LLM Mart Basic

pnpm 워크스페이스와 Vite 빌드를 설정·정비하거나, 의존성·락파일·번들 크기·폐쇄망 self-host 문제를 다룰 때 사용한다. 프레임워크 자체의 코드 작성이 아니라 빌드/패키징 층이 대상이다.

Go 코드를 작성·리뷰·리팩터링하거나 에러 처리, 동시성, 테스트, net/http 서버, go:embed 를 다룰 때 사용한다. Go 1.22+ 기준.

세션을 끊고 다음 세션에 넘긴다 — 협업 인프라(GitHub·GitLab·Forgejo·Jira·Plane·Slack)가 있으면 재개 가능한 상태를 그 트래커 이슈 본문으로 남기고 HANDOFF.md 에는 이슈 URL 을 가리키는 포인터만 적어 URL 하나로 건네고, 없으면 HANDOFF.md 자체에 전체 내용을 적어 그 파일로 끝낸다. 작업 이슈를 닫은 경계에서는 전용 이슈 대신 다음 작업 이슈에 코멘트 하나로 남긴다. 넘겨받은 쪽은 포인터나 파일을 읽고 적힌 단언을 실제와 대조한 뒤 이어서 착

에이전트 영속 메모리를 실제 근거(코드·DB·이슈 트래커·파일시스템)와 대조해 낡은 기억을 교정하고 죽은 기억을 아카이브 후보로 보고하는 감사 스킬. 메모리가 ~30개 파일을 넘었을 때, 큰 스택/인프라 변경(라이브러리 교체, 버전 업그레이드, 서버 이전, 스키마 삭제) 직후, 두 메모리가 서로 모순돼 보일 때, 또는 "메모리 정리해줘 / 감사해줘"(clean up / audit my memory) 요청에 사용한다.

사용자가 모바일 웹/앱의 기획서 / 화면설계서 / 스토리보드(storyboard) / 와이어프레임(wireframe) / IA / 화면기획을 요청할 때 도메인 불문(쇼핑, 커뮤니티, 예약, 뉴스, O2O, ...) 사용한다. PPT 스타일 16:9 슬라이드로 구성된 자체 완결형 HTML 파일 하나와, 화면 ID 를 키로 하는 Business Rules 마크다운 명세(검증·인터랙션·엣지케이스)를 산출한다.

mobile-web-planner의 Storyboard와 Business Rules를 동작하는 웹앱으로 구현할 때 사용한다. 프론트는 Next.js App Router 또는 Vite + React SPA 중 선택하고 화면·규칙 ID 추적표, 빌드, 핵심 User Flow 검증까지 완료한다. 기획 문서 없는 일반 React 컴포넌트 작업은 react-expert, Vite 설정만 다루는 작업은 frontend-build를 쓴다.

React 컴포넌트를 설계·구현·리팩터링하거나 상태 관리, useEffect 남용, 리렌더 성능, 접근성 문제를 다룰 때 사용한다. React 19 기준.

사용자가 "홈페이지 만들어줘" 등 단일 요청으로 서비스 전체 제작을 원할 때 기획(mobile-web-planner), 구현(nextjs-implementer), 보안(finguard)을 순차적으로 위임하고 통제하는 총괄 PM 메타 스킬. 스킬 룰 개정이나 진화는 skill-evolve 에 맡긴다.

사용자가 "녹음시작" / "녹음 시작" / "강의 녹음" / "회의 녹음" / "전사 시작" 이라고 하거나, 실시간 강의·회의·교육 세션의 녹음과 요약을 요청할 때 사용한다. 실시간 전사와 함께 오디오 원본(m4a)도 저장한다. 실행을 멈추는 "녹음종료" / "녹음 끝" 도 이 스킬이 처리한다.

피드백 기반으로 SKILL.md를 자동 개선하는 메타 스킬

SQLite 파일을 직접 읽고 쓰거나, 읽기 전용 조회·WAL·잠금·마이그레이션·동적 테이블명 주입 같은 SQLite 고유 문제를 다룰 때 사용한다.

