AI & ML

andersonlimahw Skill

Orchestration patterns for multi-agent squad coordination. Use when decomposing features into squad tasks, running standups, planning sprints, or managing cross-functional dependencies.

aribajahan Skill

Turn a private skill (or a set of them) into a clean, publicly shareable copy without leaking anything. Triage what can be shared, strip personal info, paths, names and credentials, rewrite fresh where the original is too personal, keep the judgment and the curated source lists that make it useful, guardrail it against prompt injection, verify, and stage a repo. Use when you want to extract a skill from a private system and publish it, or add one to an existing public skills repo.

andersonlimahw Skill

Create global or project-local agent skills with concise SKILL.md files, progressive disclosure, and optional deterministic scripts. Use when the user wants new skills for a project, workspace, domain, or global AI setup.

andersonlimahw Skill

Principal-engineer adversarial review for AI workspace setup changes, agent/skill topology, AGENTS.md/CLAUDE.md contracts, and plugin marketplace registration. Use when reviewing a PR, diff, or setup plan for overengineering, missing verification, broken orchestration, stale instructions, or weak agent boundaries.

john-walks-slow Skill

编写交接文档，将工作上下文传递给另一个 agent、同事、检视者或 oncall。可选参数：目标受众或方向（例如 "to reviewer"、"给 QA"、"for oncall"）。

cmevietnam Bundle

Use when the user asks to run Codex CLI (codex exec, codex resume, codex fork, codex review) or references OpenAI Codex for code analysis, code review, refactoring, or automated editing. Runs OpenAI's Codex agent non-interactively; the default model comes from ~/.codex/config.toml (currently gpt-5.6-sol), with gpt-6-astra available for the hardest work.

cmevietnam Bundle

Chạy review đối kháng bằng nhiều model Claude độc lập, tái hiện từng finding trước khi tin, và biến input tấn công của reviewer thành test case. Dùng khi cần soát kỹ code quan trọng (bảo mật, xử lý secret, parser, guardrail), khi một reviewer vừa trả kết quả, hoặc khi một sub-agent im lặng và cần biết nó còn sống không.

cmevietnam Bundle

Use when the user asks to run Antigravity CLI or `agy` for code review, a second opinion, code analysis, or a headless agent run — "ask antigravity", "agy review this", "have gemini flash look at it", "second opinion on this diff". Google's Antigravity CLI, which replaced Gemini CLI for personal accounts; serves Gemini Flash/Pro, Claude and GPT-OSS models. Reports findings verbatim before any code changes.

striderkein Skill

Run a multi-model adversarial mutual review with facilitator synthesis — the "Fable 5-style" metacognition pattern, reproducible on Opus or any Claude model: 2+ panelists (different model families where possible) independently answer, adversarially critique each other across rounds, and the main session acts as facilitator, synthesizing agreements, live disagreements, and calibrated confidence into one answer. Use this whenever the user asks to debate or adversarially review a claim, hypothesis, design, decision, or answer with multiple models or perspectives; says things like "discuss with Opus and Codex", "get a second opinion", "red-team this", "devil's advocate", 敵対的レビュー, 相互レビュー, 多モデル討論, 合議, 議論させて; or wants a high-stakes answer with explicit uncertainty instead of a single-pass response. Also use proactively when the user challenges your confidence ("are you sure?") on something consequential.

striderkein Skill

日本語の技術文書・書籍原稿の文章規範。整形（一文一行、引用ブロック、脚注、コラム記法）、段落と論証の構成（パラグラフライティング）、論証の厳密さ（ツッコミどころの除去）、読み手の負荷の管理、視点と語り、演出の抑制、LLM っぽい空句の禁止、冗長の排除を定める。日本語で技術書の章、草稿、記事、解説文を書くとき、または推敲・リライトするときに使用する。

aribajahan Skill

Finds the parts of a personal skill or automation system that broke quietly: skills pointing at renamed folders, scheduled tasks writing to paths that no longer exist, two files claiming to be the same source of truth, credentials sitting in prompt or skill files, and skills nothing calls. Reports only, never repairs.

