发现 Skills

Master skill for TypeScript backend development. Decision framework for APIs (tRPC/REST), authentication (Auth.js/Passport), database (Prisma), validation (Zod), logging (Pino), testing (Vitest), and deployment (Docker). Routes to specialized skills for implementation. Use as entry point for any backend task.

Use to combine multiple sources into a single paragraph or argument — the strongest plagiarism shield. Patterns for agreement, disagreement, supplementation, and gap-naming across sources.

Execute generated code.

Browser automation via HTTP server. Supports multiple concurrent sessions for multi-user testing, saved flows, and profile-based auth persistence.

Browser automation, debugging, and performance analysis using Puppeteer CLI scripts. Use for automating browsers, taking screenshots, analyzing performance, monitoring network traffic, web scraping, form automation, and JavaScript debugging.

Routing reference for live ad-platform operations at Creekside Marketing — both READING data and MAKING CHANGES. Meta / Facebook / Instagram is handled via PipeBoard MCP's Meta connector (verified live, full read + write surface). Google Ads is handled via PipeBoard MCP's Google Ads connector under a separate namespace (verified live 2026-04-23 — `list_google_ads_customers` returned 33 Creekside accounts); full read + write surface. Both MCPs are provided by PipeBoard. A read-only Google Ads Python SDK pipeline backs up historical data in Supabase. Use whenever a user or agent asks how to pull, query, read, create, update, pause, duplicate, or otherwise manage ads, ad sets, campaigns, creatives, audiences, pixels, keywords, or lead forms on either platform. Do NOT use for historical trend queries against Supabase tables (e.g. meta_insights_daily, google_ads_insights_daily) — this skill is for live, direct-from-platform operations.

This skill should be used when the user asks to 'specify a feature', 'write requirements', 'define what we're building', 'capture requirements', 'document the spec', or before invoking /architecture-tech-lead for non-trivial features. Produces formal specifications (WHAT/WHY) that feed into architecture and planning phases.

>-

>-

Update mechanisms and distribution strategies for AI agents and automations. Use when the user mentions: update agent, upgrade agent, agent versioning, semantic versioning agents, auto-update, rollback agent, migration, breaking changes, distribute updates, agent release, MCP update, plugin update, docker update, npm update agent, pip upgrade agent, agent rollback, version pinning

Audits Claude Code skills, hooks, agents, and rules in a single pass — structural lint, quality scoring (1-5 across 6 weighted dimensions), and prioritized improvement recommendations. Use when reviewing, auditing, scoring, linting, improving, or validating any customization. Produces one sectioned report with pass/fail checks, per-dimension scores, weighted total, quality tier, and P1-P5 action items.

Render DNA codes to Pencil .pen frames. Does ONE thing well. Input: DNA code + component type (hero, card, form, etc.) Output: .pen frame ID + screenshot Use when: design-exploration or other orchestrators need to render visual proposals using Pencil MCP backend.

Best practices for integrating the Pillar SDK into React and Next.js applications. Use when adding an AI assistant panel, help center widget, or embedded product assistant.

Design and execute chaos engineering experiments using Litmus or Chaos Mesh. Test system resilience through controlled fault injection, validate hypothesis-driven tests, and improve failure recovery. Use before major product launches, after architecture changes to validate resilience, during GameDays or disaster recovery drills, to validate assumptions about failure modes, or as part of an SRE maturity program.

Parse and validate PILES (Puzzle Input Line Entry System) notation for specifying piece fusion groups in jigsawR. Covers syntax validation, parsing into group lists, plain-language explanation, adjacency verification against puzzle results, and round-trip serialization. Use when validating user-supplied PILES strings before passing to generate_puzzle(), debugging fusion group issues, explaining the notation to users, or testing round-trip parse/serialize fidelity.

Build or modify Home Assistant sensors/entities that must persist native values and custom attributes across restarts using RestoreSensor/RestoreEntity plus ExtraStoredData (e.g., history, timestamps, scenarios); apply when persistence of non-native fields is required.

Review, prune, or rewrite global rules files such as global AGENTS.md or ~/.claude/CLAUDE.md. Use when deciding whether durable behavior rules belong in the global context layer across tasks, projects, and sessions; when diagnosing repeated collaboration friction; or when producing directly applicable global-rule edits. Exclude project-local AGENTS.md/CLAUDE.md setup, skill routing fixes, task context packing, and tooling enforcement. Formerly named context-engineering-global.

Digital Audio Workstation usage, music composition, interactive music systems, and game audio implementation for immersive soundscapes.

Manage OpenClaw Gateway on Windows PC

Provides expertise for analyzing DWARF debug files and understanding the DWARF debug format/standard (v3–v5). Triggers when understanding DWARF information, interacting with DWARF files, answering DWARF-related questions, or working with code that parses DWARF data.

Möbius inversion on posets and lattices: alternating sums, chromatic polynomials, incidence algebras, and centrality predicates.

A package for creating diverse scenes in MuJoCo. Compose environments with objects, terrains, and obstacles for robot training.

Tool selection decision framework, domain validation (visual vs audio), cost awareness, and when to use which tool. Use when deciding which tools to use for a given task, especially when working with MCP servers.

Design and build Model Context Protocol (MCP) servers — choosing between tools, resources, and prompts; schema design that LLMs can actually use; error handling that produces recoverable failures; auth patterns; transport selection; pagination and large-output handling; and the security considerations specific to exposing tools to autonomous agents. Use this skill whenever the task involves building an MCP server, exposing functionality to Claude or another MCP client, debugging "the model isn't calling my tool" or "the model is calling my tool wrong," reviewing an MCP server design, or any work where the goal is letting an LLM interact with a system via MCP. Trigger on terms like "MCP," "Model Context Protocol," "MCP server," "MCP tool," "Anthropic connector," "expose this to Claude," and similar.