Code evolves fast. Documentation rots quietly. codeCanon mathematically proves whether your README files, API docs, and embedded screenshots match your JS/TS codebase—and auto-generates git patches to fix them.
Deterministic extraction pipeline combined with Gemma 4 Vision multimodal verification.
AST and regex parsing extracts verifiable ground truths—Next.js & Express routes, package.json scripts, .env keys, ports, and Node engine constraints.
Instant deterministic matching flags OK and STALE claims in milliseconds, escalating only ambiguous SUSPECT claims to the AI Judge to save latency and token costs.
Integrates Gemma 4 Vision to OCR and semantically inspect terminal screenshots and UI captures in docs, detecting outdated CLI flags or broken endpoints.
When drift is confirmed, codeCanon synthesizes clean, deterministic unified git apply patches to auto-repair documentation without human intervention.
Includes a native desktop Tkinter interface (scripts/gui.py) with real-time terminal logging, file tree drift inspection, and one-click patching.
Packaged as a plug-and-play agent skill allowing autonomous AI coding agents to verify markdown accuracy before committing pull requests.
Scans your source tree in milliseconds using AST visitors and regex parsers. It builds a canonical table of truth containing active endpoints, scripts, environment variables, and ports.
Deterministic Matcher → Classifies claims into OK or STALE immediately with zero token costs. Only claims flagged as SUSPECT or referencing embedded terminal images are sent to Gemma 4. It performs visual semantic verification and generates ready-to-apply patch diffs.
Gemma 4 Vision OCR → Validates screenshots and outputs unified git apply diffs. App Router & Pages Router route detection
REST endpoint verbs & middleware parameter mapping
Build commands, dev ports & client environment variables
package.json scripts, dependencies & engines