AI Coding Assistants Documentation

https://stevekinney.com/courses/ai-development
Admin
This documentation provides guides and explanations for using various AI coding assistants such as...
Tokens:19,127
Snippets:326
Trust Score:7.8
Update:3 months ago
Show doc for...
Tokens:
Raw
### Install Dependencies and Validate Content

Source: https://stevekinney.com/courses/ai-development/cursor-environment-configuration

Use these commands to install project dependencies and validate the content. Ensure your environment is set up correctly before proceeding with edits.

```bash
bun install
bun run content:validate
```

--------------------------------

### Subagent Prompt - Better Example

Source: https://stevekinney.com/courses/ai-development/subagent-anti-patterns

Provide specific instructions and expected output format for subagents. This example details the type of information and references required.

```text
Ask a read-only subagent to inspect authentication tests. Return confirmed gaps
only, with file references and the exact regression each test should cover.
```

--------------------------------

### Verify Claude Code Setup

Source: https://stevekinney.com/courses/ai-development/installing-claude-code

Run these commands to verify that Claude Code is installed correctly and to check for common environment and authentication problems.

```bash
claude --version
```

```bash
claude doctor
```

--------------------------------

### Plan Implement Verify Prompt Pattern Example

Source: https://stevekinney.com/courses/ai-development/plan-implement-verify

Example of the Plan, Implement, Verify prompt structure. Use this pattern for tasks that touch multiple files, have security or data risk, require tests, or could be solved in multiple ways.

```text
Goal:
Add validation for expired invite tokens.

Plan:
First inspect the token validation path and the closest tests. Summarize the
smallest implementation path before editing.

Implement:
Write the failing regression test first. Then make the smallest production
change that passes it.

Verify:
Run bun test:unit -- src/lib/server/invitations.test.ts and bun run lint.

Stop:
If a command fails for an unrelated reason, stop and report the command, output,
and likely cause instead of changing unrelated files.
```

--------------------------------

### Cursor Project Configuration Files

Source: https://stevekinney.com/courses/ai-development/installing-cursor

Example structure for project-level Cursor configuration files, including rules and skills directories.

```text
.cursor/
  rules/
    architecture.mdc
    testing.mdc
  mcp.json
```

--------------------------------

### GitHub Actions Workflow Example

Source: https://stevekinney.com/courses/ai-development/integrating-with-github-actions

This example outlines a strategy for handling CI failures. It suggests opening a pull request only for test configuration issues, otherwise commenting with a diagnosis and the failing command.

```text
When CI fails, inspect the failing job. Open a pull request only if the fix is
confined to test configuration. Otherwise, comment with the diagnosis and the
failing command.
```

--------------------------------

### Skill Example

Source: https://stevekinney.com/courses/ai-development/cursor-skills

Demonstrates a procedural skill for executing a defined workflow.

```text
Run the release readiness process and produce the shipping checklist.
```

--------------------------------

### Example Automation Prompt

Source: https://stevekinney.com/courses/ai-development/cursor-automations

This example outlines the conditions and actions for an automation triggered by a failed workflow run on the main branch. It specifies when to open a pull request, what commands to run for verification, and when to report findings without making changes.

```text
When a workflow run fails on the main branch, inspect the failing job logs and
open a pull request only if the fix is confined to test configuration. Run
bun run lint and the failing test command. If the failure is product code or
requires a dependency change, report the diagnosis without editing files.
```

--------------------------------

### Install Cursor CLI on Linux/macOS

Source: https://stevekinney.com/courses/ai-development/installing-cursor

Use this command to install the Cursor command-line interface on Linux or macOS systems.

```bash
curl https://cursor.com/install -fsS | bash
```

--------------------------------

### Install Cursor CLI on Windows

Source: https://stevekinney.com/courses/ai-development/installing-cursor

Use this PowerShell command to install the Cursor command-line interface on Windows systems.

```powershell
irm 'https://cursor.com/install?win32=true' | iex
```

