GitHub / GitLab

Use this skill when the user asks to inspect repos, issues, pull requests, merge requests, code history, branches, or to make a small code change in a hosted repo.

Choose the account deliberately

Use the current credential manifest to choose an account authorized for this conversation that fits the user's intent. Run commands with execute and an explicit sandbox_id; keep that ID throughout checkout, edits, checks, and push. Personal and shared accounts are both valid choices; neither is an automatic fallback when the other fails. If the intended account is unclear before a write, ask. Do not infer permission from a login merely being present on the computer.

• Personal login: use the authorized gh, glab, or Git login. Check the active provider account (for example, gh api user --jq .login) and the Git transport's auth configuration; a CLI API identity alone does not prove which account Git will use.
• Shared org credential: if listed in the prompt, select its slug and use the core-hosted smart HTTP remote in the use-shared-credential skill. This uses the configured shared account, even when a personal OAuth connector is live. Its name is an admin label, not a verified upstream username.
• Connected app: use only advertised capabilities. API access does not by itself establish native Git transport access or configure the CLI's active login.

For an existing checkout, inspect the remote and applicable credential-helper, SSH, proxy, and HTTP-header configuration. A reused checkout may still select a previous account. Commit author metadata is separate from transport identity. Resolve an unknown identity before a write; do not probe access by pushing.

Logging in

Use the native login only after an explicit authentication rejection. A TLS timeout, connection reset, rate limit, or GitHub server error is not evidence of logout or permission expiry. Do not consume another credential grant to repair transport.

For a confirmed missing or expired GitHub or GitLab login, follow the interactive-login skill: it runs gh auth login / glab auth login in a temporary home as a background job, relays the code and URL, and saves the login to the keychain so later commands use it through execute.credentials.

Read-only work

Check auth and inspect repo state:

gh auth status
gh repo view OWNER/REPO --json name,description,url,defaultBranchRef
gh issue list --repo OWNER/REPO --state open --limit 20
gh pr list --repo OWNER/REPO --state open --limit 20

For GitLab:

glab auth status
glab repo view GROUP/PROJECT
glab issue list --repo GROUP/PROJECT
glab mr list --repo GROUP/PROJECT

Use git log, git show, and git diff for codebase-change summaries. Cite commit hashes, PR/MR numbers, and links.

Checkout and edits

Clone or fetch only the repo the user asked for:

gh repo clone OWNER/REPO repo
cd repo
git checkout -b codex/small-change

Keep changes scoped. Run the repo's tests. If a push or PR/MR creation is requested, prepare the branch and summary first.

Writes require approval

Pushing branches, creating PRs/MRs, merging, closing issues, editing labels, changing repo settings, releases, or workflows are writes. Ask for approval before running the write command.

After approval:

git push -u origin codex/small-change
gh pr create --repo OWNER/REPO --title "..." --body-file pr.md

For GitLab:

git push -u origin codex/small-change
glab mr create --repo GROUP/PROJECT --title "..." --description-file mr.md

Report the final URL and leave enough context for review.

Wait for GitHub CI

Use the bundled helper instead of a bare gh pr checks --watch or a handwritten shell loop. Use the absolute directory reported by the skill read in place of <skill dir>. It is one blocking watcher process, not repeated agent turns:

node "<skill dir>/scripts/watch-ci.mjs" --repo OWNER/REPO --pr NUMBER --head REVIEWED_HEAD_SHA

Capture the full head SHA from gh pr view NUMBER --repo OWNER/REPO --json headRefOid before review. Use that same SHA for the watcher and the final merge gate. Start the watcher with execute using the selected sandbox_id and background:true, then subscribe with background action watch and its returned process_id. Keep one watcher per PR. Set the process lifetime longer than the helper's 30-minute deadline; use --timeout-ms for a different bounded deadline. Use only credentials authorized for that background process; a live foreground grant does not authorize background work.

The helper polls GitHub internally and emits JSON status records. It retries only recognized transient transport/server errors with bounded backoff. Pending checks remain pending. Real CI failures, authentication/authorization failures, changed heads, unknown errors, exhausted retries, and deadline expiry all stop with nonzero status. Explicit API rate limits stop as rate_limited, not authentication errors. Empty check sets, skipped checks, and cancelled checks are not successful evidence. Do not interpret an exit code alone as "checks concluded": read the status record. Do not suppress a nonzero exit or print a success marker after it. Keep set -e.

Success means the checks observed for the expected head passed, not permission to merge. Re-read checks and review threads immediately before merging. Wait for async reviewers on that exact head; resolve findings and obtain a fresh-context adversarial review. After any push or head change, discard old success/review evidence and repeat. Only merge when all gates pass, with GitHub's atomic head guard:

gh pr merge NUMBER --repo OWNER/REPO --squash --match-head-commit REVIEWED_HEAD_SHA

Never use auto-merge or bypass repository gates. This helper does not retry GitLab commands or change their authentication behavior.