LLM Mart Basic

当实现完成、所有测试通过、需要决定如何集成这份工作时使用

MCP 服务器构建方法论 — 系统化构建生产级 MCP 工具，让 AI 助手连接外部能力

收到代码审查反馈后、实施建议之前使用，尤其当反馈不明确或技术上有疑问时——需要技术严谨性和验证，而非敷衍附和或盲目执行

完成任务、实现重要功能或合并前使用，用于验证工作成果是否符合要求

当在当前会话中执行包含独立任务的实现计划时使用

遇到任何 bug、测试失败或异常行为时使用，在提出修复方案之前执行

在实现任何功能或修复 bug 时使用，在编写实现代码之前

当需要开始与当前工作区隔离的功能开发，或在执行实现计划之前使用——通过原生工具或 git worktree 回退机制确保隔离工作区存在

在开始任何对话时使用——确立如何查找和使用技能，要求在任何响应（包括澄清性问题）之前调用 Skill 工具

在宣称工作完成、已修复或测试通过之前使用，在提交或创建 PR 之前——必须运行验证命令并确认输出后才能声称成功；始终用证据支撑断言

在 Claude Code / OpenClaw / Cursor 中直接运行 agency-orchestrator YAML 工作流——无需 API key，使用当前会话的 LLM 作为执行引擎。当用户提供 .yaml 工作流文件或要求多角色协作完成任务时触发。

当你有规格说明或需求用于多步骤任务时使用，在动手写代码之前

当创建新技能、编辑现有技能或在部署前验证技能是否有效时使用

Management-consulting writing craft — McKinsey SCR (Situation·Complication·Resolution), Minto Pyramid/MECE, BCG bold-bullet executive summary, so-what upfront, numeric precision, Forrester Landscape. Use when writing or reviewing an executive summary for decision-makers or a roll

Encyclopedic neutral-reference writing craft — NPOV (attribute facts not opinions, due weight, neutral faction labels, verdict restraint), summary style and Coatrack avoidance, wikilink conventions (link density, first-mention, slug alias, abbreviation glossing). Use when writing

Guideline-authoring craft for instruction SoTs (.claude/ guides, CLAUDE.md, plan files) — operative rules vs recital, MUST/SHOULD/MAY force tiers, pruning, bloat control, blind review protocol, deliberation-narrative ban. Use when writing, editing, or reviewing an agent instructi

Journalism and argumentation writing craft — inverted pyramid, lede, nut graph, kicker, explainer framing, PAGE frames, Toulmin argument (claim/rebuttal/qualifier), Hegelian dialectic, BBC due impartiality. Use when writing or reviewing news/explainer pieces, landscape overviews,

Forces the laziest solution that actually works, simplest, shortest, most minimal. Channels a senior dev who has seen everything: question whether the task needs to exist at all (YAGNI), reach for the standard library before custom code, native platform features before dependenci

Verifiable-attribution and citation-discipline craft — atomic claim decomposition, evidence grading (primary/analysis/forecast tiers), claimant attribution, citation typing (cites/references/contradicts/defines), source anchoring (Xanadu). Use when writing or reviewing a referenc

Set up and review CI, artifact publishing, versioning, and release governance for Minecraft 26.x or legacy 1.21.x mods and Paper plugins. Use for pipelines and releases, not gameplay implementation or server operations.

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.

Assemble the governance record for a range — commits, overrides, ADRs, sprint auto-decisions, open questions, checkpoint findings — into one dated audit packet. Read-only.

Lightweight Q&A about the project — answer from context and return, no routing, no state change.

Periodic multi-reviewer sweep of the whole codebase — surfaces a triaged checkpoint report.

Sanctioned lane for non-behavioral work — docs-only edits, dependency bumps, reverts. Type-scaled gates; no TDD demanded of prose.

Clean up an already-merged local branch after proving containment. Confirm each discard and preserve unique work.

Show the codeArbiter command catalog — the public command list and what each routes to.

Create a verified local Git commit when committing changes is requested. Not for explaining commit history, drafting a message only, or postponing a commit. Applies every commit gate and never implies a push or PR.

Stop everything and surface a rule conflict — persona vs. docs vs. code. Present both sides and the conflict-hierarchy level; the user resolves. No silent reconciliation.

Audit stale provenance-tracked docs on request. Report first; re-scout or re-baseline only for selected docs.

Build project context from an existing codebase through isolated scouts, resolve gaps, and preserve initialization gates.

Investigate an unexplained defect or unexpected behavior without changing application code. Use for root-cause diagnosis and an evidence-backed handoff. A no-action close records a board note. Not for implementing a known fix, new features, or explanation-only questions.

Develop greenfield project context through a layered interview, preserve decisions, and initialize only after the required gates.

Verify the active host install, package, command ownership, enforcement, and harmless live-fire probe. Read-only.

Start a feature: brainstorm a spec, get it approved, then drive it test-first through the pipeline. The one entry to implementation.

Fix a confirmed bug: a failing regression test first, then a minimal fix, then the rest of the tdd gates.

Opt this repo into codeArbiter — scaffold the root-level .codearbiter/ state store.

Read-only 3-metric governance glance — override rate, small-lane rate, sprint low-confidence ratio — each with a trend arrow vs. the prior 20-commit window.

Sanctioned, logged bypass of a gate or hard rule — one audit line, then proceed.

Open a PR or finish branch disposition; route CI watching and post-merge cleanup to their owners. Merge and discard need explicit authority.

Zero-onboarding, read-only dry-run of the reviewer fleet against the current uncommitted diff. Predicts reviewers, runs the state-free secret scan, writes nothing.

PiG (Pi in Go) is a faithful Go port of upstream Pi, the TypeScript codebase behind the Pi coding agent. It is a parity-bound translation, not a rewrite: upstre…

An AI Agent that lives in your pocket. Local-first and privacy focused.

Unofficial skill that teaches coding agents to build with TypeSafe AI's Jev: typed decisions, calibrated confidence, and prior art from 150+ community projects.

Adaptive Test-time Learning and Autonomous Specialization

Prediction-market trading engine — Wang Transform pricing on 291K+ contracts; paper-traded across Kalshi · Polymarket · Solana DFlow (Jito bundles) · 633 tests

Knowledge Management for Humans and Agents

Open-source Claude Cowork / Codex / WorkBuddy alternative — a local-first AI office agent that turns one request into real PPTX, DOCX, XLSX and HTML files. Runs…

DeepAgent Code: AI coding agent with persistent memory and control plane

Awesome Jev — evidence-graded index of TypeSafe System One: SDKs, MCP tools, agents, apps and open models. 20 languages, rebuilt every 2 hours.

CLI for Telegram — agent-friendly, daemon-based, with webhook event push.

AI deep-research agent that turns any question into a cited report: plans searches, reads real sources, verifies evidence. Self-hosted, multi-provider, Docker-r…

Event-stream AI Agent framework for building your persona bot 🍊

Give the agent a machine. Just not yours. Each AI coding agent gets its own isolated machine with root, Docker, and systemd - active defense detects and stops t…

Local Emperor-style AI agent with Vue WebUI, multi-provider LLMs, streaming chat, tools, skills, memory, and token telemetry.

Open-source AI reverse-engineering agent platform and MCP server for Ghidra, Frida, x64dbg and Rizin — automated PE/APK/binary analysis, CTF and malware researc…

"Never send a human to do a machine's job" - Open Source AI hacking agent

Prismer Cloud

My Personal Blog (Robotics)

Tau Coding Agent - like Pi, but twice as much

Open-source alternative to OpenAI Dots: self-hosted AI chat, tools, approvals, connectors, and computer tasks.