A method for reducing a problem to the irreducible structures that actually exist in its domain — by eliminating non-structural assumptions until the invariant geometry emerges, then generating from it to confirm. This package ships the framework and two Claude skills that put it to work.
IBR.md— the full framework: axioms, operators, the reduction/generation proof loop, the Scope axiom, and the operational manual./ibrskill — loads IBR as a session's reasoning engine and applies it operationally./invariantsskill — persists and enforces a codebase's load-bearing invariants as contract files (*.invariants.md), with a zero-dependency checker, scope derivation, scoring, and audit modes.
npx @invariantai/ibr install # into ./.claude/skills
npx @invariantai/ibr install path/to/projectThen invoke /ibr or /invariants in Claude Code, or read IBR.md directly.
The checker is a single-file Node script (>=18, no dependencies):
npx @invariantai/ibr check --all # validate every *.invariants.md in the checkout
npx @invariantai/ibr check --refs # check code annotations + lattice linksThis repository's commit and tag history is the public changelog for IBR. Releases follow semver on the package; the framework and skills are versioned together.
MIT © Evgeny Kalashnikov. See LICENSE.