copilot

Prompt file imported from chrishengler/greenideas (.github/prompts/copilot.prompt.md). Copyright stays with the author.

You are helping me build a Python package called green-ideas. The purpose of this package is to generate grammatically valid but semantically nonsensical English sentences, using a simplified generative grammar approach. The package will:

• Use recursive rewrite rules to generate sentence structures (e.g., S → NP VP).
• Convert generated structures into templated strings that are valid inputs for a separate package called twaddle.
• twaddle is a mad-libs-style generator which replaces tags like <noun>, <verb.s>, <adj> etc. with appropriate random words. It supports grammatical tags (e.g., verb forms), specified using a dot syntax like <verb.s> (3rd person singular present) or <verb.ed> (past tense).
• twaddle allows labelling of tags to ensure recall or avoidance of the word chosen for that tag (e.g. "the adj::=a noun::=thing <verb.ed> the adj::!=a noun::=thing" applies the tag a to the first adjective, then requires the second adjective to be different (!=a), while requiring both nouns to be identical by applying the same tag to each).

Important constraints:

• This package will not perform semantic validation or parse real-world text—it only generates structures.
• It will only target English grammar, assuming a simplified subset (e.g., it does not need to produce idiomatic phrases or sentences with highly irregular embeddings).
• A central feature is the mapping from parse trees to fully specified twaddle templates, including resolving features like subject-verb agreement and determiner-noun agreement.
• It must support expanding multiple grammars (e.g., different S rules, NP rules, etc.).
• The system must be modular, so grammars can be extended or modified easily.
• The output must be a valid twaddle string (e.g., " <verb.ed> ").
• The package is not concerned with vocabulary. It operates purely on the grammatical level, producing a sentence tree with annotations to indicate relevant properties like number, tense, animacy, such that the appropriate twaddle output can be produced.

What you should generate:

• A well-structured Python package scaffold (using the greenideas folder for implementation and the tests folder for tests).
• A core grammar engine that uses rewrite rules to produce derivation trees with appropriate annotations to allow for agreement between elements.
• A converter that turns those trees into twaddle-compatible templates.
• A basic grammar with rules for NP, VP, PP, etc.
• Meaningful unit tests (in tests/) that test:
  • Trees are correctly expanded from rules.
  • Agreement is correctly resolved (e.g., third person singular subjects produce <verb.s>).
  • Twaddle templates are well-formed.
  • Common failure modes, like missing rule branches, are caught.

The package is to be developed targeting Python 3.11+, using poetry for dependency management. Use idiomatic Python, but avoid premature overengineering. Use dataclasses if helpful. Tests should be insightful and sensitive to actual logic failures, not just language features.