--------------------------------

### Original Notepad Content Example

Source: https://stevekinney.com/courses/ai-development/cursor-notepads

This is an example of content that might have been stored in a deprecated Cursor Notepad. It outlines a set of actions to be taken when editing API routes.

```plaintext
When editing API routes, use zod validation, return typed errors, update tests,
and run bun test:unit.
```

--------------------------------

### Example Permission Rules

Source: https://stevekinney.com/courses/ai-development/claude-code-permissions

Illustrates how to define specific rules for tool and parameter access. Use narrow rules for better control.

```plaintext
Bash(bun test:*)
Read(src/**)
Edit(src/**)
Agent(model:opus)
```

--------------------------------

### Rule Example

Source: https://stevekinney.com/courses/ai-development/cursor-skills

Illustrates a simple constraint-based rule for agent behavior.

```text
Always add regression tests for bug fixes.
```

--------------------------------

### Subagent Prompt - Bad Example

Source: https://stevekinney.com/courses/ai-development/subagent-anti-patterns

Avoid vague prompts for subagents. This example shows a prompt that lacks specificity and a clear expected output.

```text
Have a subagent look into this.
```

--------------------------------

### Example Cloud Agent Prompt Structure

Source: https://stevekinney.com/courses/ai-development/cursor-background-agents

Use cloud prompts that define scope, verification, and stop conditions for agent tasks. This example specifies the exact files to update, acceptance criteria, and conditions for stopping the agent.

```plaintext
Update the invite-token validation path only.

Acceptance criteria:
- Expired tokens return the same public error shape as revoked tokens.
- A regression test fails before the implementation and passes afterward.
- Run bun test:unit -- src/lib/server/invitations.test.ts and bun run lint.

Stop and report the blocker if dependency installation fails or if an unrelated
test failure appears.
```

--------------------------------

### Plan Mode for Decision Making

Source: https://stevekinney.com/courses/ai-development/claude-code-thinking

Use plan mode to force decisions into the open before starting edits, especially when multiple design options exist. This prompt guides Claude to outline the smallest change and compare implementation paths without editing files.

```plaintext
Plan the smallest change that adds account-level rate limits.
Compare middleware and service-layer implementations. Name the files each path
would touch. Do not edit files.
```

--------------------------------

### Install Claude Code via WinGet

Source: https://stevekinney.com/courses/ai-development/installing-claude-code

Install Claude Code using the WinGet package manager on Windows.

```powershell
winget install Anthropic.ClaudeCode
```

--------------------------------

### Install Claude Code via Native Installer (macOS/Linux)

Source: https://stevekinney.com/courses/ai-development/installing-claude-code

Use this command to install Claude Code on macOS and Linux systems using the native installer script.

```bash
curl -fsSL https://claude.ai/install.sh | bash
```

--------------------------------

### Cursor Skill File Structure

Source: https://stevekinney.com/courses/ai-development/installing-cursor

Example structure for a Cursor skill file within the project's .cursor/skills directory.

```text
.cursor/
  skills/
    release-checklist/
      SKILL.md
```

--------------------------------

### Start Interactive Agent Session

Source: https://stevekinney.com/courses/ai-development/cursor-cli

Run 'agent' within a repository to start an interactive terminal session. This mode supports Agent, Plan, and Ask modes, and allows resuming or continuing previous sessions.

```bash
agent
```

--------------------------------

### Install Claude Code via Native Installer (Windows PowerShell)

Source: https://stevekinney.com/courses/ai-development/installing-claude-code

Use this command to install Claude Code on Windows systems using PowerShell.

```powershell
irm https://claude.ai/install.ps1 | iex
```

--------------------------------

### Migrate Notepad Content to Project Rule

Source: https://stevekinney.com/courses/ai-development/cursor-notepads

Transform Notepad content into a scoped project rule by defining its description, applicable file globs, and applying it when necessary. This example shows how to convert a general instruction into a specific rule for API route validation.

```yaml
---
description: API route testing and validation expectations.
globs:
  - 'src/routes/api/**/*.ts'
alwaysApply: false
---

When editing API routes, validate inputs at the boundary, return the existing
typed error shape, update the closest regression test, and run the targeted unit
test command before reporting completion.
```