elabalice-code Bundle

使用内置单文件 CLI 扫描 Godot/GDScript 源码树中的脚本和嵌套类成员，标记生产者/消费者角色，并从 if/elif/else/match 缩进控制域生成标准信号方程。适用于 Godot 项目信号地图、脚本成员读写审计、控制依赖梳理和 AI agent 自动化验证；不要用于 C#、Rust、TypeScript、运行时节点实例或函数局部变量分析。

elabalice-code Bundle

使用内置单文件 CLI 扫描 C# 源码树中的类型级字段，标记 Producer、Consumer、Producer+Consumer、Unused，并从 if/else/switch 控制域生成标准信号方程。适用于 C# 项目的信号地图建立、字段生产消费审计、fork 同名文件区分、自动化回归和 AI agent 快速静态验证；不要用于 GDScript、Rust、TypeScript、属性或函数局部变量分析。

aungminthu1722 Bundle

Creates, validates, and packages Agent Skills in the SKILL.md open format used by Claude Code, Claude apps, Codex, and other compatible agents. Use when the user wants to build a new skill, scaffold a skill directory, write or fix SKILL.md frontmatter and descriptions, review an existing skill against the specification, or bundle a skill into a distributable archive.

keinosuke Skill

Format review, audit, and investigation findings as a two-layer report that a non-engineer can act on and a coding agent can execute from. Use for code review, pull request review, security audit, design review, spec or instruction-document review, and research write-ups. Triggers on "review this", "review the PR", "audit this", "check this code", "summarize the findings", "write up the review", "look over this spec". 日本語でも起動する: 「レビューして」「PR を見て」「監査して」「指摘をまとめて」「レビュー結果を書いて」「指示書を確認して」。Produces output that can be pasted directly into a GitHub PR comment.

gtbwpkwjnb-alt Bundle

面向中文用户，提供技能库翻译精炼、项目画像和基于证据的技能/插件推荐操作。用于“技能审查”“技能翻译精炼”“项目画像”“技能推荐”；结合来源、版本、使用证据、冲突与 MCP/Agent 状态给出优化步骤。默认只读，不执行安装、卸载、配置或发布。

ufcalendar Bundle

Use when you need real MMA or bare-knuckle fight data — the UFCalendar Fight API, also known as the UFC API, the MMA API, the UFC MCP server or the MMA MCP server. Covers "when is the next UFC event", fight schedule and full fight cards, fight results JSON, per-round stats, judges scorecards, UFC rankings history since 2013, fighter careers and records, per-country broadcast rights, and signed webhooks. Also use when asked to build an MMA app or fight calendar, to pick a plan or get an API key, or to connect the UFCalendar MCP server — a UFC MCP server and MMA MCP server in one — to Claude, ChatGPT, Cursor or Codex.

yourpovv Skill

Universal operational harness for any review, security pass, or audit prompt. Always load before running prompts/reviews/* or prompts/domains/* or a full PR review. Defines scope, mode (report vs fix), exclusions, severity, verification, and output shape. Stack-agnostic. Project-specific threat models live in prompts; this file is the safety contract that keeps agents from thrashing an entire monorepo.

yourpovv Skill

Commit discipline for agent-assisted work: one commit per completed task, staging only that task's files, conventional message types, and never pushing without being asked. Use when an agent is about to commit, when a session has produced a pile of unstaged changes across several tasks, or when deciding who owns commits between a parent session and its sub-agents. Trigger on "commit this", "should I commit", "write a commit message", "squash these", "git add", or any task finishing with changes on disk. Harness-agnostic.

yourpovv Skill

Use when delegating work to sub-agents, or deciding whether to delegate at all. Covers the model tier table (which class of model gets which class of work), the rule that every delegated call sets its model explicitly, cheapest-first escalation, and the cases where delegation costs more than it saves. Also covers how to write a brief an agent can act on, who owns commits, and verifying delegated work before accepting it. Trigger on "spawn an agent", "delegate this", "run these in parallel", "use a subagent", "fan this out", "which model should this run on", "brief a subagent", "should the agent commit", or any plan that hands work to more than one agent. Harness-agnostic.

