Step 4: Refine and customize

Improve output quality by editing prompts, adjusting context, and using custom templates.

Edit the prompts

Each section has its own prompt file in prompts/. These are plain text files you can edit directly.

prompts/
├── description_prompt.txt
├── installation_prompt.txt
├── usage_prompt.txt
└── contributing_prompt.txt

Open any prompt to see its structure — each has a {REPO_CONTEXT} placeholder where the script injects repository information, plus instructions that shape the AI's output.

Prompt editing techniques

Specify your audience:

# Add to description_prompt.txt
Target audience: Backend developers integrating this library into existing Python applications.

Add formatting requirements:

# Add to usage_prompt.txt
Format all commands as: $ command (with dollar sign prompt)
Show expected output after each command as a plain code block (no language hint).

Exclude content you'll write manually:

# Add to contributing_prompt.txt
Do NOT include a code of conduct section — I will write this separately.
Do NOT include issue templates — link to CONTRIBUTING.md instead.

Tune the tone:

# Add to description_prompt.txt
Tone: casual and friendly. This is an open source side project, not enterprise software.

After editing a prompt, regenerate only that section:

python readme_generator.py --input ./my-project --sections description --output README.md

Improve context quality

The AI output is only as good as the context it receives. Two ways to improve it:

Add a stub README: Create a brief README.md in your project with a one-paragraph description. The script reads it and passes it as additional context.

# my-project

A CLI tool for converting Markdown files to PDF with custom styling.
Built for technical writers who work in docs-as-code workflows.

Add docstrings to entry points: If your main entry file has no comments or docstrings, add a module-level docstring before running the generator.

"""
my-project: Convert Markdown files to styled PDF.
Supports custom CSS, MkDocs theme compatibility, and batch processing.
"""

Use a custom template

The default template assembles sections in a fixed order. For a different layout, copy and modify the default template:

cp templates/readme_template.md.j2 templates/my-template.md.j2

Edit templates/my-template.md.j2. Available variables:

Variable	Content
`{{ description_section }}`	Full description section markdown
`{{ installation_section }}`	Full installation section markdown
`{{ usage_section }}`	Full usage section markdown
`{{ contributing_section }}`	Full contributing section markdown
`{{ project_name }}`	Project name string
`{{ language }}`	Primary language string
`{{ license }}`	License type string
`{{ date }}`	Generation date (YYYY-MM-DD)

Example custom template adding a badges line:

{{ description_section }}

![License](https://img.shields.io/badge/license-{{ license }}-blue)
![Language](https://img.shields.io/badge/language-{{ language }}-green)

{{ installation_section }}

{{ usage_section }}

{{ contributing_section }}

Run with your template:

python readme_generator.py --input ./my-project --template templates/my-template.md.j2

Iteration workflow

A productive iteration cycle:

Edit one prompt file
Regenerate that section: --sections <name>
Review the output
Repeat until satisfied
Move to the next section

Avoid regenerating all sections at once during iteration — it's slower and costs more API tokens.

Next step: Best practices →