rdc

RDC Skills Library — Generic Version

This directory contains a generic, project-agnostic version of the RDC (Research, Design, Code) skill suite. All skills have been extracted from a reference implementation and genericized for use in any monorepo project.

Install

Option A — Plugin marketplace (recommended for new installs)

From any Claude Code session:

/plugin marketplace add LIFEAI/rdc-skills
/plugin install rdc-skills

All user-invocable skills become available as slash commands. Agent playbooks in guides/agents/ are dispatched internally by rdc:build and are not user-invocable.

Option B — Directory plugin (LIFEAI dev machines)

If you have the source repo locally, point the plugin directly at the directory in ~/.claude/settings.json. Skills load live from disk — no install step needed, changes take effect on next session start:

"extraKnownMarketplaces": {
  "rdc-skills": {
    "source": { "source": "directory", "path": "C:\\Dev\\rdc-skills" }
  }
},
"enabledPlugins": {
  "rdc-skills@rdc-skills": true
}

⚠️ DO NOT mix install methods — duplicates will occur

If rdc-skills@rdc-skills is enabled in your plugin settings AND you have manually installed files at ~/.claude/skills/user/rdc-*.md, every skill loads twice (or more if stale plugin cache versions exist). Symptoms: each skill appears 2–3× in the skill list.

Fix:

# Remove manual install
rm ~/.claude/skills/user/rdc-*.md

# Remove stale plugin cache (keep only the current version)
rm -rf ~/.claude/plugins/cache/rdc-skills/rdc-skills/0.7.*

Use one method only: plugin (Option A or B) OR manual install — never both.

Installer profiles

scripts/install-rdc-skills.js supports profiles:

node scripts/install-rdc-skills.js --profile core
node scripts/install-rdc-skills.js --profile lifeai

• core is safe for a clean machine with only rdc-skills installed. It wires portable hooks only: RDC output-contract enforcement, foreground-process policy, commit-message hygiene, and non-blocking work logging.
• lifeai is for the LIFEAI/regen-root workstation. It additionally wires the regen-root cwd lock, clauth/Supabase work-item gates, overnight queue guard, and LIFEAI session hooks.
• auto is the default. It chooses lifeai only when a sibling regen-root project is detected; otherwise it chooses core.

To add managed startup instructions to a project:

node scripts/install-rdc-skills.js --project-root /path/to/project --write-startup-blocks

This writes .rdc/guides/rdc-skills-startup.md and managed RDC sections in CLAUDE.md and AGENTS.md.

Legacy install scripts (deprecated since v0.6.0)

scripts/install.sh and scripts/install.ps1 copy files into ~/.claude/skills/user/. These exist only for environments that cannot use the plugin system. Running them alongside the plugin creates duplicates. Do not use them if you have the plugin installed.

Path Variables

Throughout the skills, the following substitutions have been made to make them portable:

Original	Generic	Your Project
/full/path/to/monorepo/	{PROJECT_ROOT}/	Your monorepo root
docs/guides/agent-bootstrap.md	{PROJECT_ROOT}/docs/guides/agent-bootstrap.md	Your agent guide
develop branch	{development-branch}	Your development branch
main branch	{production-branch}	Your production branch
Specific apps	your-app-name	Your app names

Skills Overview

23 skills organized into 5 categories:

Orchestration Skills (workflow drivers)

• rdc:build, rdc:plan, rdc:preplan, rdc:review, rdc:overnight, rdc:fixit, rdc:status, rdc:report

Agent Guides (dispatched by rdc:build, not user-invocable)

Located in guides/agents/ — plain markdown playbooks spawned as sub-agents:

• frontend, backend, data, design, infrastructure, content, cs2, viz, setup, verify

Bridge Skills (Planning → CLI)

• rdc:handoff, rdc:prototype, rdc:workitems

Utility Skills

• rdc:collab

Design Skills

• rdc:design — RDC/Studio design, tokens, palettes, themes, and Rampa CLI-assisted color systems

Reference

• rdc:help

Key Features

TDD Enforcement (Built In)

• rdc:build includes mandatory post-wave test gate via vitest
• rdc:review reports test coverage delta
• All agents write tests FIRST: red → implement → green
• Packages with implementation but 0 new tests are flagged

Core Rules (Non-Negotiable)

1. NEVER run pnpm build — crashes the system. Use npx tsc --noEmit + vitest only
2. NEVER let agents overlap on the same files
3. ALWAYS create work items BEFORE starting work
4. ALWAYS read guide files before agent dispatch
5. ALWAYS update work items in real time
6. Tests FIRST, then implementation — red → green
7. Push after every logical block

How to Adapt for Your Project