Scores, audits, and disciplines natural-language artifacts across Claude Code, Codex CLI, and Antigravity. Tier-aware scoring with per-tool overlays, 51 rules, 6 vocabulary principles, cross-repo dashboards, registry-free drift detection.
NLPM's own vocabulary as a noun-verb map. Click into the self-audit for the interactive panel.
Deterministic, penalty-based quality scoring for every NL artifact — skills, agents, commands, rules, hooks, CLAUDE.md. Anchored in 51 rules with bright-line definitions and bad/good examples.
Opt-in controlled-vocabulary enforcement. Bootstrap a registry from your corpus, declare canonical noun/verb pairs, and let `/nlpm:check` flag drift. Registry-free `/nlpm:vocab-drift` runs as a no-commitment health check.
Manifest-vs-disk consistency, broken references, orphan components, behavioral contradictions. The single check that no other validator covers systematically across Claude Code, Codex CLI, and Antigravity.
Continuous audits of public Claude Code plugins (Codex and Antigravity discovery coming next). Findings, vocabulary drift, security signals — all on one cross-repo dashboard.
`bin/nlpm-check` is a single-file Python (stdlib only). Drop into pre-commit hooks or CI without Claude Code installed.
Audits feed `rule-health.py`. Rules with high false-positive rate surface for refinement; rules with strong real-world warrant become exemplars cited in the docs.
It treats natural-language artifacts as code. A 100-point rubric, 51 rules anchored in primary sources, cross-component manifest-vs-disk checks, opt-in vocabulary discipline (R51) backed by a six-principle framework derived from OntoClean, DDD, ISO 25964, and BPMN/Event Storming. Everything runs locally as Claude Code slash commands, and there's a continuous auditor pipeline that scores public plugins and feeds findings back into the rulebook.
