starlight-architect

Plugin starlight-architect from marketplace frankx-architect-skills in frankxai/skills. Commands read from the repo's manifest — 5 days ago.

Skills (15)

• claude-md

  Author, audit, and maintain CLAUDE.md memory files the way Anthropic recommends — under 200 lines, only-always-true facts, situational knowledge moved to path-scoped rules or skills. Use when creating/editing any CLAUDE.md, when a CLAUDE.md grows past ~200 lines, when rules stop being obeyed, when s

• claude-sdk

  Build autonomous AI agents using Claude Agent SDK with computer use, tool calling, MCP integration, and production best practices for Anthropic models

• langgraph-patterns

  Build production-grade agentic workflows with LangGraph using graph-based orchestration, state machines, human-in-the-loop, and advanced control flow

• openai-agentkit

  Build production-ready multi-agent systems using OpenAI AgentKit and Agents SDK with best practices for agent orchestration, handoffs, and routines

• mcp-2025-patterns

  Current best practices for Model Context Protocol server design, implementation, and integration. Updated patterns for 2025 MCP ecosystem including multi-server orchestration, security, and performance.

• mcp-architecture

  Design and implement Model Context Protocol servers for standardized AI-to-data integration with resources, tools, prompts, and security best practices

• mcp-least-privilege

  Use when building, auditing, or reviewing the authorization of an MCP server - "is my MCP server's OAuth set up right", "probe my remote MCP endpoint", "what scopes should my MCP server advertise", "protected resource metadata", "WWW-Authenticate for MCP", "can my MCP server pass the user's token to

• model-routing

  Intelligent model selection - routes tasks to Haiku (fast/cheap), Sonnet (balanced), or Opus (complex/strategic) based on task complexity analysis

• agent-stop-conditions

  Use when designing, reviewing or debugging the limits of an autonomous agent run - a headless claude -p job, an Agent SDK loop, a scheduled or unattended agent - or after a runaway run that looped, retried, overspent or never finished. Fires on "max turns", "max budget", "agent stuck in a loop", "ke

• agentic-orchestration

  Patterns for multi-agent coordination, task decomposition, handoffs, and workflow orchestration. Best practices for building and managing agent systems.

• fail-closed-eval-gate

  Use when a Claude Code plugin or skill change is about to merge or release and CI must block it unless the plugin measurably beats the no-plugin baseline. Fires on: 'gate this release on evals', 'fail CI if the plugin regresses', 'claude plugin eval passed but the delta is negative', 'can I trust th

• run-receipt

  Use when an agent run completes, fails, or is being audited. A multi-step run is not done until a receipt exists showing what ran, what failed, and what was not observed. No receipt = it did not happen. Triggers include "agent run complete", "audit this run", "incident investigation".

• tool-allowlist

  Use when deciding which tools an AI agent may call, writing or reviewing a Claude Code permissions block (allow, ask, deny in settings.json), scoping Bash, file read, file edit, web fetch or MCP tool access for one task, or after an agent ran a command it should not have. Triggers on "tool allowlist

• agent-design-review

  Use when the user asks to review, grill, pressure-test, or sanity-check an AI agent or multi-agent system design before building it — triggers include "grill me on my design", "design review", "review my agent architecture", "is this architecture sound", "pressure-test this before I build". Interrog

• defuddle

  Extract clean markdown content from web pages using Defuddle CLI, removing clutter and navigation to save tokens. Use instead of WebFetch when the user provides a URL to read or analyze, for online documentation, articles, blog posts, or any standard web page. Do NOT use for URLs ending in .md — tho

Similar plugins