proml-cli 0.1.0

Prompt Markup Language toolkit for structured LLM workflows

ProML (Prompt Markup Language)

Author: Johan Caripson

ProML is a structured markup language for Large Language Model prompts. The project now ships a full toolchain: a formal specification, a production-ready parser and runtime, a CLI, a local registry, a constrained-decoding adapter layer, and developer ergonomics such as formatting, documentation, and editor support.

✨ Highlights

• Formal spec & docs — docs/ contains the language guide, minimal grammar, and 29 governing principles for prompt engineering.
• Reference parser — proml/parser.py builds an AST, validates block order, semver, repro tiers, policies, pipelines, and test definitions.
• Strict I/O test runner — proml_test.py parses .proml files, enforces JSON Schema/regex/grammar constraints, and runs caching-aware assertions.
• Constraint engine — pluggable validators for regex, JSON Schema, and CFG grammar; ships with a Guidance-compatible adapter for decoder-time enforcement.
• Engine profiles & caching — structured metadata for model, temperature, token limits, and cost budgets with hash-based cache keys and adapter registry (OpenAI, Anthropic, Local, Ollama, Stub).
• CLI & registry — proml command (init, lint, fmt, test, run, bench, publish, import) plus a YAML registry for semver-aware module discovery.
• Developer experience — schema-aware formatter, VS Code extension skeleton, MkDocs plugin, and example prompts under test_prompts/.

🚀 Quick Start

Clone the repo and ensure dependencies are installed (Python 3.10+ recommended). The CLI runs directly via python3 -m proml.cli.

Install locally in editable mode:

python3 -m pip install -e .

For full tooling (docs, tests, Guidance adapter) use the dev extras:

python3 -m pip install -e .[dev]

Format & Lint

python3 -m proml.cli fmt test_prompts/sentiment_analysis.proml
python3 -m proml.cli lint test_prompts

The formatter is schema-aware and preserves inline comments thanks to ruamel.yaml.

Run Tests

python3 proml_test.py test_prompts/sentiment_analysis.proml
python3 proml_test.py test_prompts/caching_example.proml

Execute a Prompt

python3 -m proml.cli run test_prompts/sentiment_analysis.proml \
  --input comment="I love this product" --provider stub

Use --provider stub for offline evaluation; registered adapters handle OpenAI, Anthropic, Local backends, and more.

Registry Workflow

python3 -m proml.cli publish test_prompts/sentiment_analysis.proml
python3 -m proml.cli import com.example.sentiment --version ^1.0.0

Published modules are tracked in proml_registry.yaml with integrity hashes and reproducibility metadata.

🧰 Developer Tooling

• Formatting & comments: schema-aware formatter retains inline annotations across blocks.
• Guidance adapter: proml.adapters.GuidanceGenerationAdapter lets you enforce regex/grammar/schema constraints inside Guidance programs.
• MkDocs plugin: mkdocs_proml_plugin.py renders .proml modules into documentation pages automatically.
• VS Code extension: tools/vscode/ provides syntax highlighting and snippets for .proml files.

📚 Documentation

Browse the specification and principles in the docs index. The minimal spec includes a formal EBNF grammar, block invariants, determinism tiers, policy semantics, pipelines, and testing requirements.

🔬 Tests

The repository includes targeted tests for the parser, formatter, and adapters under tests/. Run them with:

PYTHONPATH=. python3 tests/test_formatter.py
PYTHONPATH=. python3 tests/test_guidance_adapter.py

🌐 Publishing Docs

The project is preconfigured for MkDocs out of the box. After installing mkdocs and mkdocs-material, run mkdocs serve to preview the docs site locally. A Hugo configuration is also included for alternative static site generation.

🙌 Contributing & Next Ideas

Issues and PRs are welcome! Impactful directions include:

• Package the CLI as a distributable (pip install proml).
• Build richer adapters (Guidance grammar, OpenAI function-calling integration).
• Expand formatter lint rules (comment style, schema normalization hints).
• Publish example modules and tutorials demonstrating real-world prompt workflows.

Have fun building reproducible, testable LLM workflows with ProML.