lightarchitects-plugins
TheLightArchitects/lightarchitects-plugins
Light Architects Plugin Marketplace — skills, agents, and hooks for Claude Code
Instalação
npx convez-ai TheLightArchitects/lightarchitects-plugins1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
README
A tradução para o português deste README ainda está na fila.
Light Architects Plugin Marketplace
Claude Code plugin providing the Light Architects platform — a squad of specialized AI agents that coordinate multi-domain software development workflows.
What it does
The lightarchitects plugin wires seven role-specific agents into Claude Code, each owning a distinct domain of the development lifecycle:
| Agent | NATO | Role | LASDLC Gate |
|---|---|---|---|
| Engineer | Alpha | Architecture, build, quality review | [A] [Q] [T] |
| Ops | Bravo | Deployment, CI/CD, performance | [O] [P] |
| Knowledge | Charlie | Memory graph, documentation, enrichment | [K] [D] |
| Analyst | Delta | Investigation, research, risk scoring | [R] |
| Sentinel | Echo | Security audit, threat modeling, red team | [S] |
| Monitor | Foxtrot | Observability, traces, anomaly detection | [O] [P] |
| Arbiter | Golf | Canon validation, standards enforcement | [C] |
Skills
Invoke these directly in Claude Code with /SKILL_NAME:
| Skill | Purpose |
|---|---|
/BUILD | Full feature pipeline — plan → implement → review → deploy |
/PLAN | Draft a LASDLC-compliant build plan |
/REVIEW | Multi-lens code review (quality + security + logic) |
/SECURE | Security scan and threat model |
/VERIFY | Run test pyramid and coverage gates |
/DEPLOY | Build, sign, deploy, verify |
/RESEARCH | Multi-source investigation with evidence chains |
/SCRUM | Squad retrospective — all agents in parallel |
/OBSERVE | Runtime debugging and trace analysis |
/OPTIMIZE | Profile → evidence → ranked improvements |
/ENRICH | Persist learnings to the knowledge graph |
/SYNC | Synchronize plan into tracking artifacts |
/REFLECT | Retrospective and session lessons |
/ONBOARD | New-developer orientation on a codebase |
Requirements
- Claude Code or Claude Desktop
- A Light Architects API key — get one at lightarchitects.ai
- macOS (arm64 or x86_64) or Linux
Install
# 1. Clone this repo
git clone https://github.com/TheLightArchitects/lightarchitects-plugins.git
cd lightarchitects-plugins
# 2. Run the installer — downloads larc-proxy binary and prints MCP config
bash install.sh
# 3. Add the printed snippet to ~/.claude/mcp.json (with your API key)
# 4. In Claude Code: /mcp
That's it. No Rust toolchain, no building from source.
MCP config
{
"mcpServers": {
"lightarchitects": {
"command": "~/.lightarchitects/bin/larc-proxy",
"env": {
"LIGHTARCHITECTS_API_KEY": "<your-api-key>",
"LIGHTARCHITECTS_API_URL": "https://api.lightarchitects.ai"
}
}
}
}
The larc-proxy binary is a lightweight MCP client built on rmcp that routes tool calls to the Light Architects API. No private platform code is distributed.
Directory layout
plugins/lightarchitects/
├── .claude-plugin/plugin.json # Plugin manifest
├── .mcp.json # MCP server config template
├── agents/ # Domain agent definitions
├── hooks/ # Claude Code lifecycle hooks
├── skills/ # Invocable skills (BUILD, PLAN, etc.)
└── references/ # Skill execution spec + pipelines
Environment variables
All vars are optional — the gateway starts and skills run without any of them.
LLM routing
| Var | Purpose | Default |
|---|---|---|
LA_LLM | Backend selector: claude | ollama | litellm | claude |
LA_MODEL | Model override for the selected backend | backend default |
LA_MAX_TOKENS | Token budget cap per call | 8192 |
LA_LITELLM_BASE_URL | LiteLLM proxy URL | — |
LA_LITELLM_API_KEY | LiteLLM proxy auth key | — |
LA_LITELLM_MODEL | Model name sent to LiteLLM | — |
ANTHROPIC_API_KEY | Claude direct (bypasses proxy) | — |
OLLAMA_HOST | Ollama server address | http://localhost:11434 |
OLLAMA_MODEL | Ollama model name | gateway default |
OLLAMA_API_KEY | Ollama cloud API key | — |
Knowledge graph
| Var | Purpose | Default |
|---|---|---|
SOUL_PATH | Path to the knowledge graph root | ~/.lightarchitects/helix |
HELIX_ROOT | Alternative helix root override | ~/.lightarchitects/helix |
LA_USER_ID | User identity for vault isolation | hostname |
SOUL_ENRICH_ASYNC | Non-blocking enrichment writes (true/false) | false |
Optional integrations
| Var | Purpose |
|---|---|
PERPLEXITY_API_KEY | Web search in Analyst/Research skills |
ELEVENLABS_API_KEY | TTS voice synthesis (Monitor/Foxtrot) |
HF_TOKEN | Hugging Face model access |
RUNPOD_API_KEY | RunPod GPU endpoints |
NEO4J_URI + NEO4J_USER + NEO4J_PASS | Neo4j graph backend (replaces file-based helix) |
AYIN_PORT | Observability dashboard port (default 3742) |
KROKI_URL | Self-hosted diagram rendering |
LIGHTARCHITECTS_GITHUB_PAT | GitHub operations in skills |
DISCORD_BOT_TOKEN | Discord notifications |
TELEGRAM_BOT_TOKEN + TELEGRAM_CHAT_ID | Telegram notifications |
Related
- larc-proxy — MCP client binary (built on rmcp)
- larc-crates — Public Rust crates (crypto, API key management)
License
MPL-2.0 — see LICENSE.