LS-2024-001 | REV.1.0.0 | STATUS: OPEN

TYPE: COMMUNITY | CLASS: OPEN-SOURCE

40.7128°N / 74.0060°W | 2024.08.13

[ 000 ] — COMMUNITY SPECIFICATION SHEET

LANDSLIDE⛰

Weird ideas, made real.

LANDSLIDE is an open community for people who like building strange, ambitious things — especially around multi-agent systems and human-machine collaboration. Visual work is welcome too: generative art, interactive experiments, anything that looks as good as it runs.

ORG-ID LandslideLab

PLATFORM GitHub

REPOSITORIES 10

MEMBERS OPEN

LICENSE OPEN-SOURCE

FOCUS AGENT SYS / HMC

ENTRY NO GATEKEEPING

LAST-UPDATE 2026.08.14

ACCEPTING MEMBERS

NO EXPERIENCE REQUIRED

[ 001 ] Community Overview SEC.A / ROW.01

If you have a crazy idea, or just want to help build someone else's, this is the place. LANDSLIDE exists at the intersection of ambitious engineering and creative experimentation — where multi-agent architectures meet human intuition, and where code becomes art.

We believe the most interesting work happens when diverse minds collide: researchers, artists, engineers, hobbyists, students. No credentials required. No interviews. No gatekeeping. Just curiosity and the willingness to build something strange.

Everything in this org is open source — that's the point. Every line of code, every experiment, every failure is public. We learn in the open.

[ 002 ] Focus Areas & Research Directions SEC.B / ROW.02

◇

DIR.001

Multi-Agent Systems & Agent Governance

Building control planes for agent frameworks. Human-in-the-loop approval systems. MCP + REST transports for agent orchestration.

◇

DIR.002

Human-Machine Collaboration

Tooling and experiments that blur the line between human creativity and machine capability. Augmentation, not replacement.

◇

DIR.003

Visual & Generative Works

Art that is code, code that is art. Generative art, interactive experiments, visual computing. If it looks as good as it runs, we want it.

◇

DIR.004

Whatever You Bring

The weirdest ideas often become the most interesting projects. If you can articulate it, we'll help you build it. No idea is too strange.

[ 003 ] Active Projects & Repositories SEC.C / ROW.03

Sphinx / SphinxGo

The missing middle layer between agent frameworks and the humans who must supervise them. Sits between LangGraph / OpenAI SDK / CrewAI / any MCP-capable agent and the reviewer, providing the control plane every framework leaves out.

DIFFICULTYSHA3-256 decision chain + Ed25519 signatures; REST + WebSocket + MCP + web console in one self-contained binary (Go, zero-cgo); framework-agnostic HITL approval protocol.

ARTIFACTSphinx (Python) + SphinxGo (Go module v0.1.0 on pkg.go.dev); single-binary HITL control plane; Apache-2.0.

MARKETenterprises shipping agents need auditable human oversight; HITL approval with tamper-evident decision capture is the #1 governance requirement in agentic deployments — and it must work across every framework.

RESEARCHdecision capture as an audit primitive: how can machine decisions become evidence? Sphinx formalizes the human-in-the-loop contract between agent frameworks and accountability.

Python / Go
PUBLIC
AGENT GOV

Tiltmeter / TiltmeterGo

Agent Skill Trust Platform: sign, govern and debug agent skills. Ed25519 signing/verification, governed registry with approvals and audit, and replay debugging for agent skills.

DIFFICULTYEd25519 key management + keyless ambient OIDC auth (GitHub Actions token exchange); SQLite governed registry; deterministic replay debugger; pure-Go implementation with no cgo.

ARTIFACTTiltmeter (Python/FastAPI) + TiltmeterGo (Go module v0.1.0 on pkg.go.dev); signing CLI + verify gate + registry; 170 tests.

MARKETthe agent-skills economy is exploding with unverifiable, supply-chain-risky skills. Tiltmeter answers "what will this skill actually do — and did it change?" — trust infrastructure for machine-actionable code.

RESEARCHverification and governance for executable knowledge: signing, replay and audit as the substrate for trustworthy agent ecosystems.

Python / Go
PUBLIC
SKILL TRUST

saltglass

Agent knowledge crystallization & governance: distill agent sessions into human-approved, signed, sovereign knowledge crystals — local-first, encrypted, vendor-neutral.

DIFFICULTYprovenance chains with Ed25519 signing; knowledge half-life management (expiry / re-validation / conflict); human approval gates; age-encrypted local-first storage with dual output (agent YAML + human AGENTS.md / SRS cards).

ARTIFACTsaltglass (org repo); crystallization pipeline + governance + verification gate; Apache-2.0.

MARKETagent sessions generate ephemeral context that vanishes or leaks. saltglass turns them into governed, reusable, human-approved knowledge assets — the missing memory-and-governance layer for agentic teams.

