cr8script

An English-shaped scripting language for LLMs and anyone who wants quick scripts without Python's footguns. One file. No build. No imports. Reads the way you'd describe the work out loud.

Website: cr8script.com — language tour, animated atlas, and a playable demo.

Quick start

python3 cr8script.py examples/tour.cr8        # the language tour
python3 cr8script.py                          # REPL
python3 cr8script.py --test                   # run the golden suite

Python 3.9+. No pip, no venv, no flags.

A taste

let sales = [
  { product: "widget", region: "east", amount: 12.50 },
  { product: "gadget", region: "east", amount:  8.00 },
  { product: "widget", region: "west", amount: 15.00 },
  { product: "doodad", region: "east", amount: 99.00 },
]

let by_product = sales
  | group by product
  | summarize { total: sum(amount), n: length(items) }
  | sort by total descending

for each row in by_product
  show f"{row.product}: {row.total} across {row.n} order(s)"
end

doodad: 99 across 1 order(s)
widget: 27.5 across 2 order(s)
gadget: 8 across 1 order(s)

examples/tour.cr8 walks the rest of the language end-to-end.

What makes it different

• Reads like English. is greater than, is at least, for each, where, sort by. No ==, !=, >=, <=.
• Honest types. No truthy/falsy. "5" + 3 is an error. 5 is "5" is an error. if 0 then is an error.
• Honest decimals. 0.1 + 0.2 is exactly 0.3. One number type.
• No null surprises. Only nothing. Indexing a missing key returns nothing; typing the wrong field name is a hard error with a "did you mean" hint.
• Immutable by default. let is forever; var opts into change.
• Pipelines as a first-class verb set. where, sort by, take, map, group by, summarize — bare names inside resolve to record fields.
• Errors that teach. Every error names the line, the value, and a one-line hint.
• LLM-shaped diagnostics. --check-json emits structured { line, message, hint } so a model can self-correct before running.

For language models

Three files are aimed at agents and the humans wiring them up:

• LLMS.md — condensed grammar + rules sheet. Read this first.
• AGENTS.md — system-prompt template, tool definitions, and the self-correction loop spelled out for an integrator.
• examples/agent_loop/ — a broken .cr8 file, the actual --check-json JSON output, and the corrected version. The end-to-end demo of cr8 → diagnostics → fix.

The self-correction loop:

python3 cr8script.py --check-json file.cr8     # structured diagnostics
python3 cr8script.py file.cr8                  # run

Status

v1.1. Single-file Python interpreter (cr8script.py, ~2.8k lines): lexer, parser, tree-walking evaluator, REPL, static checker, ten golden tests. The language is independent of Python — only the bootstrap is.

Built-in modules: math, http, time, json, csv. Top-level: length, sum, count, average, min, max, range, to_text, to_number, keys, type, assert. Strings support f"..." interpolation.

Out of scope until pulled by a real use case: regex, file I/O, dates, modules / imports, async. Not on the roadmap: classes (values, not objects).

Layout

cr8script.py     single-file interpreter
LLMS.md          rules-and-grammar sheet for LLMs
AGENTS.md        system-prompt template + agent tool definitions
examples/
  hello, tour, load_test, make_game, make_mindmap
  agent_loop/    broken → diagnostics → fixed (the LLM self-correction demo)
  api_ingest     fetch JSON, summarize, emit CSV
  validate       list of records → { ok, errors } per item
  report_md      structured data → markdown report
testdata/        golden tests
game.html        a small playable demo (output of make_game.cr8)

More

python3 cr8script.py --check file.cr8          # static checks (human output)
python3 cr8script.py --check-json file.cr8     # static checks (JSON output)
python3 cr8script.py --lex file.cr8            # token dump (debugging)
python3 cr8script.py --ast file.cr8            # AST dump  (debugging)