用户明确说“规格已确认，开始实现”或要求按已审批 CM specs 开发时使用。新任务默认由 JS workflow 驱动 N1-N8，完成开发、独立审查、QA 与文档同步；模糊点子、未审规格和单独一句“继续”不能触发编码批准。

后端 API 工程师 Skill，执行服务端 API 层开发（路由、业务逻辑、鉴权中间件、缓存、队列），自动适配语言和框架

用户说“检查工作流是否安装正确”“为什么找不到 cm 命令”时使用。默认查询 npm 稳定版，有新版自动升级已管理的 CM 安装，再检查插件、核心 Skills、兼容包装与模板引用；不测试或修改业务代码。

智能合约工程师 Skill，执行合约开发、测试、部署，自动适配 EVM/Solana/Move 等链和开发框架

数据库工程师 Skill，执行数据模型设计、migration、查询优化，自动适配 ORM 和数据库类型

发布/运维工程师 Skill，执行 staging 部署、冒烟验证、发布记录与生产发布待决清单编制，自动适配部署栈；生产发布与基础设施变更强制人工确认

文档同步 Skill，开发完成后自动更新 README、.claude/ 配置、specs CHANGELOG，保持文档与代码一致

金融专家 Skill，覆盖 Web3 与证券/资产/交易领域的正确性审核、营销合规红线识别、合规问题清单生成、业务验收协同；把关型角色，只举旗不定性

用户说“修复这个可复现 bug”或要求根据失败报告修代码时使用。执行红灯测试、根因定位、最小修复、独立审查和回归；尚未确认的问题先用 cm-test，新功能和架构重设计转交 cm-prd。

Discovery worked. Ping worked. Every TCP connection timed out, and later the tunnel only worked when someone had a terminal open.

Every VM came back. The cluster did not. Declarative systems converge on config, and the datapath isn't config.

A surprising share of AI-in-the-terminal failures aren't the AI. They're zsh, and a version of bash from 2006.

A Claude Code plugin turns standalone project configuration into a namespaced, installable extension that teams and communities can update as one unit.

None of the safety came from the model. It came from six boring habits.

Skills package instructions and references. Subagents run work in a separate context and return results. They solve different problems and can be composed deliberately.

Six hours in, one step left, everything green, and the incident that didn't happen

CLAUDE.md carries persistent project context. Skills load reusable procedures when relevant. Separating stable facts from task-specific workflows keeps both easier to maintain.

Twenty minutes recovering secrets that never existed, and the one sentence from a human that ended it

An API request routing a model's tool call through an approval gate to a remote MCP server

31 config keys, two audits, and why the first one was wrong in both directions

The official MCP Registry stores standardized server metadata rather than package code. Publishers verify a namespace, describe installation or remote access, and submit immutable versions.

Everyone looks at the Dockerfile. The file that actually leaked the key was the project file.

Remote MCP authorization uses established OAuth standards, but secure integration still requires issuer validation, least-privilege scopes, protected token handling, and server-side enforcement.

"Copy it over and switch the reference" is two steps, and the outage lives in the one nobody checks

stdio fits local processes and prototypes. Streamable HTTP fits hosted services and shared integrations. The right choice follows where the capability runs and who must reach it.

The most important rule wasn't about what I could change. It was about what I was allowed to display.

Tools perform operations, resources expose readable context, and prompts provide reusable templates. Choosing the correct primitive makes an MCP server easier to understand and govern.

Use MCP Inspector to connect to local or remote servers, inspect capabilities, call tools, read resources, test prompts, and diagnose failures before release.

Build an MCP server in TypeScript with focused tools, validated schemas, local and remote transports, Inspector tests, and production security controls.

{{SKILL_ENTRY:refactor}}

{{SKILL_ENTRY:release}}

Review a diff with the reviewer fleet, funneled to one triaged verdict. Targets the current working diff, a path, or an inbound GitHub PR.

Exploratory spike on a throwaway branch — answer a named question with disposable code. Never merges; exits to a findings note or {{CMD:feature}}.

Autonomous sprint — one interactive spec gate, then plan-to-PR execution with every auto-decision SMARTS-scored and logged. Hard gates remain true stops.

