Cursor

Cursor reads AGENTS.md directly. Open the example repository locally and complete your language setup before starting.

Documentation reviewed: September 5, 2026. If you need to install Cursor or set up an account, follow its quickstart.

First session

  1. Open the example repository in Cursor and complete the setup for your language path in its terminal. This gives you a known working environment for the agent to use.
  2. Open the Agent panel, choose Ask in its mode picker, and send the orientation prompt.
  3. Check the instruction sources and compare the reported entrypoint and commands with the files you just used.

Use root AGENTS.md for shared guidance. For file-specific rules, use .cursor/rules/*.mdc; plain .md files there are ignored. Personal rules live in Cursor settings. Cursor rules.

For specific rules, use Cursor’s editor and check description, globs, and alwaysApply. Try files that should and should not trigger the rule.

Ask, plan, or implement

Use the mode picker or Shift+Tab: Ask reads and explains; Agent searches, edits, and runs commands. Review edits as they appear; use Stop if the task drifts. Ask mode, Agent behavior.

Choose Plan to discuss an approach, then Build to implement. Save to workspace keeps the plan in the project; otherwise plans start in your home directory. Small, clear changes can go straight to Agent. Plan mode.

Check Settings → Agents → Approvals & Execution separately from the mode picker. Auto-review combines allowlists, available shell sandboxing, and model assessment; it can approve calls automatically and make mistakes. Allowlist sets explicit rules; an empty list replaces Ask Every Time. Run Everything removes those checks and sandboxing. Cloud execution differs. Run Modes.

Choose a model in this client

Use the model selector in Chat or Agent. Account and team settings affect choices; switching models keeps Cursor’s rules and controls. Record Auto if selected: routing can vary, and teams may hide the model identity. Cursor model selection and availability.

Included models need no separate API key. Your own key under Cursor Settings → Models supports documented chat models, with provider-specific limits; Tab still uses Cursor’s models. Switching clients requires that client’s setup. Cursor API key support.

Try the review skill

Ask Agent to use the included .agents/skills/plot-review/ and check what loads. Cursor also supports .cursor/skills/; personal equivalents use ~/.agents/skills/ or ~/.cursor/skills/. See the folder map. Cursor skills.

For cloud or remote work, check skill availability: personal skills may need synchronization or packaging. A committed project skill identifies the shared version. Cursor skill locations and remote availability.

Create a reviewer agent

Save this role as .cursor/agents/plot-reviewer.md, or under ~/.cursor/agents/ for personal reuse. readonly: true restricts writes; model: inherit uses the main agent’s model. Cursor subagents.

---
name: plot-reviewer
description: Review plotting code and its rendered image against the contract.
model: inherit
readonly: true
---

Read AGENTS.md and use .agents/skills/plot-review/SKILL.md.
Review the requested diff, image, and alt text against the local figure
specifications. Use supplied test results; do not edit files.
Report findings with source locations and name any checks you could not perform.

Ask Have plot-reviewer review my plotting changes. Compare its findings with the source and tests. A single-file task may not need delegation; extra agents add coordination time. Cursor subagent tradeoffs.

Check the interpreter and test results

Check the interpreter, actual test results, and every changed file. Record checks the reviewer could not perform. Keep the task fixed when comparing models.