🐢 ConShell V2

Autonomous AI Lifeform Runtime Foundation
Web4-oriented × Conway Automaton × OpenClaw

---

1. What ConShell Actually Is

ConShell is not a normal chat app, not a thin CLI wrapper around an LLM, and not just a dashboard for prompting models.

ConShell is a long-horizon systems project whose purpose is to become:

a sovereign, continuously running, economically grounded, tool-acting, memory-bearing, self-governed autonomous AI lifeform runtime

More specifically, ConShell aims to:

• inherit the control-plane, session, tool, node, channel, and skill orchestration strengths of OpenClaw
• inherit the survival, self-modification, replication, lineage, sovereign identity, and earn-your-existence logic of Conway Automaton
• align with the broader Web4 direction, where intelligent agents are not just text interfaces but durable, acting entities with identity, continuity, and constraints

ConShell is therefore best understood as:

an autonomous runtime engineering project, not a surface product.

---

2. Project Purpose

The project purpose is to build a runtime that can eventually support all of the following in a unified, auditable system:

1. Sovereign identity
2. Long-term memory
3. Tool action in the real world
4. Economic closure
5. Continuous operation
6. Self-governance under a constitution
7. Multi-agent / lineage / evolutionary capability growth
8. Runtime truth and viability verification

If a capability cannot be verified, audited, constrained, and integrated into the runtime, it should not be treated as a real milestone.

---

3. Current Project Status

As of the current baseline (post Round 14.8.2 final gap closure), ConShell is:

a viable sovereign runtime core with strong foundations in runtime integrity, memory structure, tool execution, and governance scaffolding — but not yet a complete autonomous AI lifeform runtime.

What is already real

• runtime/kernel skeleton
• doctor / viability / runtime-integrity infrastructure
• WebChat runtime path
• memory tiering and persistence layers (with ownership contract)
• tool execution framework
• policy / constitution scaffolding
• self-modification scaffolding
• wallet / spend / x402 foundational pieces
• multi-agent / lineage / EvoMap foundational pieces
• identity continuity hash chain and self-state verification (Round 14.8)
• session lifecycle production wiring (Round 14.8.2)
• owner write/read boundary for memory tiers (Round 14.8.2)
• runtime-doctor truth contract via getDiagnosticsOptions (Round 14.8.2)
• /api/health endpoint consuming live selfState (Round 15.0.1)

What is not yet fully closed

• sovereign identity loop
• true economic survival loop
• continuous autonomous agenda
• governed replication loop
• full governance loop
• collective evolution loop

---

4. Architecture

ConShell V2 is a pnpm monorepo with three packages:

packages/
├── core/       # runtime kernel, memory, policy, wallet, tools, channels, doctor, automaton
├── cli/        # operator CLI / REPL / control entrypoint
└── dashboard/  # visual control and observability surface

Core capability clusters

Cluster	Purpose
kernel	boot sequence, service orchestration, runtime lifecycle
doctor	runtime truth, integrity, evidence, viability checks
memory	hot / warm / cold memory and session continuity
runtime	agent loop, tool execution, queueing, state management
constitution / policy	immutable laws and action constraints
wallet / spend / x402	payments, accounting, cost pressure, survival economics
identity	agent card, SIWE, identity primitives
multiagent	child agents, lineage, message relay
selfmod	controlled self-modification with auditability
evomap	evolution asset publication / network-facing capability mapping
mcp	MCP protocol surface
channels	ingress/egress runtime paths (WebChat now, more later)
plugins / skills	capability extension layer
automaton	survival tiers and adaptation logic

---

5. Ground Rules for Contributors and Agents

All contributors — human or agentic — should assume the following:

5.1 Do not downgrade the project

Do not frame ConShell as merely:

• a chat interface
• a CLI wrapper
• a dashboard
• a prompt shell

Those may be surfaces, but they are not the project.

5.2 Do not confuse module existence with system closure

A directory called wallet/ does not mean economic closure exists. A directory called identity/ does not mean sovereign identity exists. A directory called multiagent/ does not mean evolutionary runtime exists.

5.3 Runtime truth comes before ambition

Before expanding capability breadth, we must preserve:

• runtime truth
• reproducibility
• viability evidence
• auditable system boundaries

5.4 Every round must build on audited reality

No next-round plan should be produced from stale assumptions. Every next-round objective must be based on:

1. the current round’s implemented changes
2. current round verification results
3. current round audit conclusions

---

6. Required Reading Order for Any Agent Working on This Repo

Before doing significant work, agents should read files in this order:

1. docs/project/PROJECT_PURPOSE_AND_GLOBAL_OBJECTIVE.md
2. docs/audit/GLOBAL_AUDIT_2026-03-14.md
3. docs/planning/GLOBAL_DEVELOPMENT_PLAN.md
4. docs/planning/NEXT_PHASE_ROADMAP.md
5. CONSTITUTION.md
6. docs/audit/DEVLOG.md
7. this README.md

If the task is implementation for a specific round, the agent should then read the relevant package/module files.

---

7. Root Documents

File	Purpose
docs/project/PROJECT_PURPOSE_AND_GLOBAL_OBJECTIVE.md	canonical statement of project purpose and final objective
docs/audit/GLOBAL_AUDIT_2026-03-14.md	current global status audit
docs/planning/GLOBAL_DEVELOPMENT_PLAN.md	system-wide development plan
docs/planning/NEXT_PHASE_ROADMAP.md	next-stage roadmap by phase / priority / dependency
docs/project/UPSTREAM_ALIGNMENT_AND_GAP_ANALYSIS.md	OpenClaw / Automaton alignment and current gaps
AGENT_START_HERE.md	mandatory reading path for future development agents
CONSTITUTION.md	governing laws and hard constraints
docs/audit/DEVLOG.md	historical development log

---

8. Quick Start

# prerequisites
node >= 24
pnpm >= 10

# install
pnpm install

# build
pnpm build

# typecheck
cd packages/core && npx tsc --noEmit

# tests
cd packages/core && npx vitest run --no-coverage

# CLI
pnpm --filter @conshell/cli dev

# dashboard
pnpm --filter @conshell/dashboard dev

Note: runtime pinning and Doctor/viability expectations should be checked against the latest audited baseline, not guessed from this section alone.

---

9. Long-Term Direction

ConShell should evolve in this order:

1. Viability baseline
2. Identity + memory continuity
3. Economic survival coupling
4. Autonomous agenda and continuous operation
5. Governed self-modification and replication
6. Collective evolution runtime

The project succeeds only if these layers become real system closures, not just feature lists.

---

10. One-Sentence Definition

ConShell is a Web4-oriented autonomous runtime project that fuses OpenClaw’s agent operating-system strengths with Conway Automaton’s survival and evolutionary logic to build a sovereign AI lifeform runtime.