nlpm

What NLPM checks

NLPM treats natural language artifacts as programs that can be linted. Just as ESLint scores JavaScript and ruff scores Python, NLPM scores the markdown files that drive AI behavior: skills, agents, commands, rules, hooks, prompts, CLAUDE.md, and memory files. It ships as a Claude Code plugin with eight slash commands, each doing one job. The ls command discovers and inventories artifacts in a repo. The score command grades quality on a 100 point scale where every issue carries a fixed penalty, so the same artifact always yields the same number. The check command runs cross component consistency checks, including the manifest versus disk problem where a SKILL.md exists on disk but is missing from plugin.json and therefore invisible after install. The fix command auto repairs what is mechanical. The trend command tracks score history over time, and the test command runs NL artifact tests against spec files in a test driven style. A security-scan command looks for risk in executable artifacts, and an init command sets up a project. The rubric applies one universal floor plus overlays for Claude Code, Codex CLI, and Antigravity, so the same artifact is judged against the rules that apply to the tool that will load it. This tier aware design is what lets NLPM support multiple agent harnesses from one rule catalog.

Scoring system and author tooling

Scores in NLPM start at 100 and decrease. The bands are 90 to 100 for production ready, 80 to 89 for good with minor gaps, 70 to 79 for adequate, 60 to 69 for weak, and below 60 for a rewrite. The default pass threshold is 70 and can be configured in a project file. The scorer auto classifies each artifact by its path and applies the matching rules, covering more than 20 artifact types and the 50 Rules of Natural Language Programming. For plugin and skill authors, NLPM provides a standalone Python validator called bin/nlpm-check that has no Claude Code dependency and runs in pre commit hooks or CI on any tool's artifacts. It is a single Python 3.11 or newer file with no external dependencies, and it runs the deterministic subset of the check command, including the manifest versus disk consistency check that the README says no other validator currently covers. Templates ship for a drop in git pre commit hook and a GitHub Actions workflow. A PostToolUse hook reminds you to score a file after writing or editing it, though it is advisory and does not block writes. Configuration supports relaxed, standard, and strict levels that set the threshold at 60, 70, and 80 respectively, and individual rules can be overridden with a local config file.

Self evolving auditor pipeline

NLPM also runs as a self evolving GitHub Actions pipeline that audits real plugin repositories, contributes fix pull requests, harvests teaching examples from clean ones, and feeds learnings back into its own rule catalog. Repositories that audit clean at a score of 90 or higher produce a teaching artifact under an exemplars directory, and a weekly job opens a human gated pull request adding real world example links to the rules file. A two stage drift detector re validates every audit's rule id against the rubric and against the rule's title keywords, catching mislabeled ids. The README states that a 2026 sweep found 990 mislabeled rule ids across 128 historical audits. As of 2026-05-19 the auditor pipeline had filed 278 pull requests across 44 distinct accepting repositories, with a 71 percent acceptance rate, and two repositories had crossed into rule adoption where the maintainer credited NLPM in a changelog. The project is part of the xiaolai plugin marketplace and also reachable through Anthropic's official community marketplace, though the community copy can lag the maintainer's version by up to about 24 hours. The pipeline is an unusual example of a linter that improves its own rules by mining the ecosystem it checks.

Editorial conclusion

NLPM is released under the ISC license and, at the time of writing, the repository recorded 130 stars and was last updated on 2026-08-24.