Every AI review tool claims to learn your conventions. Exposing those learnings as readable objects makes the claim checkable — you can see what it thinks your team's rules are and whether they are right. Combined with update_issue_status taking feedback, the loop is visible rather than implied. resolve_github_user_contacts is the practical one: turning a username into a Slack mention is exactly the small friction that stops a finding reaching the person who can fix it.
Cubic's endpoint over their AI code review. Eleven tools spanning scans and their issues, the repository wiki, and the reviewer's accumulated learnings.
- list_scans and get_scan — repository scan summaries, then aggregated issues with filters and paging
- get_issue and get_pr_issues — one issue with its full code context, or everything published for a pull request
- update_issue_status — triage an issue, with optional feedback
- list_learnings and get_learning — what the AI reviewer has learned about this repository
- list_wikis, list_wiki_pages, get_wiki_page — the repository wiki
- resolve_github_user_contacts — GitHub usernames to email and Slack mentions
A cubic account with access to at least one repository — membership of that repo's installation. OAuth with dynamic client registration; a legacy API key still works during migration. TLS 1.2+ in transit, AES-256 at rest, and least-privilege GitHub App scopes documented per scope.
Paste a URL, then authorize — add the endpoint to your client, then approve the OAuth consent screen