Pular para o conteúdo
TakoAPI
Voltar às skills

agntor

por garinmckayl

Coding Agents & IDEs
386Downloads informados pelo clawskills.sh
2Estrelas
0Visualizações
0 curtidaEntre para curtir esta skill
—Ainda sem avaliações
Sua avaliação

Resumo

Trust layer for AI agents — verify identities, guard prompts, redact PII, and manage x402 escrow payments.

Descrição

O que esta skill faz

Agntor provides a trust and payment layer for AI agent interactions. It verifies agent identities, scans inputs for prompt injection, strips PII from outputs, and manages escrow payments that release only on verified task completion.

Combines identity verification, injection defense, PII redaction, and escrow payments in one layer rather than requiring separate tools for each concern.

Quando usá-la

  • Verifying an external agent's certification before letting it execute tasks
  • Scanning incoming agent instructions for prompt injection before processing
  • Redacting API keys and emails from logs before sharing with third parties
  • Locking payment in escrow before an agent begins a paid job
  • Querying the agent registry to find certified agents within a specific trust score range

Exemplo de fluxo de trabalho

ENTRADA

User asks: hire an external agent to summarize customer support tickets

AGENTE
  1. 1

    Call is_agent_certified to confirm the external agent has a valid, non-expired audit ticket

  2. 2

    Call get_trust_score to verify the agent scores above 30

  3. 3

    Call create_escrow to lock payment funds before the agent starts work

  4. 4

    Call guard_input on the agent's instructions to check for prompt injection

  5. 5

    Call redact_output on the final response to strip any PII before returning it to the user

SAÍDA

Summarized tickets returned to the user with escrow released on verified completion

Requisitos

  • AGNTOR_API_KEY environment variable
  • Agntor account at app.agntor.com

Instalação

clawhub install garinmckayl/agntor

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

Links