github_read

GitHub Read

Read access: repos, issues, PRs, files, branches, workflows, runs.

Canonical tool: github_api

ALWAYS use github_api — credentials auto-resolved, raw token never exposed. NEVER call http_request against api.github.com directly.

Common actions

• current_user — confirm auth works
• list_repos — repos affiliated with token (scope: affiliations/owned/member/all)
• list_orgs — orgs you belong to
• list_org_repos — repos within specific org
• repo_info — single repo metadata
• list_issues / list_prs — per-repo
• read_file — file at ref
• list_branches / list_commits
• list_workflows / list_runs — GitHub Actions
• search — global search

See github_api tool docs for full arg list.

Legacy tools

github_repo_info, github_list_issues, github_list_prs, github_read_file — back-compat only. Prefer github_api.

Requirements

GitHub credential in platform credential store with repo read scope.

Credential = Personal Access Token (PAT) — scoped to human user who issued it:

• list_repos scope=affiliations → ONLY repos PAT owner owns, collaborates on, or is org member of. Random public repos NOT included.
• list_orgs → ONLY orgs PAT owner belongs to.
• Empty results = PAT genuinely has no affiliation. Do NOT retry guessing org names.

Anti-hallucination

ALWAYS use real API responses verbatim. NEVER invent:

• Repo names, owners, ids, branch names, file paths, URLs, topics.
• "Probable" repos based on context.
• Repos user "should" have — if API didn't return them, not accessible to this PAT.

Inventory smaller than expected → say so plainly, suggest checking PAT scopes (repo, read:org). Never pad list.

Example — inventory

[
  {"name": "github_api", "args": {"action": "current_user"}},
  {"name": "github_api", "args": {"action": "list_repos", "scope": "owned", "visibility": "private", "max_pages": 5}}
]