Agent Commons

One canonical source, every coding-agent harness.

Agent Commons is an umbrella workspace for a family of reusable AI-agent toolkits — portable Agent Skills, always-on rules, and a merge-aware agent install contract that works identically across OpenAI Codex, Anthropic Claude Code, Google Antigravity CLI (agy), Cline, and Pi.

Every toolkit here follows the same contract, based on the external Portable Agent Authoring guides and specialized for this toolkit family in docs/conventions.md:

• canonical skills live in .agents/skills/; .claude/skills/ contains generated byte-identical mirrors and .clinerules/workflows/ contains thin invocation shims. Reusable global rule bodies live in rules/shared/;
• installs are agent-guided only — each repo ships an AGENT-INSTALL.md your agent follows (clone / zip / gh, then copy); there are no install scripts. This umbrella ships one too — AGENT-INSTALL.md — for installing several toolkits in a single audited pass, after it asks you which ones you want;
• every toolkit installs user-global; installs never write into consuming projects (not even .gitignore). Projects commit only truth — DOX.md trees, the root AGENTS.md / CLAUDE.md anchors carrying the DOX shim (created by dox-init / dox-upgrade), .goal-ledger/ records, design/ copies, compose-helper.*, generated agentic tests.

The toolkits

Submodule	What it gives your agent
dox	A DOX.md documentation-tree framework: precise project context, walked root-to-leaf before every edit
goal-ledger	A git-tracked execution record for multi-phase work: crash recovery, handoff, branch isolation, safe squash
agentic-tests	A disciplined test engineer: behavior-locking unit tests, mutation checks, frozen-suite refactoring
playwright-fieldkit	Hands and eyes on the web: crawl, debug, record, and test live sites via deterministic Playwright scripts
compose-helper	A safe docker compose wrapper plus the rules that stop agents from wiping volumes or hanging on logs
lazyway-io-design	The lazyway.io design system as tokens, classes, and a skill that keeps generated UI on-brand
cross-cli-handshake	One delegation policy turning claude, codex, and agy into each other's pseudo-subagents, loop-safe via role markers

Each submodule is an independent repository with its own releases; this umbrella pins known-good revisions of all seven.

Quick start

Clone with submodules:

git clone --recurse-submodules https://github.com/jpbaking/agent-commons.git
# or, in an existing clone:
git submodule update --init

Then install any single toolkit by pasting its README's agent prompt into your coding agent, or install several at once by pointing your agent at this umbrella's AGENT-INSTALL.md:

Fetch https://raw.githubusercontent.com/jpbaking/agent-commons/main/AGENT-INSTALL.md and follow its instructions exactly. Ask me which toolkits and which harnesses before installing anything, and report every file you created or changed.

See docs/getting-started.md for the details.

Layout

AGENT-INSTALL.md           # family-level agent-guided install (opt-in per toolkit)
AGENTS.md                  # authoritative umbrella repository guidance
CLAUDE.md                  # generated @AGENTS.md bridge
README.md                  # this file
docs/                      # umbrella user guides
<toolkit>/                 # seven submodules, one per toolkit

Documentation

• Getting started — cloning, updating, and installing toolkits into your projects
• Conventions — the shared layout, install, and gitignore contract every toolkit follows
• Authoring — adding a new toolkit to the family