Pular para o conteúdo
TakoAPI
Voltar às skills

agent-self-governance

por bowen31337

DevOps & Cloud
826Downloads informados pelo clawskills.sh
0Visualizações
0 curtidaEntre para curtir esta skill
—Ainda sem avaliações
Sua avaliação

Resumo

Self-governance protocol for autonomous agents: WAL (Write-Ahead Log), VBR (Verify Before Reporting), ADL.

Descrição

O que esta skill faz

Five protocols that address common agent failure modes: WAL persists context before LLM compaction, VBR prevents false task-complete claims, ADL tracks persona drift, VFM monitors token cost against task value, and IKL captures infrastructure facts before they leave context. Each protocol has its own Python script, defined triggers, and log format.

Rather than relying on agent memory or ad-hoc notes, each failure mode gets a dedicated scriptable protocol with explicit triggers, commands, and log formats that survive context resets.

Quando usá-la

  • Recovering key decisions after LLM context compaction mid-session
  • Catching false task-complete claims before they reach the user
  • Routing monitoring tasks to budget models instead of burning Opus tokens on simple summarization
  • Logging GPU server specs discovered during SSH setup before they drop out of context
  • Detecting sycophantic drift in agent responses over long sessions

Exemplo de fluxo de trabalho

ENTRADA

User says: 'No, use Podman not Docker'

AGENTE
  1. 1

    Run python3 scripts/wal.py append <agent_id> correction 'Use Podman not Docker' before responding

  2. 2

    Acknowledge the correction to the user

  3. 3

    Update affected configuration or commands to reference Podman

  4. 4

    Run python3 scripts/vbr.py check task123 command 'podman ps' to verify Podman is available

  5. 5

    Log the result with python3 scripts/vbr.py log <agent_id> task123 true 'Podman confirmed available'

SAÍDA

Correction is persisted to WAL before the agent responds, and tooling availability is verified before proceeding

Instalação

clawhub install bowen31337/agent-self-governance

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

Links