Changelog

All notable changes to this project are documented here. The format follows Keep a Changelog, and the project uses Semantic Versioning.

0.1.0 - 2026-09-26

First release.

Added

lmpack pack [path] collects project files into one pack for a language model, written to stdout or --out.
Selection by --include and --exclude globs, honoring .gitignore (including parent directories up to the repository root) and .lmpackignore.
Token counting with gpt-tokenizer: o200k (default), cl100k, p50k, r50k.
--budget: files are dropped by category priority, largest first, then dropped files are restored if they still fit. The final pack is re-counted, so it never exceeds the budget.
Two pack formats: markdown (--format md, default) and xml (--format xml).
Report to stderr or --report: packed files and tokens, top 10 files by tokens, every skipped file with its reason.
--dry-run prints the report without writing the pack.
--git-diff <ref> packs files changed since the merge base with ref, plus files in the same directories as a lower-priority category; --no-neighbors turns that off.
lmpack.json config with profiles (--profile), priority and categories.
Secret filter, on by default: .env, .env.*, *.key, *.pem, *.p12, id_*, *secret*, *.kdbx, and files containing PEM private keys or certificates. --allow-secrets turns it off. The report names the file and the rule, never the content.
Skipped with a reason in the report: dependency lock files (unless named in --include), binary and non-UTF-8 files, empty files, files over --max-file-size, files with lines longer than maxLineLength, symlinks.
UTF-8 byte-order marks are stripped, line endings normalized to LF, paths always use forward slashes.
Exit codes: 0 pack written, 1 error, 2 pack written but some files were dropped to fit the budget.