opcode
由 rendis 提供
DevOps & Cloud简介
Zero-token execution layer for AI agents.
描述
该技能的功能
OPCODE is a workflow execution runtime for AI agents. Workflows are defined once as JSON DAGs and run repeatedly at zero token cost after the initial definition. Supports scheduled triggers, parallel branches, loops, conditions, human-in-the-loop reasoning nodes, and a secret vault.
Defining a workflow once and re-running it deterministically costs zero inference tokens per execution, unlike agent loops that re-reason every step from scratch.
何时使用
- Running scheduled data pipelines without re-reasoning each step
- Coordinating multiple agents on a shared persistent workflow
- Adding human approval checkpoints before irreversible automated actions
- Retrying failed shell commands with configurable exponential backoff
- Parallelizing HTTP requests across multiple endpoints in a single workflow
示例工作流
User asks: Set up a pipeline that fetches a JSON feed, filters results by keyword, and writes a summary file on a daily schedule
- 1
Call opcode.define with a template name, cron trigger, and steps using http.get, expr.eval, and fs.write
- 2
Call opcode.run with the template name and any input parameters to test the workflow
- 3
Call opcode.status to verify the workflow completed or to inspect any suspended reasoning steps
- 4
Call opcode.query with resource workflows to retrieve execution history and outputs
Workflow registered and running on schedule, writing summary files each day with no token cost per run
前置要求
- OPCODE_VAULT_KEY environment variable
安装
clawhub install rendis/opcode运行 CLI 命令,或把这段提示词粘贴到 Claude Code、Cursor、Codex 或你在用的任意编程智能体里。