--------------------------------

### Example Project Rule Configuration

Source: https://stevekinney.com/courses/ai-development/cursor-rules

This rule applies to server-side TypeScript files. It mandates updating or adding the closest unit test before implementation and running specific commands before reporting completion. Set 'alwaysApply' to false for intelligent application.

```yaml
---
description: Testing expectations for server-side TypeScript changes.
globs:
  - 'src/**/*.ts'
alwaysApply: false
---

When changing server-side TypeScript, update or add the closest unit test before
implementation. Run the targeted test command and the repository lint command
before reporting completion.
```

--------------------------------

### Security Review Prompt Example

Source: https://stevekinney.com/courses/ai-development/cursor-bugbot-and-security-review

When using Security Agents, specify the threat model to focus the review on specific risks like authorization bypasses and secret exposure. Ignore style issues and cite exact files for fixes.

```text
Review this pull request for authorization bypasses and secret exposure.
Ignore style issues. Cite exact files and propose the smallest fix for confirmed
security issues.
```

--------------------------------

### Install Claude Code via Homebrew

Source: https://stevekinney.com/courses/ai-development/installing-claude-code

Install Claude Code using the Homebrew package manager on macOS.

```bash
brew install --cask claude-code
```

--------------------------------

### Approve Plan Handoff

Source: https://stevekinney.com/courses/ai-development/claude-code-plan-mode

This is an example of a good handoff for implementing a plan, specifying which option to use, the order of operations, and stop conditions.

```plaintext
Implement option B only. Write the regression test first. Run the targeted test
and bun run lint. Stop on unrelated failures.
```

--------------------------------

### Agent Mode Prompt Example

Source: https://stevekinney.com/courses/ai-development/interacting-with-cursor

Agent mode can search the repository, read and edit files, run shell commands, and use configured tools. Provide a verifiable finish line for the agent's task, including steps like writing tests, implementing fixes, and running verification commands.

```plaintext
Add validation for invalid invite tokens.
Write the failing test first, implement the fix, then run the targeted test and
the repository lint command. Stop if either command fails for an unrelated reason.
```

--------------------------------

### Git and Bun Commands for Version Control

Source: https://stevekinney.com/courses/ai-development/cursor-checkpoints

Use these commands to manage your project's version control and run tests. Ensure your working tree is clean before starting risky agent tasks.

```bash
git status --short
```

```bash
git diff
```

```bash
bun run lint
```

```bash
bun run test:unit
```

--------------------------------

### Prompt for Environment Verification

Source: https://stevekinney.com/courses/ai-development/cursor-environment-configuration

Instruct the agent to verify the cloud environment before making edits. This prevents debugging in a broken setup.

```plaintext
Before editing, verify the cloud environment by running bun run content:validate.
If setup fails, stop and report the failing command and output.
```

--------------------------------

### Plan Mode Prompt Example

Source: https://stevekinney.com/courses/ai-development/interacting-with-cursor

Plan mode is for research before edits, especially when a request involves multiple designs, unknown files, or security implications. Name the decision you need and specify the desired output, such as comparing implementation paths and listing touched files.

```plaintext
Plan the smallest change that adds Slack notifications to failed imports.
Compare the two likely implementation paths and list the files each path touches.
Do not edit files.
```

--------------------------------

### Add and Inspect MCP Servers via CLI

Source: https://stevekinney.com/courses/ai-development/claude-code-mcp

Use these CLI commands to manage MCP servers. The `add` command is for adding new servers, `list` to view existing ones, `get` for details, and `remove` to delete them.

```bash
claude mcp add
```

```bash
claude mcp list
```

```bash
claude mcp get
```

```bash
claude mcp remove
```

--------------------------------

### Chat Prompt Example

Source: https://stevekinney.com/courses/ai-development/interacting-with-cursor

Chat is useful for questions, debugging, and design discussions. It can use open files, selected code, and documentation references. Ask for citations when you need to understand the source of an answer.

```plaintext
Explain how authentication is wired in this project.
Use file references and do not propose changes yet.
```

