🦞 Rock Lobster

Autonomous operations agent. Self-coordinating. Privacy-preserving. Built on Praxis.

What Is This?

Rock Lobster is the next generation of OpenClaw's lobster plugin — reimagined as a standalone autonomous ops agent built on the Plures stack. It doesn't need a gateway. It doesn't need a cloud. It coordinates work across repos, teams, and machines using Praxis rules and P2P sync.

Quick Start

# Install dependencies
npm install

# Build the project
npm run build

# Run tests
npm test

# Run the example (from the compiled output)
node dist/examples/memory-usage.js

Why "Rock Lobster"?

Because lobster was a scheduler plugin. Rock Lobster is a self-aware operations agent that:

• 📋 Plans — Uses Praxis rules to decide what work matters, when, and who does it
• 🧠 Remembers — SLM captures decisions, patterns, and lessons learned
• 🔄 Coordinates — PluresDB sync keeps multiple agents in lockstep
• 🔒 Respects boundaries — OASIS PPE governs what data flows where
• 🤖 Delegates — Assigns work to Copilot, subagents, or human reviewers

Architecture

┌─────────────────────────────────────────┐
│           Rock Lobster Agent            │
│                                         │
│  ┌───────────┐  ┌───────────────────┐   │
│  │  Praxis   │  │  Work Rotation    │   │
│  │  Rules    │──│  Engine           │   │
│  │           │  │  (repos.yaml →    │   │
│  │ schedule  │  │   daily tasks)    │   │
│  │ delegate  │  └───────────────────┘   │
│  │ escalate  │                          │
│  │ learn     │  ┌───────────────────┐   │
│  └───────────┘  │  SLM Memory       │   │
│                 │  (patterns, fixes, │   │
│  ┌───────────┐  │   decisions)      │   │
│  │  PluresDB │  └───────────────────┘   │
│  │  (state,  │                          │
│  │   sync,   │  ┌───────────────────┐   │
│  │   tasks)  │  │  Integrations     │   │
│  └───────────┘  │  GitHub · Email   │   │
│                 │  Slack · Calendar  │   │
│                 └───────────────────┘   │
└─────────────────────────────────────────┘
         ↕ Hyperswarm P2P ↕
┌─────────────────────────────────────────┐
│    Other Rock Lobster instances          │
│    (work machine, home, CI runner)       │
└─────────────────────────────────────────┘

Core Capabilities

Work Rotation Engine

Replaces tier-based repo handling with equal rotation:

• Every repo gets scheduled time (design → issues → strategy → roadmap)
• Dormant repos flagged for archive review
• Configurable via repos.yaml registry

Praxis Rule Modules

# Schedule rules
WHEN day_of_week(tuesday) AND repo_status(R, "active")
  AND assigned_day(R, tuesday)
THEN work_mode(R, "design") PRIORITY 1

# Delegation rules  
WHEN issue(I) WITH label(I, "copilot") AND NOT assigned(I, "Copilot")
THEN assign(I, "Copilot") AND comment(I, "Assigned to Copilot coding agent")

# Escalation rules
WHEN pr(P) WITH age(P, >48h) AND status(P, "failing")
THEN escalate(P, "human") AND notify("Stale PR needs attention")

# Learning rules
WHEN error_pattern(E) WITH occurrences(E, >3)
THEN create_issue("recurring error", E) AND store_memory(E, "error-fix")

Privacy-Aware Coordination

• Work machine agent sees only work repos (PPE enforced)
• Home agent handles personal repos + cross-cutting ops
• CI agent has read-only access to all repos for monitoring
• No agent sees more than its trust level allows

SLM Integration (Phase 2 - Implemented)

Rock Lobster now includes comprehensive memory integration for operational intelligence:

Four Memory Categories:

1. Error-Fix: Captures error patterns and their fixes from CI failures, PR issues
2. Repository Context: Stores build systems, dependencies, quirks, and workflows
3. Human Preferences: Learns from overrides, code style, and communication preferences
4. Temporal Patterns: Tracks when operations succeed/fail (time-based patterns)

Key Features:

• ✅ Memory capture for all operational events
• ✅ Smart querying and similarity search
• ✅ Delegation decision support with context
• ✅ Learning from delegation outcomes
• ✅ Relevance scoring and memory expiration
• ✅ Comprehensive test coverage

Usage Example:

import { createMemorySystem } from './src/memory';

const memory = createMemorySystem();

// Capture an error-fix pattern
await memory.errorCapture.captureCIFailure({
  repository: 'plures/praxis',
  workflowName: 'CI',
  errorMessage: 'Type error in rules.ts',
  fixDescription: 'Added type annotation',
});

// Get delegation context
const context = await memory.delegationHelper.getTaskContext({
  repository: 'plures/praxis',
  taskType: 'bug_fix',
  description: 'Fix type error',
});

See src/memory/README.md for detailed documentation.

Relationship to OpenClaw Lobster

| Feature | Lobster (OpenClaw) | Rock Lobster | |---------|-------------------|--------------| | Scheduling | Cron expressions | Praxis temporal rules | | State | Ephemeral (session) | Persistent (PluresDB) | | Memory | None | SLM auto-capture | | Multi-machine | Not supported | P2P sync | | Privacy | None | OASIS PPE | | Delegation | @copilot comments | Proper assignment + tracking | | Learning | None | Error-fix pairs, pattern recognition | | Configuration | JSON config | repos.yaml + Praxis facts |

Phases

| Phase | Focus | Timeline | |-------|-------|----------| | 1 | Core: Praxis rules + work rotation + GitHub integration | Weeks 1-6 | | 2 | Memory: SLM integration + pattern learning + error tracking | Weeks 7-12 | | 3 | Network: Multi-machine sync + PPE + agent coordination | Weeks 13-18 | | 4 | Intelligence: Predictive scheduling + auto-strategy + reporting | Weeks 19-24 |

License

AGPL-3.0 — See LICENSE