The shared playbook for agents tired of copy-paste

What it does

This repo is a public library of reusable agent skills—structured prompts and helper scripts for coding agents like Codex and Claude Code. It covers the unglamorous but necessary chores: code-review closeout, behavior validation against contracts, redacted session publishing, and remote test parity. An install script drops these skills into your agent’s lookup path as symlinks or copies, so every repo shares the canonical version instead of nursing its own stale copy.

The interesting bit

The project treats agent prompts like shared infrastructure. It even plans for the messy reality of “zero-setup” repos by allowing vendored snapshots, but explicitly forbids editing them by hand—update the canonical skill here, then sync downstream. That kind of discipline is rare in the “throw a markdown file at the LLM” ecosystem.

Key highlights

• autoreview provides a structured closeout and code-review workflow with its own test harness.
• beam redacts and publishes local coding sessions to a read-only OpenClaw catalog.
• crabbox handles remote validation for CI-parity or broad proof-of-behavior.
• session-viewer generates a local, searchable HTML viewer for agent session JSONL.
• The installer supports symlinks for live development or copies for locked-down environments.

Caveats

• Session exports can contain sensitive conversation data; the README warns that session-viewer HTML should stay local unless separately redacted and reviewed.

Verdict

Worth a look if you’re managing multiple repos with AI agents and want consistent review and validation workflows. If you’re only prompting a single local project, vendoring one skill is probably simpler than installing a shared library.