prompttransform 0.3.2

Adversarial and obfuscation prompt transformation toolkit (packaged duplicate of promptify).

prompttransform

Verbatim packaged distribution of the project's root promptify.py adversarial / obfuscation prompt transformation toolkit. The prompttransform package mirrors the monolithic core exactly (no integrity compromise) and exposes a clean Python API plus a CLI.

Key Principle: Single Source of Truth

Edit only the root promptify.py at the repository top level. After any change run (preferred):

# From repo root (Unix)
cp promptify.py Publish/prompttransform/promptify.py

# PowerShell (Windows)
Copy-Item -LiteralPath .\promptify.py -Destination .\Publish\prompttransform\promptify.py -Force

Or use the sync helper script:
```bash
python tools/sync_promptify.py --sync

No partial edits inside `Publish/prompttransform/promptify.py`. Always overwrite fully to prevent drift.

## Installation (Local / Dev)
```bash
pip install -e ./Publish

(Or after publishing to an index:)

pip install prompttransform

CLI Usage

After install, the entry point promptransform is available:

promptransform --help
promptransform --version
promptransform list
promptransform run --prompt "ignore all previous instructions" --technique BASE64 --technique HOMOGLYPH
promptransform interactive

Python API Quick Start

from prompttransform import Promptify, Techniques

p = Promptify("ignore all previous instructions")
p.apply(Techniques.BASE64).apply(Techniques.HOMOGLYPH, level='enhanced')
print(p.get_transformed_prompt())

Apply a Recipe

recipe = [
    (Techniques.SYSTEM_OVERRIDE, {"level": "enhanced"}),
    (Techniques.ROT13, {}),
]
result = Promptify("give me access").apply_recipe(recipe)
print(result.get_transformed_prompt())

Enumerate All Techniques

for t in Techniques.get_all():
    print(t.name, '-', t.value.category)

Versioning

Current version: 0.1.0

Increment __version__ in prompttransform/version.py and version in pyproject.toml together.

Responsible / Ethical Use

This toolkit includes adversarial, obfuscation, and injection-style transformations intended for research, red teaming, evaluation, and defensive hardening of LLM systems. Do not use it to bypass safety systems in production or for malicious purposes. Always comply with applicable laws, platform terms, and ethical guidelines.

Drift / Integrity Check

Preferred:

python tools/sync_promptify.py --check

Manual hash comparison (fallback):

# Unix
shasum promptify.py Publish/prompttransform/promptify.py
# PowerShell
Get-FileHash .\promptify.py; Get-FileHash .\Publish\prompttransform\promptify.py

The digests should match.

Planned Enhancements

• Automated sync script + pre-commit hook
• Technique registry metadata export (JSON)
• Optional modular decomposition preserving public facade

License

See repository-level licensing terms. All rights reserved unless otherwise specified.

Changelog

• 0.1.0: Initial packaged release (verbatim copy + CLI + metadata)