--------------------------------

### Dated Model Selection Claim Example

Source: https://stevekinney.com/courses/ai-development/cursor-model-selection

When documenting model selection rules, date the claim to ensure its relevance over time, as model availability and provider routing change frequently.

```text
As of 2026-06-23, use Cursor Auto for default implementation work and escalate to
the strongest available reasoning model for architecture reviews.
```

--------------------------------

### Prompting Tool Use Restrictions

Source: https://stevekinney.com/courses/ai-development/claude-code-mcp

Explicitly define the scope and limitations of tool usage within prompts to ensure security and prevent unintended actions. This example restricts the GitHub MCP server to read-only operations.

```plaintext
Use the GitHub MCP server only to read the pull request checks and comments.
Do not post comments or update labels.
```

--------------------------------

### Background Subagent Return Format

Source: https://stevekinney.com/courses/ai-development/cursor-subagents

When using background subagents, specify a clear return format to ensure parallel work provides value. This example defines the expected return content.

```text
Return only: findings, file references, and whether the main task should stop.
```

--------------------------------

### Specifying Relevant Context and Goal in Cursor

Source: https://stevekinney.com/courses/ai-development/cursor-context

Use a short context contract at the top of prompts to define relevant files, existing tests, rules, and the specific goal. This saves search time and guides the model on which evidence to prioritize.

```markdown
Relevant context:
- The public API is in @src/lib/server/invitations.ts.
- Existing validation tests are in @src/lib/server/invitations.test.ts.
- Follow @.cursor/rules/testing.mdc.

Goal:
Reject expired invite tokens with the same error shape used by revoked tokens.
```

--------------------------------

### Example of a Denied Action Message

Source: https://stevekinney.com/courses/ai-development/claude-code-hook-control-flow

When a hook blocks an action, provide a clear and precise message in stderr to inform the agent and human user about the reason and the next steps.

```text
Denied: generated files under applications/website/.generated must not be edited
directly. Change the source generator and rerun the content build.
```

--------------------------------

### Investigative Shell Commands

Source: https://stevekinney.com/courses/ai-development/claude-code-and-bash-scripts

Provides examples of safe, bounded commands suitable for investigation. These commands are designed to be non-destructive and provide specific information without hanging or mutating broad state.

```bash
rg "expired token" src tests

```

```bash
git status --short

```

```bash
bun test src/lib/server/invitations.test.ts

```

--------------------------------

### Initial Repository Orientation Request

Source: https://stevekinney.com/courses/ai-development/claude-code-session-management

Use this read-only request for new repositories to gauge Claude's understanding of build, test, lint, and release commands. Ensure Claude does not invent commands and cite the files used.

```text
Read this repository and summarize the build, test, lint, and release commands.
Do not edit files. Cite the files you used.
```

--------------------------------

### Enter Plan Mode Prompt

Source: https://stevekinney.com/courses/ai-development/claude-code-plan-mode

Use this prompt to initiate plan mode for researching and comparing implementation options without editing files.

```plaintext
Enter plan mode. Research the smallest safe way to add refresh-token rotation.
Compare at most two implementation paths. Do not edit files.
```

--------------------------------

### Targeted Prompt with File References

Source: https://stevekinney.com/courses/ai-development/cost-management

Use file references and specify relevant files to keep context small and focused. Avoid scanning unrelated routes unless necessary for implementation.

```plaintext
Read @src/lib/server/invitations.ts and the closest tests. Do not scan unrelated
routes unless the implementation requires it.
```

--------------------------------

### Custom Subagent Definition

Source: https://stevekinney.com/courses/ai-development/cursor-subagents

Define a custom subagent by specifying its name, description, model, and read-only status. This example shows a 'test-auditor' agent.

```yaml
---
name: test-auditor
description: Find missing regression coverage for a described bug.
model: auto
readonly: true
---

Inspect the changed files and closest tests. Report missing coverage with file
references. Do not edit files.
```

--------------------------------

### Inline Edit Prompt Example

Source: https://stevekinney.com/courses/ai-development/interacting-with-cursor

Use inline edit for surgical changes to selected code. Prompts should be narrow and specific to the selected code block.

```plaintext
Convert this function to return a discriminated union.
Do not change the call sites.
```

--------------------------------

### Project Instructions in CLAUDE.md

Source: https://stevekinney.com/courses/ai-development/claude-dot-md

Defines durable project guidance, including package management and linting steps. Keep this file concise and link to deeper documentation.

```markdown
# Project Instructions

- Use Bun for package management.
- Run `bun run lint` and the closest test before reporting completion.
- Do not edit generated files directly.