RESEARCHthe knowledge lifecycle (crystallization, half-life, re-validation) treated as a first-class systems problem — bridging human memory, organizational learning and machine knowledge.

Python
PUBLIC
KNOWLEDGE GOV

MONOLITH

Digital landslide laboratory — a terminal-native observatory that treats a git repository as a geologic body: commits as sediment, faults as conflicts, erosion as reverts. Observe, don't fix.

DIFFICULTYgit plumbing internals; commit-as-sediment modelling; fault/erosion analysis over repository history; terminal-first UX.

ARTIFACTMONOLITH (org repo, terminal application); observation engine + analysis reports.

MARKETdeveloper analytics as geology: teams get a new lens on how software really decays, conflicts and rebuilds — observability for code as a living structure, not a static artifact.

RESEARCHmetaphor as model: can geological process models (sedimentation, faulting, erosion) explain and predict software evolution? A cross-domain exploration of complexity science and software engineering.

Terminal
PUBLIC
OBSERVATORY

yieldpoint

Dynamic power handover engine for human-machine collaboration and multi-agent systems. Autonomy as a first-class runtime policy: control moves dynamically between human and machine, driven by live confidence / risk / cost / timeout signals — not static "ask human when stuck" interrupts.

DIFFICULTYruntime policy engine over a five-level autonomy model (operator → machine-only); zero-dependency core with framework adapters (LangGraph.js / OpenAI Agents SDK / Vercel AI SDK / raw loops); built-in scenario simulation scoring quality / safety / autonomy / cost.

ARTIFACTyieldpoint on npm (v0.2.x, provenance-signed via OIDC); TS dual ESM/CJS; 155 tests; MIT.

MARKETevery production agent system needs supervised autonomy. Existing HITL tools are binary interrupts; yieldpoint is the missing primitive for dynamic, provable control transfer — the governance ask of 2026 agent deployments.

RESEARCHformalizes dynamic authority handover as an engine, grounding the human-agency autonomy literature (Knight Institute levels) in runnable, testable code — a core building block of accountable human-machine collaboration.

TypeScript
npm
HMC

arena

Framework-agnostic multi-agent scenario simulation and evaluation engine. Same scenario suite, many orchestration strategies, comparable quality / safety / autonomy / cost / latency metrics — with statistical significance and CI regression gates.

DIFFICULTYdeterministic message-bus scheduler (seeded RNG, timeout/deadlock detection); YAML/TS scenario DSL; A/B strategy comparison with significance tests; RegressionGate + GitHub Action; CLI (run / compare / report).

ARTIFACT@noahisarider/arena on npm (v0.2.x, provenance-signed); GitHub Action; 189 tests, 96.6% coverage; MIT.

MARKETJS/TS has no framework-agnostic measurement layer for multi-agent systems: LangGraph-, AI-SDK- and SaaS-bound tools evaluate LLMs, not orchestration strategies. arena makes "which strategy is safer/cheaper" an answerable, CI-enforced question.

RESEARCHreproducible scenario suites turn multi-agent collaboration into an empirical science — the instrumentation layer for studying handover policies, role structures and emergent coordination.

TypeScript
npm
MULTI-AGENT EVAL

[ 004 ] Join Procedure — One Step SEC.D / ROW.04

Open an Issue

Create a new issue in the LANDSLIDE repository with the following title format:

Join request: <your GitHub username>

Tell Us (Optional)

Share what you want to build or explore. This is optional but fun — it helps us understand your interests and connect you with relevant projects.

Get Invited

A maintainer will review your request and send you an invitation. That's it. Your membership is public by default; you can set it to private anytime in your org settings.

[ 005 ] Ground Rules & Community Standards SEC.E / ROW.05

R.01 Be kind. Treat everyone with respect and patience.

R.02 Be curious. Ask questions, explore ideas, challenge assumptions.

R.03 Give credit where it's due. Acknowledge contributions openly.

R.04 Everything is open source. That's the point. No closed doors.

R.05 Build in public. Share your progress, failures, and learnings.

R.06 No idea is too weird. If you can articulate it, we'll explore it.

[ 006 ] Getting Started — Onboarding Sequence SEC.F / ROW.06

Join via Issue

Open a join request issue as described in Module 004.

Introduce Yourself

Say hello in an issue or discussion. Tell us who you are and what excites you.

Find Your Spark

Explore existing projects or open an issue with your own idea. Build. Break. Learn. Repeat.

[ CTA ] — CALL TO ACTION

Let's make
something strange.

No experience requirements. No interviews. No gatekeeping.
Just bring your curiosity and your weirdest ideas.

→ Open Join Request on GitHub △ Explore Sphinx Project

REF: LS-2024-001 / JOIN.PROC.v1.0
github.com/LandslideLab