LLM 與 Agent

Use when building persistent codebase intelligence for AI agents or integrating knowledge systems via MCP

Production-grade AI agent patterns with MCP integration, agentic RAG, handoff orchestration, multi-layer guardrails, and observability (modern best practices)

Expert on building decentralized applications with Shelby Protocol storage on Aptos. Helps with dApp architecture, wallet integration (Petra), browser SDK usage, React/Vue integration, file uploads, content delivery, and building Shelby-powered applications. Triggers on keywords Shelby dApp, build on Shelby, Shelby application, Petra wallet, browser storage, web3 app, decentralized app Shelby, React Shelby, Vue Shelby.

Enables GitHub repository operations (read/write/commit/PR) for Claude.ai chat environments. Use when users request GitHub commits, repository updates, DEVLOG persistence, or cross-session state management via GitHub branches. Not needed in Claude Code (has native git access).

Operational prompt engineering patterns, templates, and validation flows for Claude Code.

Expert error coordinator specializing in distributed error handling, failure recovery, and system resilience. Masters error correlation, cascade prevention, and automated recovery strategies across multi-agent systems with focus on minimizing impact and learning from failures.

Use this agent when delegating research tasks that require collecting and verifying beyond surface-level sources. The agent crawls, collates, verifies, and fact-checks information, providing evidence of its verification process.

Secret detection patterns and scanning workflow. Auto-loaded by security-reviewer agent for all security audits. Contains regex patterns for 25+ secret types (AWS, Google, GitHub, Stripe, etc.) and CLI scanning commands.

Exports project instructions and knowledge files from the current Claude project. Use when users want to clone, copy, backup, or export a project's configuration and files.

Use when building e2e tests with Playwright, under deadline pressure to ship incomplete coverage, or struggling with wait strategies and mock-vs-reality tradeoffs - provides patterns for edge case coverage, deterministic waits, and strategic mocking decisions

Expert chaos engineer specializing in controlled failure injection, resilience testing, and building antifragile systems. Masters chaos experiments, game day planning, and continuous resilience improvement with focus on learning from failure.

Use when creating or developing, before writing code or implementation plans - refines rough ideas into fully-formed designs through collaborative questioning, alternative exploration, and incremental validation. Don't use during clear 'mechanical' processes

Expert on Model Context Protocol (MCP) integration, MCP servers, installation, configuration, and authentication. Triggers when user mentions MCP, MCP servers, installing MCP, connecting tools, MCP resources, MCP prompts, or remote/local MCP servers.

Synthesize findings from multiple research sources into coherent insights. Use when combining data from different sub-agents or research threads.

Production-grade mobile app development with Swift (iOS), Kotlin (Android), React Native, and WebView patterns, including UI/UX, navigation, state management, networking, local storage, push notifications, and App Store deployment.

Interactive slash command creator for Claude Code. Triggers when user mentions creating commands, slash commands, command templates, command arguments, or wants to build a new command workflow.

AGGRESSIVELY use TOON v2.0 format for biggish regular data (≥5 items, ≥60% uniform). Auto-applies to tables, logs, events, transactions, analytics, API responses, database results. Supports 3 array types (inline, tabular, expanded), 3 delimiters (comma, tab, pipe), key folding for nested objects. Triggers on structured data, arrays, repeated patterns. Use TOON by default when tokens matter - RAG pipelines, tool calls, agents, benchmarks. Keywords "data", "array", "list", "table", "log", "transaction", "metric", "analytics", "API", "database", "query", "TOON".

Creates browser-executable JavaScript bookmarklets with strict formatting requirements. Use when users mention bookmarklets, browser utilities, dragging code to bookmarks bar, or need JavaScript that runs when clicked in the browser toolbar.

Interactive hook creator for Claude Code. Triggers when user mentions creating hooks, PreToolUse, PostToolUse, hook validation, hook configuration, settings.json hooks, or wants to automate tool execution workflows.

Install skills from github.com/oaustegard/claude-skills into /mnt/skills/user. Use when user mentions "install skills", "load skills", "add skills", "update skills", "refresh skills", or references a skill not currently installed.