Skip to main content
Skills are built-in capabilities that ship with cubic’s agent integration. They activate based on what you ask your coding agent — no slash commands needed, just describe what you want in natural language.

​
Install skills

First connect the MCP server with OAuth. Then install the skills and slash commands:

​
Best prompts to start with

These prompts cover the workflows most teams use every day:

​
Built-in skills

​
Use the prompts by stage

  • Before you push Ask your coding agent to "review my code" or "loop until clean" to catch obvious issues early.
  • On an open pull request Ask "check cubic comments" after cubic finishes the GitHub review. The agent will fetch all unresolved threads, analyze them, and fix only issues that are real and worth addressing. It will then push the changes and automatically resolve the handled threads.
  • When learning the codebase Ask questions like "how does auth work?", "where is rate limiting handled?", or "what does this team care about in reviews?".
  • When handling codebase scan results Ask "show open codebase scan findings for this repo" or provide an exported CSV and ask the agent to investigate or fix it. The agent uses each finding’s stable issue ID to retrieve the full report, so older CSV links with a mismatched scan ID still work. A CSV is optional: when you ask about the current repository, the agent detects its Git remote and fetches findings directly.
For the step-by-step local-review flow and the copy-paste AGENTS.md / CLAUDE.md guidance, see Local CLI review.

​
Slash commands

If you prefer explicit commands, use these. Claude Code uses /cubic:.... Other supported agents typically use /cubic-....
The /cubic-run-review and /cubic:run-review commands require the cubic CLI. Use npx @cubic-dev-ai/cli install -g on Windows or curl -fsSL https://cubic.dev/install | bash on macOS and Linux. See Local CLI review for setup details.