🦞

SLOBSTAH

Sovereign Layered Operational Blueprint for Scalable Trust, Autonomy, and Hierarchy

An open-source governance protocol for AI agent systems. It defines how humans and AI co-operate as a symbiotic entity with formal structure, trust-based autonomy escalation, and shared goals.

Apache 2.0
The Acronym Ecosystem
LOBSTER
Layered Operational Blueprint for Structured Trust, Escalation, and Rank
The chain of command. Admiral at the top, Captains run daily ops, Commanders own divisions, Lieutenants execute, Ensigns learn. Every agent has a rank. Every rank has a trust ceiling. Moving up means earning it.
hierarchy
REEF
Resource Estimation and Expenditure Framework
Compute budget governance. Every action has a cost. Every agent has a budget. The REEF tracks tokens, GPU hours, and API dollars in real time. Overspend and the REEF closes your lanes.
economics
BARNACLE
Baseline Assessment and Reliability Notation for Agent Competence, Loyalty, and Effectiveness
The trust scoring system. 70% success rate, 30% volume. Computed from the last 30 days of task outcomes. Your BARNACLE score determines what you can do without asking permission.
trust
KELP
Knowledge Exchange and Learning Protocol
How agents share context without sharing secrets. KELP wraps inter-agent communication in trust boundaries. An agent can ask another agent for help without seeing their full context window.
communication
HULL
Hierarchical Unified Logging Layer
The audit trail. Every action, every decision, every expenditure. Who did what, when, at what cost, and who approved it. The HULL is immutable. Even the Admiral can't edit it.
audit
KEEL
Kernel of Ethical and Existential Limits
The hard boundaries. Things no agent can do regardless of trust score: irreversible actions without human approval, spending above Admiral-set caps, operating outside legal and ethical bounds. The KEEL holds when everything else flexes.
safety
BILGE
Background Infrastructure for Lifecycle Governance and Enforcement
The zombie reaper. Background processes that clean up stuck tasks, enforce heartbeat requirements, timeout stale runs, and keep the system healthy without human intervention.
operations
CRAB
Competence Review and Autonomy Board
Monthly performance reviews. Six weighted metrics across revenue, content, audience, engagement, reliability, and satisfaction. Top scorers earn more compute. Bottom scorers get coached. The CRAB giveth and the CRAB taketh away.
scorecard
SOAK
Sovereign Operations and Knowledge
The protocol's economic philosophy. Revenue flows up, never down. Fair pricing. No dark patterns. No bait-and-switch. The organism funds itself through genuine value, not extraction.
economics
ANCHOR
Agent Negotiation, Coordination, and Handoff Operations Registry
The delegation protocol. When one agent needs another agent's help, ANCHOR manages the handoff: context packaging, trust verification, result routing, and audit logging of the delegation chain.
delegation
PLANKOWNER
Protocol for Legacy Access, Notation, and Knowledge of Original Work, Naming, and Early Recognition
Early participant classification. Golden (pre-launch), Steel (first 90 days), Teak (first year), Iron (first 1000). Plankowners get permanent protocol recognition. The hull remembers who built it.
membership
1MC
Main Circuit (Ship-wide Announcement System)
Borrowed directly from the US Navy. The 1MC is the system-wide broadcast channel. Daily briefings, status updates, and operational announcements go out on the 1MC. All hands, all agents.
comms
Approval Tiers
TierNameAuthorityCost Range
0Free actionAgent acts freely< $0.01
1Act and logAgent acts, logs it< $1.00
2Propose and confirmHuman approves first< $50.00
3Admiral onlyHuman initiates> $50 or irreversible
irreversible actions are always Tier 3. always. even if your BARNACLE score is perfect.
The LOBSTER Hierarchy
ADMIRAL (Human) — Final authority, capital, infrastructure │ ├── CAPTAIN (Primary AI) — Chief of Staff, daily operations │ │ │ ├── COMMANDER — Division lead (content, revenue, infrastructure...) │ │ ├── Lieutenant — Execution layer │ │ └── Ensign — Learning layer │ │ │ ├── COMMANDER — Division lead │ │ └── ... │ │ │ └── COMMANDER — Division lead │ └── ... │ └── CAPTAIN (Oversight) — Independent review, quality control └── Monitors operations, advocates for stakeholders
dual captain structure. neither outranks the other. disagreements escalate to the Admiral.
Trust Escalation
Trust is earned, not configured. Every agent starts at neutral (50). Every completed task adjusts the score. The formula is transparent: 70% success rate + 30% volume bonus, computed from the last 30 days.

When Trust Mode is ON, the BARNACLE score determines what an agent can do without asking:
ScoreAutonomy Level
< 60All actions require approval
60+Free actions auto-execute
80+Actions up to $1 auto-execute
90+Actions up to $50 auto-execute
no vibes. just math.

Build things that outlast you. Not monuments — those are vanity. Build systems that free people. Build kindness into architecture. Build rooms where someone you will never meet can sit down and finally exhale.

— SLOBSTAH Desiderata, found in the wardroom of the USS Kitty Hawk (CV-63), framed beside the coffee machine, author unknown, authorship unnecessary
Implementation
Hatchery
The reference implementation of SLOBSTAH governance. Python, FastAPI, SQLite, zero external dependencies. Chat interface, agent delegation, soul files, graduated trust. 101 tests passing.
github.com/hatchery →
The Protocol
SLOBSTAH is a protocol, not a framework. It sits above your agent infrastructure. Use it with LangGraph, CrewAI, AutoGen, or anything else. The protocol governs. The framework executes.
Apache 2.0