Agentic-Dev-Tooling

A portable library of AI development tools: skills, agents, hooks, and guardrails.

Website: www.concordvoice.com | Wiki: Agentic-Dev-Tooling wiki | Security: [email protected]

---

Introduction

Concord Voice is a privacy-first place to talk with your friends, your community, or your team. It gives you voice calls, video, screen sharing, and chat. Your device encrypts each message before it leaves, so our servers store your messages only as scrambled data that we cannot read. You can use our hosted service, or you can run the same software on your own hardware.

Concord Voice uses AI coding assistants every day. This repository shares the tools that keep those assistants safe and consistent. Anyone can use them.

One source tree installs into several AI coding tools. Each tool supports a different subset, and CAPABILITIES.md lists the gaps honestly. Read it before you trust a hook to survive the trip to another platform.

For guides and background, see the Agentic-Dev-Tooling wiki.

---

Technical overview

A portable, attributed library of AI dev-tooling — skills, agents, hooks, guardrail scripts, MCP config templates, and an installer — refined over months of real use running a multi-service product's development lifecycle, and generalized here for any team to adopt. One canonical Claude-native source tree compiles to Claude Code, OpenAI Codex, GitHub Copilot, Google Gemini CLI, and Google Antigravity via install.sh. Every layer's actual, verified per-platform coverage — including the gaps — is documented in CAPABILITIES.md; read it before you rely on a hook or subagent surviving conversion to a non-Claude platform.

Quickstart

Clone the repo, cd into it, then run one install.sh command per platform you use. Every command accepts --project <path> (default: current directory); claude additionally accepts --user for a user-scope install into $HOME/.claude (the other subcommands have no user-scope target and exit 2 if you pass --user to them). Every install is safe to re-run: files that already match are left alone, files you've hand-edited are skipped unless you pass --force.

./install.sh claude      [--project <path>|--user] [--with-fleet] [--force]
# installs skills, agents, hooks, and repo-config docs natively into .claude/;
# merges hook wiring into settings.json non-destructively. --with-fleet adds
# the Claude-desktop-only multi-session supervision pack.

./install.sh codex       [--project <path>] [--force]
# compiles via rulesync into AGENTS.md, .agents/skills/, .codex/hooks.json,
# config.toml.

./install.sh copilot     [--project <path>] [--force]
# compiles via rulesync into .github/copilot-instructions.md and hooks (only
# UserPromptSubmit-shaped hooks survive); .vscode/mcp.json is written as an
# empty scaffold ({"servers": {}}) — populate it by hand (see CAPABILITIES.md).

./install.sh gemini      [--project <path>] [--force]
# writes GEMINI.md directly (no rulesync target exists for Gemini CLI, which
# was sunset for consumer use in 2026-06). Prefer antigravity unless you're
# on Gemini Enterprise Code Assist.

./install.sh antigravity [--project <path>] [--force]
# compiles via rulesync into AGENTS.md, .agents/skills/, .agents/hooks.json
# (partial — one hook event dropped), and an empty MCP scaffold; project-scope
# permissions are dropped entirely (see CAPABILITIES.md).

./install.sh imports
# prints fetch-from-source instructions for third-party marketplace plugins
# (superpowers, mattpocock-skills, pr-review-toolkit, ponytail) and the
# graphify CLI; if the claude CLI is present, it also prints the
# marketplace-add/install commands to run — it does not execute them for you.

./install.sh doctor
# checks prerequisites (git, jq, gh, npx) and prints a summary of key
# capability gaps.

以上为 README 的开头部分。其余内容请阅读原文。 查看完整 README ↗