Daily repo hygiene — review the day's repo state, then perform the cleanups under per-action confirmation. Fast-forward only, never destructive without a yes.

Show the project's current state at a glance — stage, open tasks, open questions, overrides since the last checkpoint, current branch. Read-only.

Wire codeArbiter's statusline into ~/.claude/settings.json, or remove it.

The sanctioned task-board mutator — add a queued task, start one (flips to in-progress and stamps the date, minting a dotted ID on pick-up), or mark an in-progress task done. The only blessed write to open-tasks.md.

{{SKILL_ENTRY:security-architecture}}

{{SKILL_ENTRY:tribunal}}

Watch a PR's CI to completion — diagnose on red, notify and offer the merge on green. Never auto-merges.

Copy a file OUT of a running sandbox box to the host — host-initiated egress only (docker cp). The reverse, a host→container bind, is impossible by construction.

Tear down a sandbox box — remove its container and named volume. --keep-volume leaves the volume; with no id, prune reclaims any leaked ca.sandbox=1-labeled object. Cached images are retained.

Run a single command inside a running sandbox box and capture a JSON result — exitCode, separate stdout/stderr, and a truncated flag past the byte cap. The scriptable exec seam.

Open an interactive shell inside a running sandbox box at /work/repo. Read-only root, non-root user, no host-FS access — explore the untrusted code interactively, then exit.

Pull an untrusted repo into an ephemeral, host-FS-isolated Docker container — clone into a named volume, build a dep-cached image, run under structural isolation. Network defaults to offline. Requires Docker and nixpacks.

Vet a new or changed third-party dependency for license, provenance, and supply-chain risk before any install runs.

Inspect ADR health read-only; optionally select one ADR with --adr N.

Record user-decided ADRs or inspect their health read-only. Preserve attribution and acceptance evidence.

Atom Agent, Open-Source Governed AI Agent Platform for Self-Hosted Automation

The agent engineering intelligence harness, optimized tools, memory system, subagents and mixture of models packages ⚚

Sesori iOS/Android app and the Sesori Bridge CLI — drive Claude, Codex, OpenCode, Cursor, Pi, OMP, Hermes coding sessions from your phone

🧠 RepoBrain (formerly Antigravity) — Give your repo a brain. ChatGPT for your codebase: works in Claude Code, Cursor, Codex, Windsurf & more.

Use ChatGPT (Codex), Claude, and Grok (X Premium) subscriptions as DeepSeek Harness LLM providers — OAuth login in the web UI, no API keys

Latitude is the open-source AI monitoring platform.

Open-source AI browser agent for Chrome and Firefox (monorepo) 🧠

Official Model Studio CLI（阿里云百炼 CLI）built for AI Agent frameworks, exposing models, search, multimodal, and workflow capabilities as structured tool calls.

Curated DeepSeek Harness (DSH) plugins, extensions, tools, skills, clients, runtimes, integrations, and verified references — English and Chinese.

Turn papers, code, and docs into presentation-ready, natively editable PPTX in Codex / Claude Code. Native charts and equations, speaker notes, click-build anim…

Grix : Work with agents like talking to people.

Local First Ai Agent. Optimized for Local Ai models. Long context window. Proper tools callings. Runs privately on your device.

Model Context Protocol server that integrates AgentQL's data extraction capabilities.

AI that ships your tickets.

Metadata-driven CLI for AI Agent Bots — 48 operations across 7 domains, structured JSON envelope I/O, zero interactive prompts.

An open-source, extensible, self-hosted agent workspace with multi-runtime support for Codex, Claude Code, and more, plus reusable local apps for custom interfa…

Open-source Windows desktop client and GUI for DeepSeek Harness — zero-setup installer with Codex, plugins, skills, SSH, mobile remote access, and 11 skins.

The open-source AI workbench for scientific research

DeepSeek-native AI coding agent for your terminal. Engineered around prefix-cache stability — leave it running.

A Fully free agentic browser driver for AI , few tools, full control, real stealth, top-tier token efficiency.