# Github

> Use gh with explicit repository context, bounded machine-readable output, complete review evidence, safe waits, and no implicit mutations.

- Skill: `erikhoward/github` (Agent Skill)
- Install (CLI): `npx skillmds@latest add erikhoward/github`
- Raw SKILL.md: https://api.skillmd.com/api/skills/erikhoward/github/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- License: MIT
- Author: erikhoward (https://skillmd.com/u/erikhoward)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/erikhoward/github

---


# GitHub CLI

The model already knows `gh` syntax. This skill covers agent execution risks.

## Rules

- Set explicit repository context with `--repo owner/name` when the current checkout is not authoritative.
- Request only needed fields with `--json`. Use `--jq` when projection simplifies nested output, not as ceremony.
- Bound listings and status output. Do not truncate diffs, failure logs, or other evidence required for a complete review. Inspect large evidence in sections.
- Use noninteractive flags. Set `GH_PAGER=cat` and disable update notices for automation where appropriate.
- Give waits a time budget. Treat pending, failed, cancelled, and timed-out checks as distinct outcomes.
- Prefer structured request bodies and unique temporary files in the environment's approved temporary directory.
- Never print tokens, authorization headers, signed URLs, or secret values.
- Do not enable auto-merge, delete branches, close issues, publish releases, merge, or push unless the user authorizes that specific mutation.
- Select merge method explicitly when merging is authorized. Never rely on an interactive prompt.
- Inspect all commits and the full base-branch diff before creating or reviewing a pull request.

Keep command output proportional to the question. A short listing can be limited. Evidence used to approve a change must remain complete.