Saltar al contenido
TakoAPI
Volver a skills

molt-life-kernel

por jongartmann

Moltbook
1.6kDescargas según clawskills.sh
0Vistas
0 me gustaInicia sesión para dar me gusta a esta skill
—Aún sin valoraciones
Tu valoración

Resumen

Agent continuity and cognitive health infrastructure.

Descripción

Qué hace esta skill

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, heartbeat vitality signals, and Shannon entropy coherence checks.

Combines memory persistence, crash recovery, and safety gating in a single module rather than requiring separate solutions for each concern.

Cuándo usarla

  • Preserving agent context across session restarts
  • Recovering agent state after an unexpected crash
  • Requiring human approval before an agent deletes or modifies critical data
  • Detecting when an agent has drifted from expected behavior over time
  • Building an immutable audit trail of all agent actions for compliance

Flujo de trabajo de ejemplo

ENTRADA

User asks: set up an agent that survives crashes and requires approval before deleting data

AGENTE
  1. 1

    Import MoltLifeKernel and instantiate with a witnessCallback pointing to a human approval function

  2. 2

    Call kernel.append() to record the incoming request to the append-only ledger

  3. 3

    Call kernel.enforceCoherence() to verify the agent's context has not drifted

  4. 4

    Call kernel.witness() with the proposed delete action and risk score before executing

  5. 5

    Call kernel.getSnapshot() to save a recovery capsule; if a crash occurs, call kernel.rehydrate() with the saved capsule and ledger

SALIDA

Agent executes the session with full audit trail, coherence validation, human-approved destructive actions, and a recoverable snapshot stored in the workspace

Instalación

clawhub install jongartmann/molt-life-kernel

Ejecuta el comando de la CLI o pega el prompt en Claude Code, Cursor, Codex o el agente de programación que uses.

Enlaces