Latin: "the one who knows all" — aspirational name for a humble goal: don't lose context.
Model changes. Records compound.
The record layer for AI-driven production. A Claude Code plugin that auto-tracks the files you make, the sessions you run, the tests you inherit, and the decisions that outlive any single chat.
git records what changed. omniscitus records why, who, and what's next — the layer that outlasts every editor swap.
A codebase is a living system. No single person can hold the full picture.
When you delegate more to AI, the bottleneck moves from writing to remembering. omniscitus handles the remembering so you can stay on judgment.
Traditional development relies on the developer's mental model — holding every file, every decision, every test in working memory. omniscitus externalizes that burden into a structured, always-current knowledge layer that persists across sessions, across team members, and across time.
Each feature is designed to work silently, activated when needed, never in your way.
Per-directory file tracking with team attribution
Every file write and edit is recorded automatically via hooks. Split per top-level directory to minimize merge conflicts. Tracks who made each change via git identity.
Topic-based session history with ontology support
No more duplicate session files. Work accumulates into cohesive knowledge units by domain. Define your own domain taxonomy via ontology for consistent classification across the team.
Code tests + LLM prompt evaluation
Language-agnostic test definitions for code. LLM judge-based evaluation for prompts with multi-criteria scoring, manual overrides, and execution logs. Works as a metadata layer over existing test infrastructure — no migration needed.
Onboarding, identity, and cloud migration
New member clones repo, runs /team-init — done.
Every change is attributed via git identity.
Start with git-based sync, migrate to cloud when ready via /cloud-setup.
Visual admin at localhost:3777
Browse blueprints, history units, and test definitions in a dark-themed dashboard. Zero external dependencies. File tree, change timelines, task checklists — all in one place.
Everything is accessible through simple commands inside Claude Code.
/omniscitus-migrate once.
It bootstraps your project’s blueprints and history from existing files. One-time only.
/wrap-up
/follow-up
/blueprint-sync
/birdview
Everything else in the table below is optional — invoked only when the situation calls for it.
| /wrap-up | Record session into topic-based units |
| /follow-up | Review pending tasks across units |
| /blueprint-sync | Sync blueprints with filesystem |
| /birdview | Visual dashboard at localhost:3777 |
| /test-add | Code test scaffold for any file |
| /test-add:prompt | LLM judge evaluation scaffold |
| /omniscitus-migrate | Bootstrap from existing project |
| /team-init | Onboard new team member |
| /ontology-init | Define domain taxonomy |
| /cloud-setup | Generate cloud sync config |
A single .omniscitus/ directory at your project root. Clean, self-contained, non-destructive.
Birdview’s Constellation tab renders your tracked files as nodes in a 3D document space. It’s more visual than practical — the tree picker is faster for real work — but it’s a fun way to get a feel for the shape of your codebase.
/birdview in your project to see your own.
Restart Claude Code after install — omniscitus activates automatically.
Zero configuration. Respects .gitignore. Non-destructive overlay. Remove anytime.
Run /omniscitus-migrate to bootstrap from git history, docs, and tests.
Run /team-init to set up identity, verify hooks, and get started.
If omniscitus is useful, a star helps others find it.