Saltar al contenido
TakoAPI
Volver a skills

agentkeys

por alexandr-belogubov

DevOps & Cloud
462Descargas según clawskills.sh
0Vistas
0 me gustaInicia sesión para dar me gusta a esta skill
—Aún sin valoraciones
Tu valoración

Resumen

Secure credential proxy for AI agents.

Descripción

Qué hace esta skill

AgentKeys is a credential proxy for AI agents. API calls route through the proxy, which injects real secrets server-side before forwarding to the target. The agent never handles plaintext API keys, tokens, or passwords.

Decoupling secrets from agents means revoking access or rotating keys requires no changes to agent code.

Cuándo usarla

  • Sending emails via Resend without embedding the API key in an agent
  • Revoking a compromised agent's API access without rotating the underlying credential
  • Auditing which agent made calls to a third-party service and when
  • Sharing one credential across multiple agents without duplicating secrets
  • Rotating API keys without redeploying or reconfiguring agents

Flujo de trabajo de ejemplo

ENTRADA

User asks: Send a transactional email via Resend without giving the agent direct API access

AGENTE
  1. 1

    Reads AGENTKEYS_PROXY_URL and AGENTKEYS_API_KEY from environment

  2. 2

    Sends POST to the proxy endpoint with X-Credential-Name: resend and X-Target-Url pointing to the Resend API

  3. 3

    AgentKeys decrypts the real Resend credential server-side

  4. 4

    Proxy injects the credential into headers and forwards the request to Resend

  5. 5

    Returns the API response and logs the call in the audit trail

SALIDA

Email sent via Resend; the real API key was never exposed to the agent

Requisitos

  • AgentKeys account at app.agentkeys.io
  • AGENTKEYS_PROXY_URL env var (value: https://proxy.agentkeys.io)
  • AGENTKEYS_API_KEY (workspace API key from AgentKeys Settings) or AGENTKEYS_PROXY_TOKEN (proxy token assigned to an agent in the AgentKeys dashboard)

Instalación

clawhub install alexandr-belogubov/agentkeys

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

Enlaces