Skip to main content

Auto-formatter for LLM prompt files

Project description

promptfmt

Auto-formatter for LLM prompt files.

PyPI License: MIT Python 3.9+

What It Does

promptfmt is an opinionated auto-formatter for LLM prompt files, similar to what Prettier does for JavaScript or Black does for Python. It normalizes whitespace, section delimiters, template variable syntax, line wrapping, and structural formatting for YAML/JSON prompt files.

After formatting, promptfmt verifies semantic equivalence -- it re-parses the formatted output and confirms that the meaning of the prompt has not changed (same messages, roles, variables, and metadata).

Installation

pip install promptfmt

Dependencies: prompttools-core >= 1.0, typer >= 0.12, rich >= 13.0, watchfiles >= 0.21

CLI Commands

promptfmt format

Format prompt files in-place or check formatting.

# Format a single file
promptfmt format prompts/greeting.yaml

# Format all prompt files in a directory (recursive)
promptfmt format prompts/

# Check only -- exit 1 if any file would change (for CI)
promptfmt format prompts/ --check

# Show a unified diff of changes
promptfmt format prompts/ --diff

# Quiet mode (no output except errors)
promptfmt format prompts/ -q

Options:

Option Default Description
--check false Check only, do not write files. Exit code 1 if changes needed.
--diff false Show unified diff of changes.
--delimiter-style ### Target delimiter style: ###, ---, ===, ***, ~~~
--variable-style double_brace Target variable syntax: double_brace, single_brace, angle_bracket
--max-line-length 120 Maximum line length for wrapping. 0 to disable.
--quiet, -q false Suppress non-error output.

promptfmt init

Generate a default .promptfmt.yaml configuration file in the current directory.

promptfmt init

This creates a .promptfmt.yaml with default settings:

# promptfmt configuration
delimiter_style: '###'
variable_style: double_brace
max_line_length: 120
wrap_style: soft
sort_metadata_keys: true
indent: 2
exclude:
  - 'vendor/**'
  - '*.generated.*'

Formatting Rules

promptfmt applies five rule categories in order:

1. Whitespace Normalization

  • Normalize line endings to LF
  • Strip trailing whitespace from every line
  • Remove leading blank lines at file start
  • Collapse 3+ consecutive blank lines to 2
  • Ensure file ends with exactly one newline

2. Delimiter Normalization

Normalizes section delimiters (---, ===, ***, ~~~) to a single consistent style. Code blocks (```) are preserved and not modified.

Supported styles: ###, ---, ===, ***, ~~~

3. Variable Syntax Normalization

Converts all template variable references to a single syntax style. Inline code spans (`code`) are preserved and not modified.

Style Syntax Example
double_brace {{var}} Hello {{name}}
single_brace {var} Hello {name}
angle_bracket <var> Hello <name>

Common HTML tags (div, span, p, br, etc.) are excluded from variable detection when using angle bracket style.

4. Line Wrapping

Wraps lines exceeding the configured maximum length at word boundaries. The following are never wrapped:

  • Lines inside code blocks
  • URL lines
  • Table rows (starting with |)
  • Heading lines (starting with #)

Leading indentation is preserved on continuation lines.

5. Structure Normalization (YAML/JSON only)

For YAML and JSON files:

  • Re-serializes with consistent indentation (default: 2 spaces)
  • Sorts metadata keys with priority ordering: model, name, description, defaults appear first, then remaining keys alphabetically
  • Within message objects, sorts keys with role before content
  • Re-applies whitespace normalization after structural changes

Configuration (.promptfmt.yaml)

Key Type Default Description
delimiter_style string "###" Target delimiter style
variable_style string "double_brace" Target variable syntax
max_line_length int 120 Max line length (0 to disable)
wrap_style string "soft" Wrapping approach
sort_metadata_keys bool true Sort keys in YAML/JSON
indent int 2 Indentation width for YAML/JSON
exclude list [] Glob patterns to exclude

Programmatic Usage

from promptfmt import format_file, format_content, FmtConfig, is_equivalent

# Format a file
config = FmtConfig(delimiter_style="---", variable_style="double_brace")
result = format_file("prompts/greeting.yaml", config)

print(result.changed)      # True if content changed
print(result.equivalent)   # True if semantically equivalent
print(result.formatted_content)

# Format raw content
from prompttools_core import PromptFormat
formatted = format_content(content, PromptFormat.YAML, config)

# Check semantic equivalence between two parsed prompts
from prompttools_core import parse_file
original = parse_file("original.yaml")
modified = parse_file("modified.yaml")
print(is_equivalent(original, modified))

CI Integration

GitHub Actions

- name: Check prompt formatting
  run: promptfmt format prompts/ --check

Pre-commit Hook

#!/bin/sh
promptfmt format prompts/ --check || exit 1

GitLab CI

check-prompt-format:
  script:
    - pip install promptfmt
    - promptfmt format prompts/ --check

Exit codes:

Code Meaning
0 All files formatted (or no changes needed)
1 Files need formatting (with --check)
2 Errors occurred during formatting

License

MIT License. Author: Scott Converse.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

promptfmt-1.0.0.tar.gz (12.6 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

promptfmt-1.0.0-py3-none-any.whl (13.7 kB view details)

Uploaded Python 3

File details

Details for the file promptfmt-1.0.0.tar.gz.

File metadata

  • Download URL: promptfmt-1.0.0.tar.gz
  • Upload date:
  • Size: 12.6 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.14.3

File hashes

Hashes for promptfmt-1.0.0.tar.gz
Algorithm Hash digest
SHA256 aaf6d00be8e132ec6f3555820b8665774832a6e74fde6f8a29c9033942e99d34
MD5 bfa79fdb1367ff790456074a77cbe871
BLAKE2b-256 2137b307243aee036778d8f85e8213329d2e97bef944e385393af45f010a89cf

See more details on using hashes here.

File details

Details for the file promptfmt-1.0.0-py3-none-any.whl.

File metadata

  • Download URL: promptfmt-1.0.0-py3-none-any.whl
  • Upload date:
  • Size: 13.7 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.14.3

File hashes

Hashes for promptfmt-1.0.0-py3-none-any.whl
Algorithm Hash digest
SHA256 a19a90fb4084fa1e0f7e194644ec80ecfe6dd826108eb72363d33e1475505fd2
MD5 7c035338016454e947df51f9b75ccba9
BLAKE2b-256 bee43d5d9d9b909e3c0800adea6489963659e74a0c94950aaaede7b8a5dd8d2f

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page