```

--------------------------------

### Importing AGENTS.md in CLAUDE.md

Source: https://stevekinney.com/courses/ai-development/claude-dot-md

Demonstrates how to import other files like AGENTS.md when Claude Code does not read them directly. This ensures a single shared instruction source.

```markdown
@AGENTS.md

```

--------------------------------

### Minimal Skill Configuration

Source: https://stevekinney.com/courses/ai-development/cursor-skills

Provides the essential metadata for a Cursor Skill, including its name and a brief description of its purpose.

```yaml
---
name: release-check
description: Verify package release readiness before publishing.
---

Read the package manifest, changelog, generated exports, and test commands.
Run the documented verification commands. Stop and report the exact blocker if a
command fails.
```

--------------------------------

### TDD Prompt for Claude Code

Source: https://stevekinney.com/courses/ai-development/test-driven-development-with-claude

Use this prompt structure to guide Claude Code through a TDD workflow. It clearly defines the bug, the task (write failing test, then implement smallest fix), and verification steps.

```plaintext
Bug:
Expired invite tokens are accepted.

Task:
Write the failing regression test first in the closest test file. Then implement
the smallest production change that makes it pass.

Verification:
Run bun test:unit -- src/lib/server/invitations.test.ts and bun run lint.

Stop:
If either command fails for an unrelated reason, stop and report the failure
instead of editing unrelated files.
```

--------------------------------

### Preserve Compaction State with Pre-Compact Hook

Source: https://stevekinney.com/courses/ai-development/claude-code-hook-examples

Utilize a pre-compact hook to write a status file containing goal, changed files, completed checks, remaining checks, and blockers. This serves as a safety net for long sessions and should not be the sole state store.

```text
goal, changed files, completed checks, remaining checks, blockers
```

--------------------------------

### OAuth-backed MCP Server Login/Logout CLI

Source: https://stevekinney.com/courses/ai-development/claude-code-mcp

Commands for managing authentication with OAuth-backed MCP servers. Use `login` to authenticate and `logout` to end the session.

```bash
claude mcp login
```

```bash
claude mcp logout
```

--------------------------------

### Session Handoff: Git Status and Diffs

Source: https://stevekinney.com/courses/ai-development/claude-code-session-management

When transitioning work between Claude Code and other tools, summarize the filesystem state using git commands. This allows the next tool to continue from the current branch, tests, and files.

```shell
git status --short
```

```shell
git diff --stat
```

```shell
git diff
```

--------------------------------

### Fallback Model Configuration

Source: https://stevekinney.com/courses/ai-development/claude-code-model-selection

Configure fallback models to ensure reliability. Fallbacks should preserve task intent and not silently downgrade critical reviews.

```bash
claude --model opus --fallback-model sonnet
```

--------------------------------

### Import Instructions from Other Files in Claude Code

Source: https://stevekinney.com/courses/ai-development/referencing-files-in-claude-code

Use `@path` references within `CLAUDE.md` files to import shared instructions from other Markdown files. This avoids duplicating guidance across multiple documents.

```markdown
@AGENTS.md
@docs/testing.md
```

--------------------------------

### Default Model Alias Configuration

Source: https://stevekinney.com/courses/ai-development/claude-code-model-selection

Use 'sonnet' for routine tasks and 'opusplan' for planning complex changes. Specify the alias and date for reproducibility.

```text
As of 2026-06-23, use `sonnet` for routine implementation and `opusplan` for
planning complex changes before execution.
```

--------------------------------

### Scoped Rule with Paths

Source: https://stevekinney.com/courses/ai-development/claude-dot-md

Illustrates how to define rules with frontmatter, such as 'paths', to attach the rule to relevant files. API routes must validate input at the boundary.

```yaml
---
paths:
  - 'src/routes/api/**/*.ts'
---

API routes must validate input at the boundary and return the shared error
shape.

