numbers-le

Get every hardcoded number in a codebase into one list, with the file and line each came from, so a person can check them: the auditor confirming the tax rate in the code is the rate in the specification, the reviewer looking for the retry limit that changed, the actuary reading a retention window.

npx numbers-le src/
npx numbers-le --values --dedupe src/ > numbers.txt
cat rates.toml | npx numbers-le --stdin --format toml

This is the same command line as the Rust crate numbers-le, and it behaves identically: same flags, same walk, same positions, same output, same exit codes. Install whichever toolchain you already have. It reads documents with the parsers of the numbers-le editor extension, so the few documents the extension and the crate read differently, which the crate's SPEC lists, are read here as the extension reads them. When a parser refuses a broken document, both report it as unparsed under the same Failed to parse JSON: prefix, and the rest is each parser's own words.

What it reads

Every file. JSON, JSONC, YAML, CSV, TSV, TOML, INI and dotenv are parsed. Python, Rust, Go, Java, Kotlin, C#, C, C++, JavaScript, TypeScript, SQL and shell are read by their own literal syntax, hex, binary, octal, digit separators and type suffixes included, so u32 is a type and not the number 32. Anything else is scanned as text and reported as format unknown. Directories are walked the way ripgrep walks them: .gitignore, .ignore and hidden files are respected unless you pass --no-ignore or --hidden. A binary file is passed over and counted in the summary.

Every number is printed exactly as JavaScript prints it: 0xFF is 255, 1_000 is 1000, 1e21 is 1e+21. It reports what is there and nothing else. Which numbers matter is yours to decide.

Output

stdout carries one JSON report per file whenever it is piped or redirected, and nothing when it is a terminal. stderr lists every number with its position. --values prints just the values, one per line, wherever stdout goes.

$ numbers-le src rates.toml
rates.toml:1:18  30
rates.toml:2:7  0.2
src/pricing.ts:1:21  255
src/pricing.ts:2:22  1000
4 numbers in 2 files

{"file":"src/pricing.ts","format":"typescript","numbers":[{"value":"255","notation":"hex","line":1,"column":21},{"value":"1000","notation":"decimal","line":2,"column":22}],"diagnostics":[],"summary":{"numbers":2,"unlocated":0}}

Columns count UTF-16 code units, as an editor does. A number a parser resolved from text no run of digits spells, such as TOML's 0x1A, carries no line or column, and summary.unlocated counts them.

Exit codes

grep's convention:

• 0: numbers found.
• 1: none found. That is an answer, not an error.
• 2: the question was malformed, or --strict and a text file could not be read.

For agents

numbers-le mcp serves the Model Context Protocol on stdio with two tools: extract_numbers for a document's text and numbers_le_scan for files and directories. Hosts that want the document tool alone can install numbers-le-mcp.

numbers-le --help lists every option.

MIT licensed. Part of letools.