What is lint-plugin-evidence?

This project is a lint plugin designed to keep AI coding agents honest when they work on their own. When tools like Claude Code or Codex write code without human supervision, they can skip a requirement and still report the task as complete. Evidence Graph addresses this by making every project requirement demand an explicit acknowledgement from the code, test, or document that claims to satisfy it. The core idea is a JSDoc tag called @evidence. A developer writes a comment above a function that names the exact requirement it fulfills and states why. The compiler does not judge whether that reason is actually true. Instead, it forces the agent to commit to a specific, concrete claim. If an AI fabricates a justification, that claim sits right beside the code for a human reviewer to catch. The result is that an agent can still lie, but it cannot lie by leaving things out. Every configured obligation must be accounted for, or the build fails. The plugin runs through a TypeScript compiler wrapper called ttsc, not the standard tsc tool. You install it alongside @ttsc/lint, then write a configuration file that declares your evidence graph. The graph defines relationships, such as React components in your source folder claiming to implement sections in your documentation folder. If a documentation section exists but no component cites it, the compiler reports an error pointing to the missing citation. These violations appear in the same stream as normal type errors, with no separate checking step required. Graphs can chain together in layers. Feature documents can claim to build on requirement documents, components can claim to implement feature rules, and tests can claim to verify both. The plugin supports Markdown headings, TypeScript exports, Prisma models, and OpenAPI operations as evidence units. The full README is longer than what was shown.