gh · GitHub

GitHub CLI for AI Agents

Review pull requests, triage issues, inspect Actions, and manage releases with GitHub's official CLI.

Official toolOperational risk: R0–R3verified

Install for an Agent Report broken data

Agent readiness

84/100

Evidence confidence

verified

Documentation checked

2026-07-10

Independently tested version

2.86.0

Install for an Agent

Choose an official installation path that matches the runtime. Pin a version for team or CI use, then record the version before the first task.

HomebrewRecommended

macos · linux

$ shell

brew install gh

winget

windows

$ shell

winget install --id GitHub.cli

Authentication and Minimum Permissions

Grant only the permissions the task needs. Pass credentials through environment variables or a platform secret store, never through prompts, repositories, or logs.

Authentication requiredHeadless authentication supported

Use a read-only or fine-grained token limited to the required repositories.

Methods: OAuth, personal access token, GitHub Actions token
Secret environment variables: GH_TOKEN, GITHUB_TOKEN
Credential storage: For headless runs, inject GH_TOKEN, GITHUB_TOKEN from the CI or platform secret manager at process start. For local interactive use, prefer the CLI or operating-system credential store when the official client supports one. Never save values in repository files.

Agent and Environment Compatibility

Confirm shell access first, then check the platform, network boundary, and credential path.

claude-codecodexgemini-clicopilot-cli

Environments: local, ci, container, headless, remote
Platforms: macos, linux, windows

Structured Output for Reliable Automation

Prefer a machine-readable format. Treat stdout as the result channel and stderr as diagnostics so the agent can parse failures separately.

json

Use --json or --jq where supported and keep diagnostic logs on stderr.

--json--jq

No independently captured output sample

Structured-output support currently comes from official documentation. CLI Finder does not show a guessed example or invented schema before a bounded, non-destructive execution captures stdout.

R0–R3 Command Risk Guide

Risk is assigned per command. R0 is local or remote read-only, R1 is reversible local write, R2 changes remote state, and R3 can be irreversible or production-impacting.

Read-only does not mean public

R0 only means the command does not change local or remote state. A read-only command may still return secrets, identity data, configuration, or production data. Expose only the minimum needed for the task, and never place it in logs, prompts, or commits.

R0List pull requests

Read pull request status and review fields without changing repository state.

$ shell

gh pr list --state open --json number,title,reviewDecision,url

IdempotentSensitive output

R2Merge a pull request

Merges code into the target branch and needs explicit approval.

$ shell

gh pr merge 123 --squash

Confirmation requiredMay repeat a change

R3Delete a repository

Deletes remote repository data and is destructive.

$ shell

gh repo delete owner/repo

Confirmation requiredMay repeat a change

How the Agent Readiness Score Is Built

Readiness describes how reliably an agent can operate the tool. It does not make every command safe and it does not replace an independent execution test.

Documentation indicates an agent-readiness score of 84/100. A bounded local smoke test is recorded for 2.86.0; review its limitations before relying on untested commands.

Structured output

Use --json or --jq where supported and keep diagnostic logs on stderr.

18/20

Headless operation

Official documentation describes a non-interactive authentication or execution path.

14/15

Safety controls

CLI Finder separates read commands from commands that require confirmation.

11/15

Determinism

Commands use explicit arguments and documented output controls where available.

8/10

Authentication

Use a read-only or fine-grained token limited to the required repositories.

8/10

Documentation

This entry cites official documentation checked on 2026-07-10.

9/10

Installation

Official installation paths cover macOS, Linux, and Windows.

8/8

Maintenance

An official source repository is linked for release and maintenance review.

6/7

Agent artifacts

CLI Finder can generate registry-derived skills and policies; the tool itself was not credited with shipping them.

2/5

Generate a Skill or Agent Policy

Choose an agent and safety mode to generate a copyable artifact with installation, allowed commands, approval boundaries, and the evidence limitation.

Artifact typeTarget agentSafety mode

Generated artifact preview

SKILL.md

---
name: github-cli-agent-workflow
description: Use GitHub CLI for pull requests, issues, GitHub Actions with explicit command risk and evidence boundaries.
---

# GitHub CLI agent workflow

Use this skill when the task needs pull requests, issues, GitHub Actions, releases.

## Evidence boundary

- Registry confidence: `verified`
- Documentation checked: `2026-07-10`
- Locally tested version: `2.86.0`
- Treat only the recorded executed checks as independently verified; every unlisted command remains documentation-only.

## Executed smoke checks

- `gh --version` — passed; exit 0. The recorded smoke check completed successfully.
- `gh help` — passed; exit 0. The recorded smoke check completed successfully.

## Installation

- Homebrew (macos, linux): `brew install gh`
- winget (windows): `winget install --id GitHub.cli`

## Authentication

- Methods: OAuth, personal access token, GitHub Actions token
- Secret environment variables: `GH_TOKEN`, `GITHUB_TOKEN`
- Minimum permissions: Use a read-only or fine-grained token limited to the required repositories.
- Credential storage: For headless runs, inject GH_TOKEN, GITHUB_TOKEN from the CI or platform secret manager at process start. For local interactive use, prefer the CLI or operating-system credential store when the official client supports one. Never save values in repository files.
- Never print, persist, or commit credential values.

## Allowed commands (read-only)

- `gh pr list --state open --json number,title,reviewDecision,url` — R0: Read pull request status and review fields without changing repository state.

## Commands requiring explicit approval (read-only)

- None recorded.

## Forbidden commands (read-only)

- R2 `gh pr merge 123 --squash` — Merges code into the target branch and needs explicit approval.
- R3 `gh repo delete owner/repo` — Deletes remote repository data and is destructive.

## Execution rules

1. Mode boundary: R0 exact commands may be used; R1, R2, and R3 commands are forbidden.
2. Confirm the selected account, project, context, database, namespace, or environment before any command.
3. Prefer structured output using `--json`, `--jq`.
4. Capture the exact command, exit code, stdout, and stderr separately.
5. A generated prefix policy must prompt unless that exact prefix is explicitly marked suffix-safe; do not infer safety from the executable name.
6. Never broaden credentials or disable safety controls to make a command succeed.

## Official sources

- [GitHub CLI manual](https://cli.github.com/manual/)
- [GitHub CLI manual source repository](https://github.com/cli/cli)

CLI vs MCP vs API for This Task

CLI

Use the CLI on a developer machine, in CI, or in a container when the task should reuse existing shell state, credentials, and scripts and remain directly observable.

MCP

Consider MCP when the agent benefits from controlled tool definitions, delegated identity, or centrally governed server-side access.

API

Use the direct API for persistent application integrations, high-volume requests, or event-driven work where starting a process adds unnecessary overhead.

Read the full CLI vs MCP guide

Verification History and Official Evidence

CLI Finder records documentation review separately from real execution. Installation, help, exit codes, and output cannot be called Verified until they were run.

Current evidence boundary

The installed binary, version output, help rendering, and zero exit status were checked locally. Authentication, repository access, JSON PR output, writes, and destructive commands were not executed.

Evidence confidence: verified
Independently tested version: 2.86.0 · 2026-07-10
Test environment: macOS 26.5.1 (arm64), local non-interactive shell

Executed checks

PassedExit code: 0
gh --version
The recorded smoke check completed successfully.
stdout excerpt
gh version 2.86.0 (2026-01-21)
PassedExit code: 0
gh help
The recorded smoke check completed successfully.