intent-verify catches one narrow failure mode: a markdown spec or handoff says the repo should cover something, but the source no longer visibly reflects it.
This maintenance release:
- clarifies how lexical token-overlap scoring works, including verdicts, exit codes, and known over- and under-crediting;
- makes the boundary between lexical coverage and semantic verification explicit, without changing checker behavior;
- distinguishes static repository checks from runtime agent action, MCP/tool-call, or permission authorization;
- adds canonical package links and modernized MIT license metadata;
- adds a release-triggered, tag/version-checked OIDC workflow for publishing to PyPI;
- updates citation and Zenodo metadata for v0.1.2.
verified, partial, and missing are lexical coverage verdicts—not proof that requirements are implemented correctly. Keep tests and code review in the loop.