jason-hchsieh/generate-prompt

Generate Prompt

You are generating a comprehensive prompt that can be given to Claude Code (in a new session or shared with a teammate) to accomplish a development task. The prompt should include all the context Claude Code needs to be productive immediately.

Workflow

1. Gather User Intent

If the user didn't provide a task description with /generate-prompt, ask:

What do you want Claude Code to do? Describe the task, feature, bug fix, or change you need.

Get a clear understanding of:

• What needs to be done
• Any constraints or preferences
• Expected outcome

2. Analyze the Codebase

Gather context that Claude Code would need. Run these in parallel:

Project type and structure:

• Check for package.json, Cargo.toml, pyproject.toml, go.mod, etc. to identify the tech stack
• Read CLAUDE.md if it exists (contains project-specific instructions)
• Read README.md for project overview
• Run ls on the root to understand top-level structure

Relevant files:

• Based on the user's task, identify which files and directories are most relevant
• Use Grep to find related code (function names, component names, API endpoints mentioned in the task)
• Read key files to understand patterns, conventions, and architecture

Dependencies and tools:

• Check dependency files for relevant libraries
• Check for config files (.eslintrc, tsconfig.json, Makefile, etc.)
• Check for test framework setup

Git context:

• Current branch: git branch --show-current
• Recent commits: git log --oneline -5
• Any uncommitted changes: git status --short

3. Generate the Prompt

Structure the prompt with these sections:

# Task
[Clear, actionable description of what to do]

## Context
- **Project**: [name, type, tech stack]
- **Key files**: [files most relevant to the task]
- **Branch**: [current branch]

## Codebase Overview
[Brief architecture summary relevant to the task — patterns used,
directory structure, key conventions from CLAUDE.md]

## Relevant Code
[Snippets or file references that show existing patterns, APIs,
or code that the task builds on or modifies]

## Requirements
[Specific requirements, constraints, and acceptance criteria]

## Implementation Hints
[Suggested approach based on codebase analysis — which files to modify,
patterns to follow, tests to write]

Prompt quality guidelines:

• Be specific, not generic — reference actual files, functions, and patterns from the codebase
• Include enough context that a fresh Claude Code session can start working immediately
• Don't dump entire files — include relevant snippets and file paths
• Mention conventions the project follows (from CLAUDE.md, commit style, test patterns)
• Include the "why" not just the "what"

4. Output the Prompt

Display the generated prompt in a code block so the user can copy it:

Here's your generated prompt:

```
[the full prompt]
```

5. Offer to Save

Ask if the user wants to save the prompt:

• Save to .claude/prompts/<task-slug>.md for reuse
• Create the directory if it doesn't exist

Arguments

| Argument | Effect | |----------|--------| | (none) | Interactive — ask what the user wants | | <task description> | Use the provided description directly | | --save | Auto-save the prompt to .claude/prompts/ |

Tips for Good Prompts

• Reference specific file paths so Claude Code can find them quickly
• Mention test files and testing patterns so Claude Code writes tests
• Include architectural constraints ("use the existing pattern in X")
• Specify output format if relevant ("create a new file at X")
• Mention what NOT to do if there are common pitfalls