prompt-ast 0.2.0

Parse free-form human prompts into a canonical Prompt AST (structured JSON/YAML).

📦 prompt-ast

Parse human prompts into a canonical, structured Prompt AST.

prompt-ast is a lightweight Python library that converts free-form human prompts into a Prompt Abstract Syntax Tree (AST) — a clean, machine-friendly representation that captures intent, constraints, and expected output.

It treats prompts as first-class artifacts, not just strings.

---

✨ Why prompt-ast?

Modern LLM workflows rely heavily on prompts — but prompts today are:

• unstructured text
• hard to analyze or validate
• difficult to reuse or compare
• tightly coupled to specific tools or models

prompt-ast introduces a foundational layer that sits before agents, chains, or LLM calls:

Human Prompt
     ↓
Prompt AST (this library)
     ↓
LLM / Agent / Tooling

By normalizing prompts into a canonical structure, you unlock better tooling, governance, and reuse — without forcing opinions on how prompts are stored or executed.

---

🧠 What is a Prompt AST?

Inspired by compiler ASTs, a Prompt AST is an abstract, structured representation of a prompt.

Example:

Input

Act as a CTO. Be concise. Explain risks of migrating MySQL to RDS.

Output (JSON)

{
  "version": "0.1",
  "raw": "Act as a CTO. Be concise. Explain risks of migrating MySQL to RDS.",
  "role": "CTO",
  "context": null,
  "task": "Explain risks of migrating MySQL to RDS",
  "constraints": [
    "Be concise"
  ],
  "assumptions": [],
  "ambiguities": [],
  "output_spec": {
    "format": null,
    "structure": [],
    "language": null
  },
  "metadata": {
    "extracted_by": "heuristic",
    "confidence": 0.55
  }
}

---

🚀 Features

• ✅ Parse free-form prompts into a canonical AST
• ✅ Heuristic (offline) parsing — no API key required
• ✅ Optional LLM-assisted parsing (pluggable)
• ✅ JSON and YAML export
• ✅ CLI and Python API
• ✅ Minimal, extensible schema
• ✅ Stateless by design (you own storage, versioning, governance)

---

📦 Installation

pip install prompt-ast

Optional extras:

pip install prompt-ast[yaml]     # YAML export
pip install prompt-ast[openai]   # OpenAI-compatible LLM parsing
pip install prompt-ast[all]

---

🧪 Quick Start (Python)

from prompt_ast import parse_prompt

ast = parse_prompt(
    "Act as a senior backend architect. Be concise. Explain system design trade-offs.",
    mode="heuristic"
)

print(ast.to_json())

---

🖥 CLI Usage

prompt-ast normalize \
  "Act as a CTO. Be concise." \
  --mode heuristic \
  --format json

LLM-assisted mode (optional):

export OPENAI_API_KEY=...
prompt-ast normalize \
  "Design a scalable data pipeline." \
  --mode hybrid \
  --use-openai

---

🧩 Design Principles

prompt-ast is intentionally lean:

• ❌ No storage
• ❌ No prompt registry
• ❌ No agent framework
• ❌ No vendor lock-in

Instead, it provides a clean abstraction that other systems can build on.

This makes it ideal for:

• internal tooling
• prompt governance pipelines
• evaluation frameworks
• AI platform teams
• research & experimentation

---

🛣 Roadmap / Future Scope

The current version focuses on core normalization. Planned and possible future directions include:

Near-term

• Improved heuristic extraction accuracy
• Better ambiguity detection
• Schema refinements based on real-world usage
• More LLM adapters (Anthropic, local models)

Mid-term (opt-in tooling)

• Prompt linting (missing task, conflicting constraints)
• Prompt diffs (semantic-ish comparison)
• Prompt fingerprints / hashes
• Validation rules

Long-term (out of scope for core)

• Prompt registries
• Governance workflows
• Evaluation frameworks
• UI tooling

These are intentionally not bundled into the core library.

---

🤝 Contributing

Contributions are very welcome — especially in these areas:

• Improving heuristic parsing
• Adding test fixtures (real-world prompts)
• Refining the schema
• Documentation improvements
• CLI UX enhancements

Getting started

git clone https://github.com/<your-username>/prompt-ast.git
cd prompt-ast
poetry install
poetry run pytest

Please:

• keep changes small and focused
• add tests for behavior changes
• avoid introducing heavy dependencies

Open an issue before large changes — discussion is encouraged.

If you’re new to the project, check issues labeled good first issue — they are small, well-scoped tasks ideal for first-time contributors.

---

📜 License

MIT License

This project is released under the MIT License, which means:

• ✅ Free for personal and commercial use
• ✅ Permissive and business-friendly
• ✅ Attribution required
• ❌ No warranty provided

Perfect for individual contributors and early open-source traction.

(See LICENSE file for full text.)

---

🌱 Why this project exists

Prompts are becoming one of the most important interfaces in modern software — yet they lack the basic tooling we take for granted in other domains.

prompt-ast is an attempt to provide a small, composable foundation that the ecosystem can build upon.

If that resonates with you, you’re in the right place.

---

⭐ Support the project

If you find this useful:

• ⭐ Star the repo
• 🐛 Report issues
• 💡 Share ideas
• 🧑‍💻 Contribute code