guidance

Erk Integration

When users mention erk or worktree workflows:

• Load the Erk Integration section from references/gh.md
• Explain how gh detects repos in worktrees
• Show PR operations across multiple worktrees
• Guide on using --repo flag when needed
• Reference erk documentation for worktree-specific patterns

Git Integration

When users need to understand gh + git workflows:

• Load the Git Integration section from references/gh.md
• Explain repo detection from git remotes
• Show how current branch affects PR commands
• Demonstrate combined git + gh workflows
• Clarify when to use git vs gh commands

CI/CD Integration

When users want to integrate gh with CI/CD:

• Load the CI/CD patterns from references/gh.md
• Show authentication in CI environments: GH_TOKEN
• Provide examples of automated PR creation/merging
• Demonstrate status check monitoring
• Guide on release automation

Scripting and Automation

When users want to script GitHub workflows:

1. Load the Scripting section from references/gh.md
2. Show how to use gh in shell scripts
3. Demonstrate error handling: --jq, exit codes
4. Explain pagination for large datasets
5. Provide examples of common automation patterns:

- Bulk operations across PRs/issues - Custom notifications - Release workflows - Repository management

Advanced Features

When users need advanced capabilities:

1. Aliases: Custom commands via gh alias set
2. Extensions: Third-party gh extensions
3. GraphQL: Complex queries beyond REST API (see GraphQL API Reference)
4. Webhooks: Trigger workflows (via API)
5. GitHub Actions: Interact with workflows via gh workflow

Load the relevant sections from references/gh.md for each advanced feature.

GraphQL API Reference

When standard gh commands are insufficient, use GraphQL via gh api graphql. Load references/graphql.md for comprehensive GraphQL guidance.

When to Use GraphQL:

Use GraphQL when the porcelain commands (gh pr, gh issue, etc.) cannot accomplish the task:

• Projects V2: No REST API exists - all operations require GraphQL

- Create/update projects - Add items and update field values - Query custom fields and project data

• Discussions: No porcelain commands available

- Create/manage discussions - Add comments and replies - Query discussion categories

• Batch Operations: Query multiple resources in one API call

- Compare multiple repositories - Aggregate data across repos - Reduce rate limit consumption

• Complex Queries: Fetch nested data efficiently

- PR with reviews, comments, and status checks in one call - Issue with timeline, reactions, and linked PRs - Repository with open issues, PRs, and contributors

• Advanced Search: Complex filtering beyond basic commands

- Multi-criteria issue/PR searches - Boolean logic and date ranges - Advanced sorting options

• Custom Fields: Precise field selection

- Request only needed fields - Optimize for performance - Build custom data aggregations

GraphQL Resources:

• references/graphql.md - Complete GraphQL guide including:

- Use cases requiring GraphQL - Common patterns (variables, pagination, batching, etc.) - Complete examples (Projects V2, Discussions, batch queries) - Best practices and troubleshooting

• references/graphql-schema-core.md - Core schema types (load only when needed):

- Detailed field information for Repository, Issue, PullRequest, etc. - Projects V2 field types and mutations - Discussion types and operations - Input types and enums

Loading Strategy:

1. Start with references/graphql.md for use cases and patterns
2. Load references/graphql-schema-core.md only when detailed schema info is needed
3. Use grep patterns to find specific sections:

- Projects V2 - Project automation examples - Discussion - Discussion API examples - Batch - Batch query patterns - Pagination - Cursor-based pagination - Example [0-9] - Complete working examples

Rate Limits and API Backend

When users encounter rate limit issues or need to optimize API usage:

1. Load references/api-backend-audit.md for complete command-to-API mapping
2. Understand the difference:

- REST: 5,000 requests/hour, counted per request - GraphQL: 5,000 points/hour, cost based on query complexity - Search: 30 requests/minute (separate limit)

1. Key insights:

- PR/Issue commands use GraphQL (query cost varies) - Run/Workflow/Gist commands use REST (predictable cost) - Project commands require GraphQL exclusively - Search always uses REST with stricter limits

1. Check current limits: gh api rate_limit --jq '.resources'

Troubleshooting

When users encounter issues:

1. Check authentication: gh auth status
2. Verify repository detection: gh repo view
3. Test API access: gh api user
4. Review configuration: gh config list
5. Check rate limits: gh api rate_limit
6. Enable debug mode: GH_DEBUG=api gh <command>

Load the Troubleshooting section from references/gh.md for specific error patterns.

Command Discovery

When users ask "what can gh do?":

• Start with overview of main commands: pr, issue, repo, release, workflow
• Show command help: gh <command> --help
• Demonstrate interactive mode: most commands work interactively
• Point to gh api for anything not covered by porcelain commands
• Reference the full command reference in references/gh.md

Resources

references/

• gh.md - Comprehensive GitHub CLI mental model and command reference (~1480 lines)

- Load for all standard gh command guidance - Full workflow patterns and examples - Integration patterns (erk, git, CI/CD)

• graphql.md - GitHub GraphQL API comprehensive guide (~1000 lines)

- Load when porcelain commands are insufficient - Use cases requiring GraphQL (Projects V2, Discussions, batch queries) - Complete patterns and examples - Best practices and troubleshooting

• graphql-schema-core.md - Core GraphQL schema types (~500 lines)

- Load only when detailed schema info needed - Detailed field definitions for core types - Mutation input types and examples

• api-backend-audit.md - REST vs GraphQL API backend audit (~850 lines)

- Load when understanding which API type a command uses - Rate limit guidance and optimization strategies - Complete mapping of all gh commands to their API backends - Batch operation alternatives and caching strategies

These references should be loaded as needed to ensure accurate, detailed information. Use progressive disclosure: start with the main reference, then load specialized GraphQL docs when needed.