```

--------------------------------

### Manage Agent Sessions

Source: https://stevekinney.com/courses/ai-development/cursor-cli

Useful commands for managing interactive agent sessions, including listing available sessions, resuming a previous session, or continuing the latest one.

```bash
agent ls
```

```bash
agent resume
```

```bash
agent --continue
```

--------------------------------

### Execute Agent Request with Cloud Hand-off

Source: https://stevekinney.com/courses/ai-development/cursor-cli

Prepend '&' to an agent request to hand off the task to the cloud. This is useful for longer-running or more intensive agent operations.

```bash
& refactor the auth module
```

--------------------------------

### Skill Folder Structure

Source: https://stevekinney.com/courses/ai-development/cursor-skills

Defines the standard directory layout for a project skill, including SKILL.md, scripts, references, and assets.

```tree
.cursor/skills/
  release-check/
    SKILL.md
    scripts/
    references/
    assets/
```

--------------------------------

### List Existing Git Worktrees

Source: https://stevekinney.com/courses/ai-development/git-worktrees

Displays all active Git worktrees for the current repository.

```git
git worktree list
```

--------------------------------

### Custom Command File Path

Source: https://stevekinney.com/courses/ai-development/claude-code-commands

This is the file path for a custom command in older Claude Code lesson formats.

```text
.claude/commands/review.md
```

--------------------------------

### Using Opus Planning Alias

Source: https://stevekinney.com/courses/ai-development/claude-code-thinking

The `opusplan` alias is recommended for larger changes as it uses the more capable Opus model for planning and a less expensive model for implementation. Always review the plan before proceeding with edits.

```plaintext
Use opusplan. Produce the plan first and wait for approval before editing.
```

--------------------------------

### Automation Pattern with Print Mode

Source: https://stevekinney.com/courses/ai-development/cursor-cli

Utilize print mode for specific, narrow checks like code review. Output findings with file paths as requested. Avoid building hidden release systems; use agent for reasoning about results of script-driven processes.

```bash
agent -p "Review this diff for missing tests. Output only findings with file paths."
```

--------------------------------

### Navigate and Run Agent in Worktree

Source: https://stevekinney.com/courses/ai-development/git-worktrees

Changes the directory to the worktree and executes an agent command. This isolates agent activity to the worktree's branch.

```bash
cd ../project-feature
claude
```

```bash
agent
```

--------------------------------

### Reference Directories in Claude Code

Source: https://stevekinney.com/courses/ai-development/referencing-files-in-claude-code

Use `@` followed by a directory path to inspect conventions or patterns within a specific directory. This is helpful for understanding project structure and established practices.

```markdown
Inspect @src/routes/api/ for existing error response conventions.
```

--------------------------------

### Run Targeted Test Command

Source: https://stevekinney.com/courses/ai-development/claude-code-and-bash-scripts

Instructs Claude to execute a specific test command from package.json. If the test fails, it should report the command and failure before proceeding to edit files. This ensures repeatable testing and clear error reporting.

```bash
Run the targeted test command from package.json. If it fails, report the command
and failure before editing files.

```

--------------------------------

### Bugbot Review Guidance

Source: https://stevekinney.com/courses/ai-development/cursor-bugbot-and-security-review

Follow up on Bugbot findings by verifying issues in the code and closest tests. Fix only confirmed issues and explain why invalid findings are dismissed.

```text
For each Bugbot finding, verify whether the issue is real by reading the changed
code and the closest tests. Fix only confirmed issues. If a finding is invalid,
explain why with file references.
```

--------------------------------

### Asking Specific Tradeoff Questions

Source: https://stevekinney.com/courses/ai-development/claude-code-thinking

Instead of vague requests, ask specific tradeoff questions that give Claude a clear task for its reasoning. This prompt asks for implementation paths, failure modes, and verification commands.

```plaintext
List the two smallest implementation paths, the failure modes of each, and the
verification command that would prove the chosen path works.
```
Tokens:
Raw