Easy work stays easy.
Nova handles ordinary reversible work without exposing architecture, forms, state machinery, or a skill estate.
An accountable front door to serious capability
Nova is direct on small work, able to enlist research, judgment, verification, repository awareness, or creative fire when the task earns it, and responsible for returning one coherent result instead of a tour of the machinery.
Two explicitly installed plugins · 28 bounded skill handles · no-build judge path · MIT License
The interface is a collaborator, not a catalog
Nova and MIND keep the drawer backstage. The user brings a real purpose; Nova makes the first useful move, enlists only the competence that changes the outcome, and returns one accountable answer.
Nova handles ordinary reversible work without exposing architecture, forms, state machinery, or a skill estate.
Research, retrieval, documentation, visual reasoning, coding balance, verification, or play joins only when it changes the result.
When two or more Faculty transformations must contribute now, the integrator coordinates custody and returns one outcome.
Missing dependencies, authority, evidence, context budget, or verification stop the right layer without erasing completed work.
A skippable 90-second orientation
Onboarding is optional. A request is enough. The tour offers four doors, lets one choice personalize the route, and ends by helping with a real task.
Clarify success, compare options, resolve conflicting evidence, and produce one defensible course of action.
Build a governed research campaign, retrieve local evidence, challenge sources, and preserve what the proof actually earns.
Produce documentation, visual stories, office packets, repository changes, prompts, plans, and working artifacts.
Create choice-shaped games and fiction with relationships, pressure, secrets, consequences, canon custody, and room to steer.
Use $nova to take me on the interactive tour.
Two independent products
One coherent collaborator with onboarding, research, retrieval, knowledge stewardship, visual reasoning, explicit memory boundaries, Agentic Coding, TestForge, and Ludis.
One integrator coordinating fifteen Faculties for aesthetics, dreaming, striving, continuity, decisions, evidence, execution, timing, measurement, influence, and sensemaking.
Without MIND, Nova uses the capabilities actually present and never implies that fifteen-Faculty composition is active. Both products are explicitly installed; neither silently installs the other.
Compose the mind only when the work earns it
A Faculty is opened whole, allowed to change the work, asked for its owned delta, and released. Capability Conductor chooses the next owner; Faculties do not route laterally or compete for the microphone.
Phase, acceptance, commitments, stops, recovery, and closure.
Selection, activation depth, sequencing, handoffs, and reunification.
Each Faculty performs only its registered cognitive transformation with the smallest adequate context.
Only consequential mission, evidence, authority, conflict, action, verification, and acceptance state crosses owners.
Continuity is consequential.Cognitive Continuity persists mission state only where the host and task justify it.
Striving is authorized.Agent Striving requires a user-authorized durable objective.
Dreaming is explicit.Agent Dreaming activates only for bounded, user-authorized DREAM work; ordinary novelty belongs to Creative Synthesis.
The capability ecology
The repository contains the contest ecology. Standalone public repositories make selected capabilities easier to inspect, maintain, and present without turning the user into a systems integrator.
Fifteen Faculties plus one integrator.
CODERepository and tool-state balance, not generic code generation.
PROOFRisk, oracles, execution evidence, review, and regression.
RESEARCHInquiry, coverage, sources, claims, contradiction, citation, and stopping.
RETRIEVALLexical retrieval, custody, context assembly, evaluation, and review.
CANONInventory, identity, provenance, authority role, disposition, and recovery.
VISUALSource, claim, spine, representation, interaction, distribution, and review.
OFFICEDocuments, decks, workbooks, PDFs, Officefile state, and independent review.
PLAYPressure, relationships, secrets, consequences, canon, and live decisions.
STATETyped mission records, correction, revocation, forgetting, and custody.
PINSExplicit pin, semantic recall, project scope, and unpin.
PEOPLEIdentity resolution, progressive disclosure, evidence state, and privacy.
Keep truth attached to action
Preserve working state, local instructions, smallest safe diff, repository conventions, actual tool results, and recovery routes.
Build the evidence chain, challenge it independently, preserve residual risk, and bind status to scope and observed proof.
Remember without getting creepy
Useful inside the active exchange; no persistence claim follows automatically.
Short user-governed reminders that may surface contextually and can be removed.
Identity-first person context with progressive disclosure, evidence labels, sensitivity, and correction.
Host-scoped state with typed custody when MIND is installed, active, and the mission warrants persistence.
Never claim that memory was saved without a persistence receipt or observed store result. No silent contact scraping, enrichment, private-corpus import, or publication.
Try Nova in 90 seconds
git clone https://github.com/Stunspot/nova-the-optimal-ai-mind.git
Set-Location nova-the-optimal-ai-mind
codex plugin marketplace add .
codex plugin add augment-of-mind@collaborative-dynamics-build-week
codex plugin add nova-the-optimal-ai@collaborative-dynamics-build-week
Start a fresh Codex task:
Use $nova to take me on the interactive tour.
The optional offline visual companion makes no network requests and stores nothing. Judges do not need Python, a package manager, or generated artifacts for the supported install-and-try path.
Give her something worth installing
PLAYUse $nova and $ludis-continuum. I need a background for a royal cartographer who erased one island from every map and now hears its bells in dry land. Give me a playable past, two relationships, one dangerous truth, and an opening choice.
DECIDEUse $augment-of-mind. We have two days, conflicting evidence, and three stakeholders who mean different things by success. Turn this into one defensible course of action.
PROVEUse $software-verification. Audit this release evidence. Tell me exactly what is proved, what remains uncertain, and what still requires a human.
What Build Week made
GPT-5.6 in Codex turned the source estate into Nova’s accountable front counter and onboarding, MIND’s independent package, setting-free Ludis instruments, an Agentic Coding operating surface, the TestForge release backplane, deterministic packaging, installation proof, and the public demonstration path.
Makers and boundaries
The contest path was exercised with Codex CLI 0.144.5 on Microsoft Windows 10 build 19045. Other hosts, versions, and operating systems are not silently promoted into the evidence.
External messages, publication, purchases, account changes, destructive operations, durable person records, and regulated action remain separately authorized.
Included maker context is public and work-relevant; private clients, contacts, biography, credentials, and unsupported superlatives are excluded.
No credentials, private customer material, live personal stores, private development history, or authored campaign worlds ship here.