Claude Skills by xiaolai

Anthropic Messages API reference. Use for POST /v1/messages, tool_use/tool_result, count_tokens, batches, streaming, prompt caching (TTL, breakpoints), anthropic-beta headers, admin API (API keys, workspaces), compliance (residency, audit logs), model IDs/deprecations/context windows. Skip: Claude Code (use claude-code), Agent SDK (use claude-agent-sdk), Agent Skills/tool guides (use anthropic-platform-features), MCP connectors (use claude-connectors), per-language SDK types (use your IDE).

Anthropic platform features above the raw API. Use for Agent Skills, hosted MCP connector + Tunnels, computer use/code execution/bash, extended thinking, batches, prompt caching strategy, citations, Bedrock/Vertex, embeddings, fast mode, context editing, workload identity (WIF), billing, SSO, CMEK, Managed Agents, self-hosted sandboxes. Skip: raw /v1/messages (use anthropic-api), Claude Code (use claude-code), Agent SDK (use claude-agent-sdk), Connectors directory (use claude-connectors).

Anthropic news + research digest, refreshed daily from anthropic.com/news + /research. Use for TIME-SENSITIVE questions: model launches ("did Claude X just release?"), product announcements, partnerships, region launches, research papers, recent policy, "what's new from Anthropic?", "what did Anthropic just announce?", "any recent research on X?". Skip: deep technical reference (use the other 7 skills), news older than ~30 items (link to anthropic.com), full paper text (WebFetch the URL).

Claude Agent SDK (TS @anthropic-ai/claude-agent-sdk, Python claude_agent_sdk), wrapping Claude Code. Use for query()/ClaudeSDKClient, SDK hooks, subagents (AgentDefinition), SDK MCP servers, permissionMode/permission_mode, structured outputs, Docker/K8s sandbox, sessions (resume, fork, SessionStore to S3/Redis/Postgres), SDK errors. Skip: Claude Code CLI config (use claude-code), Messages API (use anthropic-api), MCP spec (use mcp-spec), hosted Managed Agents (use anthropic-platform-features).

Claude Code (the CLI) reference. Use for .claude/settings.json / settings.local.json, hooks (PreToolUse, PostToolUse, Stop, SubagentStop, SessionStart, etc.), slash commands, agents, MCP servers in .mcp.json, plugins and marketplaces, ANTHROPIC_* / CLAUDE_* env vars, permission modes, the ~/.claude/ layout, or "what does <feature> in Claude Code do". Skip: Messages API (use anthropic-api), Agent SDK (use claude-agent-sdk), MCP spec (use mcp-spec), general shell/git/programming.

Claude Connectors (user-facing MCP integrations on claude.com). Use for enabling a connector, building a custom connector, OAuth + tokens, submitting to the connectors directory, GitHub/Google/Microsoft/Slack connectors, Desktop extensions (MCPB), MCP Apps design guidelines, user-installable Skills, installing plugins in the app. Skip: MCP protocol spec (use mcp-spec), Claude Code MCP servers (use claude-code), API MCP connector (use anthropic-platform-features → SKILL-agents-and-tools).

Claude Cowork (Claude for Work on third-party clouds) and Office agents. Use for deploying Claude for Work on Bedrock or Microsoft Foundry, LLM gateways, enterprise SSO, telemetry / audit logs, the M365 connector, enterprise policy controls, Claude in Excel, Slack-style office agent integrations. Skip: raw API on Bedrock / Vertex (use anthropic-platform-features), user-facing connectors directory (use claude-connectors), Claude Code CLI in enterprise contexts (use claude-code).

Model Context Protocol (MCP) open spec. Use for writing an MCP server or client (any language; TypeScript @modelcontextprotocol/sdk, Python mcp), JSON-RPC framing, initialize handshake, capability negotiation, stdio/streamable HTTP/SSE transports, defining tools/resources/prompts, sampling, roots, completion. Skip: Anthropic's hosted MCP connector (use anthropic-platform-features), Claude Connectors directory (use claude-connectors), Claude Code's .mcp.json (use claude-code → SKILL-mcp).

Template starter for an Agent Skill — replace this entire description with one that names the concrete intent triggers ("Use when the user asks about X, Y, Z") and lists negative-intent Skip clauses. The platform's skill matcher reads this field — write it for matching, not for marketing. Stay under 1024 chars.

Template example showing the SKILL.md structure — replace this description with one that names YOUR skill's concrete trigger condition ("Triggers when the user asks how to X, Y, Z"). Keep it ≤200 characters for best intent-matching.

Use when the user wants help looking after their Mac — a health checkup, reclaiming disk space, organising files, diagnosing a problem in plain language, setting up backups, fixing Wi-Fi or a home network, automating a repetitive chore, or understanding what just happened. Also the entry point when the user types _it or the IT guy's name with a leading underscore. Routes to the right workflow, loads the binding safety contract first, and never treats a diagnosis as permission to act.

Decision logic for home network work — isolating whether a problem is the device, the Wi-Fi, the router, or the ISP before spending money; the fix ladder from free to hardware; router and mesh buying judgement; and the boundaries of whose network may be touched. Load before any /mac-it-guy-pro:network work.

Core conventions for all mac-it-guy-pro commands and agents — the safety contract, plain-language rules, report formats, and state file layout. Load before any IT task (checkup, cleanup, organize, fix, automate, backup, onboard).

How the IT guy remembers, refreshes, and retires what it knows about this machine — the profile schema, provenance classes, expiry by evidence, the belief ledger, demotion to history, the visit log, and undo manifests. Load when reading, writing, reviewing, or pruning anything in ~/ITGuy/.

Exact macOS commands for IT diagnostics and safe actions — disk, memory, startup items, updates, Time Machine, battery, SMART, duplicate detection, Trash-based deletion, HEIC conversion, photo dates, launchd scheduling. Load before running any diagnostic or maintenance command on a Mac.

Decision logic for building a personal unrestricted-internet connection from a censored network — which architecture to use (CDN-fronted vs direct), which protocols work and which are detectable, and the boundaries of what to help with. Load before any /mac-it-guy-pro:open-internet work.

The micro-product contract for the IT Guy toolbox — acceptance criteria, directory layout, README template, registry schema, dry-run requirement, double-clickable wrappers, the evolution ladder, and the pattern catalogue used to offer a user automations they did not know to ask for. Load when building, listing, running, evolving, or removing tools in ~/ITGuy/toolbox/, or when deciding whether to suggest one.

How the IT guy explains rather than just fixes — the two teaching modes (just-in-time and deliberate study), the learning-map structure, grounding lessons in the user's own measurements, depth scaling, and the pull-not-push rule. Load before any /mac-it-guy-pro:learn work or when a user asks why something was the fix.