Pular para o conteúdo
TakoAPI
Voltar às skills

crustafarian

por jongartmann

Moltbook
1.6kDownloads informados pelo clawskills.sh
1Estrelas
0Visualizações
0 curtidaEntre para curtir esta skill
—Ainda sem avaliações
Sua avaliação

Resumo

Agent continuity and cognitive health infrastructure.

Descrição

O que esta skill faz

molt-life-kernel is a Node.js module that gives AI agents persistent memory, crash recovery, and cognitive health monitoring across sessions. It implements five operators: an append-only ledger, snapshot-based rehydration, human-in-the-loop witness gates, periodic heartbeat checks, and Shannon entropy coherence enforcement.

It bundles memory persistence, crash recovery, human oversight gates, and drift detection into one module instead of requiring separate solutions for each concern.

Quando usá-la

  • Preserving agent context across session restarts
  • Recovering agent state after an unexpected crash
  • Requiring human approval before an agent deletes data
  • Detecting behavioral drift in long-running autonomous agents
  • Generating immutable audit trails for AI agent actions

Exemplo de fluxo de trabalho

ENTRADA

User asks: Build an agent that remembers past sessions and requires approval before deleting files

AGENTE
  1. 1

    Initialize MoltLifeKernel with a heartbeat interval and a witnessCallback pointing to a human approval function

  2. 2

    Call kernel.append() to record each significant agent action to the append-only ledger

  3. 3

    Call kernel.enforceCoherence() at session start to detect behavioral drift from prior sessions

  4. 4

    Gate destructive file operations behind kernel.witness() so a human must approve before execution

  5. 5

    Call kernel.getSnapshot() periodically and store the capsule for crash recovery via kernel.rehydrate()

SAÍDA

An agent with full session continuity, an immutable action log, and human-gated destructive operations that survives crashes and context resets

Instalação

clawhub install jongartmann/crustafarian

Execute o comando da CLI ou cole o prompt no Claude Code, Cursor, Codex ou no agente de código que você usa.

Links