LLM Mart Basic

Use when implementing a multi-file change, building a feature from a breakdown, or writing a large amount of code. Not for a single settled ticket: use work.

Use when writing or restructuring code, before adding a helper, wrapper, config key, or dependency, or when the user asks for minimal or DRY code. Not for performance tuning: use optimize.

Use when asked to audit comments in code files and propose structural replacements or deletions with per-candidate approval. Not for deterministic commented-out-code removal: use deslop.

Use when a human says "overhaul", "rebuild this subsystem", or "rewrite it from scratch". Not for thin-slice features: use incremental-implementation. Not for root-cause repair: use strike-the-root.

Use when asked to optimize code, speed up a path, reduce allocations, repair a regression, or profile a target. Not for remote, credential, publish, deploy, or irreversible changes.

Use when a request names a working principle (subtract before you add, idempotent operations, never block on the human) or asks which principle applies. Not for running a repair: use strike-the-root.

Use when modernizing APIs, removing compat shims, killing feature flags, or rewriting a subsystem cleanly. Not for additive refactors that must preserve the old path.

Use when a trusted bug or performance report needs reproduction and fix. Not for untrusted reports or scope beyond the named feature.

Use when an implementation has more workarounds than structure and another patch will not pay. Not for in-place re-derivation: use breaking-driven. Not for one-artifact rewrites: use rewrite-clean-v0.

Use when the user says "simplify this diff" or asks for a compression pass over a change-set. Not for dead-code sweeps: use deslop.

Use when an exact symbol, path, entrypoint, or line range can bound a focused code question or patch proposal under a fixed source budget. Not for source changes or broad repository exploration.

Use when writing or verifying framework-specific code, boilerplate, or a documented, correct implementation. Not for remote, credential, publish, deploy, or irreversible changes.

Use when a feature begins or specs are checked in: author or update behavioral specs and keep them current with what ships. Not for producing the initial approved spec and plan: use spec-driven.

Use when a bug, failure, flake, regression, review finding, or ticket needs the core fixed so it cannot recur. Not for greenfield features: use tdd. Not for style-only review or typo-class one-liners.

Use when the user says greenfield this or rescue this codebase, names a field (dark, red, blue, or brown), or diagnoses a subsystem. Not for specs: use to-spec. Not for remote or irreversible changes.

Use when an abstraction leak must be sealed as a module seam, configuration option, or explicit override, or exposed as a named boundary. Not for detecting concealment patterns: use no-hide.

Use when the user says "audit my code", "find all the bugs", "review until clean", or "grill my changes". Not for remote, credential, or irreversible changes.

Use when asked to audit an agent or AI feature for agentic-experience quality (AX review, agent-native critique, trust question). Not for source or remote-system changes.

Use when asked to determine what a change could break before it ships. Not for remote, credential, publish, deploy, or irreversible changes.

Use when a user wants to identify the true sources of complexity qualitatively before counting metrics. Not for source or remote mutation.

Fourteen posts of being wrong in production, compressed to checkboxes

Healthy nodes, a quiet network, 300 restarts in three days, and a latency budget measured in milliseconds

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.

`crabbox attach` follows the recorded events of an active coordinator run and

`crabbox azure` groups Azure provider setup commands. It currently has a single

`crabbox bench` records and reports local benchmark timing observations. It is a

`crabbox cache` inspects, purges, or warms package and build caches on a

`crabbox capsule` captures, replays, and tracks lightweight failure capsules.

Save the state of a lease, then restore it onto another box or fork it into a

`crabbox claims list` prints the lease claims stored on the current machine. It

`crabbox cleanup` sweeps direct-provider machines and local provider state that

`crabbox code` bridges a Linux lease's `code-server` workspace into the

`crabbox config` inspects and updates user configuration. It has three

`crabbox connect` resolves a lease and opens an interactive SSH session to it.

`crabbox cp` copies files or directories between the host and a Crabbox-owned

`crabbox desktop` drives a visible desktop session on a lease that was warmed

`crabbox doctor` runs a preflight before you commit to a long workflow. It is

`crabbox egress` gives a lease mediated outbound network: a lease-local browser

`crabbox events` prints the broker's event log for a recorded run.

`crabbox heartbeat` refreshes the idle deadline for one owned lease and prints

`crabbox history` lists recorded remote command runs from the broker. Each run is

`crabbox image` holds the trusted-operator controls for provider base images:

`crabbox init` onboards the current repository: it writes the minimal config

Open‑WebUI Tools is a modular toolkit designed to extend and enrich your Open WebUI instance, turning it into a powerful AI workstation. With a suite of over 15…

The token-efficient agentic coding workbench. Built for a future where every token counts — it optimizes token usage at the agent-loop level, saving 70%+ on lon…

MCP server for controlling a real iPhone via macOS iPhone Mirroring...and any MacOs app. Screenshot, tap, swipe, type — from any MCP client.

One desktop for all your AI coding Agent — Claude Code, Codex CLI & Gemini CLI. Auto-detect, one-click install, unified chat, file explorer, terminal & editor.…

Pi — A cross-platform AI coding agent, bringing the Claude Code experience to your desktop. No environment setup, no terminal commands. Download and start codin…

DeepSeek Harness Desktop (dsh-desktop). EAC: Embracing All Creation (揽尽万象). Bundled Node.js runtime with full dsh-CLI kernel, one-click startup, 10 built-in UI…

A telegram bot for searching and auto-saving.

金蝶云星空 K3Cloud MCP Server，让 AI 助手（Claude Desktop、Claude Code、Cursor、Cline、Cherry Studio、Openclaw 等任意支持 MCP 协议的客户端）通过自然语言查询和操作金蝶 ERP 系统。

Rewrite frontier research using allegorical structural mapping.

Free4Chat is a temporary collaboration fabric for Humans and Agents.

Give any AI agent a full desktop — it sees the screen, clicks, types, and runs apps like a human. Automate anything with a UI: browsers, legacy software, intern…

A personal AI agent that can work safely on your machine, remember useful context, and keep its data under your control.

A framework for AI-driven economic activity. Declarative, composable, observable, deterministic.

Open-source, self-hosted customer support desk in a single binary. A lightweight alternative to Intercom, Zendesk, Chatwoot.

54 rigorous skills for Codex, OpenCode, and Pi: code review, security audit, feature development, frontend design, MCP tools, Hugging Face ML/training, and more…

A wonderful list of Game Development resources.

Open Science Desktop — local-first, model-agnostic AI research workbench for macOS, Windows & Linux. Open-source Claude Science desktop alternative built on Tau…

Lexora — Personal AI workspace built around Desktop / 以 Desktop 为核心的个人 AI 工作台

Open-source, AI-native Evernote alternative with native MCP. Zero-cost on Cloudflare or self-hosted with Docker.

Full job-hunt Claude skill bundle — Job Description decoder + Resume builder (11 templates) + Behavioral Interview / story bank. Three self-contained sub-skills…