Agent Playbooks: Portable Engineering Workflows for AI Coding Agents

Research context and background

Agent Playbooks is a collection of portable, tool-agnostic engineering workflows designed specifically for AI coding agents. It provides a single, universal instruction set that eliminates the need to configure separate rules for every AI tool. By standardizing the process, it ensures agents follow repeatable, safe, and verified engineering practices rather than guessing or self-grading their work. The project consists of 38 playbooks covering bug fixing, feature development, code review, security, testing, and autonomous operations. It is built on the AGENTS.md standard, an open cross-vendor format for project-level instructions.

Benefits

Agent Playbooks solves two critical problems in AI-assisted development. First, it addresses fragmented tooling by providing one generic instruction set that acts as the single source of truth for all tools. Different AI coding tools expect instructions in different formats and locations, but Agent Playbooks unifies this process. Second, it improves agent reliability. Left to their own judgment, AI agents often fail in predictable ways such as guessing at ambiguous specifications, grading their own work, or forgetting project conventions. Agent Playbooks enforces a strict process that includes independent verification, hard-blocking destructive commands, and cryptographically signed releases.

The core engineering loop ensures every task follows a unified path. Users simply describe the task naturally, and the system classifies and routes it to the appropriate playbook. This loop includes reproducing bugs before fixing them, writing tests before building features, and independently verifying results. The system also includes safety guardrails that prevent agents from executing destructive shell commands or committing secrets.

Use Cases

Agent Playbooks can be used in various scenarios throughout the software development lifecycle. Developers can use it to fix bugs by reproducing the failure first, then fixing it, and finally re-confirming the fix. Teams can use it to add features by writing a failing test from the spec before any implementation code. Code reviewers can use it to review the real diff, check security separately, and distinguish between confirmed and assumed findings.

The tool is also useful for shipping to production by deciding a staged rollout and rollback plan first, then checking in before widening exposure. It can block destructive commands like force-push or dropping a table automatically. Developers can create custom personas for specific roles such as a Bug Hunter, Feature Builder, or Code Reviewer. The system supports onboarding agents to unfamiliar repositories and auditing projects against quality checklists.

Pricing

Agent Playbooks is free to use, fork, and modify under the MIT License. Installation requires no account, token, or signup. The repo ships the installer and documentation, and the actual playbook text is fetched from a backend on every install to ensure users always have the latest version.

Vibes

There are no public reviews or testimonials available for Agent Playbooks at this time. The project is hosted on GitHub and is in active development.

Additional Information

Agent Playbooks is currently in active development and is maintained by a single developer. The project is built on the AGENTS.md standard, an open cross-vendor format for project-level instructions. The installer checks for prerequisites such as curl, jq, base64, and ssh-keygen. It supports 15 tools including Claude Code, Cursor, Antigravity, GitHub Copilot, and Codex CLI. The project uses cryptographic signing with an Ed25519 key to ensure security and integrity. Every release is signed with a key that resides only on the maintainer's machine. The installer sends only a random local install ID to the backend, and no project data, credentials, or personal information is collected. Users can pin specific versions using version flags or by fetching the installer from a specific release tag.