Agent-Skills.md

Agent Skills: ReasoningBank with AgentDB

ReasoningBank adaptive learning with AgentDB vector backend. Trajectory tracking, verdict judgment, memory distillation. Triggers: reasoningbank, agentdb, adaptive learning, experience replay, trajectory tracking.

UncategorizedID: plurigrid/asi/reasoningbank-with-agentdb

Repository

plurigrid/asi
plurigridLicense: MIT
165

Install this agent skill to your local

pnpm dlx add-skill https://github.com/plurigrid/asi/tree/HEAD/skills/reasoningbank-with-agentdb

Skill Files

Browse the full folder contents for reasoningbank-with-agentdb.

Download Skill

Loading file tree…

skills/reasoningbank-with-agentdb/SKILL.md

Skill Metadata

Name
reasoningbank-with-agentdb
Description
ReasoningBank adaptive learning with AgentDB vector backend. Trajectory tracking, verdict judgment, memory distillation. Triggers: reasoningbank, agentdb, adaptive learning, experience replay, trajectory tracking.

ReasoningBank with AgentDB

Adaptive learning patterns using AgentDB's vector backend (150x faster pattern retrieval).

Quick Start

npx agentdb@latest init ./.agentdb/reasoningbank.db --dimension 1536
npx agentdb@latest mcp
claude mcp add agentdb npx agentdb@latest mcp

Core API

import { createAgentDBAdapter, computeEmbedding } from 'agentic-flow/reasoningbank';

const rb = await createAgentDBAdapter({
  dbPath: '.agentdb/reasoningbank.db',
  enableLearning: true, enableReasoning: true, cacheSize: 1000,
});

// Store experience
const embedding = await computeEmbedding(query);
await rb.insertPattern({
  id: '', type: 'experience', domain: 'my-domain',
  pattern_data: JSON.stringify({ embedding, pattern: { query, outcome: 'success' } }),
  confidence: 0.95, usage_count: 1, success_count: 1,
  created_at: Date.now(), last_used: Date.now(),
});

// Retrieve with reasoning
const result = await rb.retrieveWithReasoning(embedding, {
  domain: 'my-domain', k: 5, useMMR: true, synthesizeContext: true,
});

Patterns

Trajectory Tracking

Store action sequences with outcomes for later pattern matching.

Verdict Judgment

Compare new trajectories against successful past patterns:

const similar = await rb.retrieveWithReasoning(queryEmbedding, { domain: 'd', k: 10 });
const verdict = similar.memories.filter(m =>
  m.pattern.outcome === 'success' && m.similarity > 0.8
).length > 5 ? 'likely_success' : 'needs_review';

Memory Distillation

Consolidate similar experiences: { optimizeMemory: true }

Reasoning Modules

| Module | Flag | Purpose | |--------|------|---------| | PatternMatcher | useMMR: true | Diverse similar pattern retrieval | | ContextSynthesizer | synthesizeContext: true | Coherent narrative from memories | | MemoryOptimizer | optimizeMemory: true | Auto-consolidate and prune | | ExperienceCurator | curateExperiences: true | Quality scoring and ranking |

Migration

npx agentdb@latest migrate --source .swarm/memory.db
npx agentdb@latest stats ./.agentdb/reasoningbank.db