boshu2/agentops

What it solves

AgentOps provides a DevOps-like discipline for AI coding agents, making software delivery repeatable and reliable. It solves the problem of inconsistent agentic coding sessions by providing "paved paths"—structured workflows (skills) and memory systems—that guide agents from initial intent to verified results, ensuring that work is documented and context is preserved across sessions.

How it works

AgentOps uses a kit of "skills" (Markdown-based instructions) that agents load to follow specific operational loops. These skills combine with an optional ao CLI and existing engineering tools to implement a cycle of shaping, building, and judging:

• The Operational Loop: Uses plan to define expected behavior (via BDD scenarios), implement to build the change, and validate to provide independent judgment in a separate session to avoid author bias.
• Goal Management: For larger tasks, it uses an interview skill to settle outcomes and craft-goal to define the work. It integrates with Beads, a work graph tool that preserves decisions, results, and project memories outside of a chat transcript.
• Memory System: The memory skill curates findings into reviewed .context/ pages, allowing future agents to recall lessons and sources without starting over.
• Extensibility: Each skill is a standalone Markdown file, allowing users to customize, delete, or add their own instructions to create a personalized operating model.

Who it’s for

Developers using AI coding agents such as Claude Code, Codex, Cursor, OpenCode, Gemini CLI, Pi, OpenClaw, and Grok Bot who want to move beyond simple chat-based coding to a structured, repeatable engineering process.

Highlights

• Behavior-Driven Development (BDD): Integrates BDD scenarios to clarify intent before implementation.
• Independent Validation: A dedicated validate skill ensures a fresh agent in a new session judges the work against the original intent.
• Longevous Memory: Uses Beads to maintain a searchable work graph and persistent project facts across different agent sessions.
• Agent-Agnostic: Works across a wide variety of coding agents via plugins or npx installation.
• Customizable Skill Kit: Provides a library of 29 starting skills covering everything from research and refactoring to multi-model "councils" for debate and auditing.