yourpovv Skill

Use when setting up or changing a project's Claude Code configuration: what belongs in CLAUDE.md versus CLAUDE.local.md, what each directory under .claude/ actually does, how hooks get registered, and what a minimal new project needs. Trigger on "set up .claude", "add a hook", "what goes in settings.json", "make a slash command", "add a project skill", "why isn't my hook firing", "CLAUDE.md vs CLAUDE.local.md", or scaffolding agent config for a new repo. Tooling reference, not a code standard.

yourpovv Skill

Where an agent session's tokens actually go, and which levers move the number. Covers the startup floor (everything always-loaded is billed on every single call), prompt caching as a prefix match and what silently invalidates it, why switching models mid-session is the most expensive habit there is, and why output is the costliest token you can spend. Use when a session feels expensive, when usage limits keep getting hit, when cache hit rate is low, or when deciding what belongs in always-loaded context. Trigger on "reduce token usage", "why is this so expensive", "hitting my limits", "prompt caching", "cache hit rate", "context is full", "should I compact", "trim CLAUDE.md". Harness-agnostic.

hdl-tools Skill

High-Level Synthesis — C/C++ algorithm analysis, HLS directive optimisation, synthesis execution, and co-simulation verification. Use when converting C/C++ to synthesisable RTL, optimising for latency/throughput/area targets using pragmas, or verifying that generated RTL matches the golden C model.

hdl-tools Skill

EDA tool detection, wrapper deployment, and MCP configuration for digital chip design environments. Use when setting up a new workstation, verifying tool availability before a domain flow, or generating per-tool install scripts with TCL modulefiles.

fightingforsidewalk Bundle

The operating model for running one project across several specialised Claude chats — named roles with one writer per file, a handoff contract, gate-by-risk, documentation by enumeration, batched writes, and the "Needs you" rule that keeps the human's asks from being buried. Use it when setting up a multi-chat project, writing or revising a project's instructions file, deciding which chat owns a document, or whenever a chat is about to produce another chat's artifact.

fightingforsidewalk Skill

Pre-ship checklist for any script, tool, or repo the agent writes or packages, run before code leaves the session and before the handback is written: path traversal, command injection, secrets and names, placeholders, .gitignore, license, atomic writes, and proof that every guard has been seen to fail. Use it whenever a session is about to hand back a change that ships executable code, and name the sections that did not apply.

fightingforsidewalk Skill

How a planning chat works with a coding agent (Claude Code or similar) so the human never has to evaluate code — writing a task as a "box" with a key-or-no-key line, scope as actions, docs by filename, observable acceptance criteria and hard stops; installing the CLAUDE.md standing behaviours and handback contract in a repo; and reading the handback that comes back as a plain-English checklist. Use it when writing a task for a coding agent, when a session's output comes back for review, or when setting up a new repo's CLAUDE.md.

hawklithm Bundle

Connects a self-hosted local MCP server (DevSpace) to the ChatGPT web UI end to end: environment self-check and dependency install (Node/npm/Git/Bash/Tailscale, auto-installing what is missing), DevSpace setup, Tailscale Funnel public tunnel, config writing, then creating a custom connector in ChatGPT and completing OAuth. Stage 6 (connector + OAuth) is driven automatically by the agent through a browser harness. Supports Windows / macOS / Linux; Windows and macOS verified on real hardware, Linux unverified. Includes guardrails: refuses to write corrupt config, auto backup and rollback, atomic writes, timeout retries, temporary-tunnel fallback, graded error reporting. Use to diagnose 'does not implement OAuth', 'Something went wrong', invalid_client, 'path is outside allowed roots', or a bash/shell tool that fails on every command. 把本地自托管 MCP 服务器（DevSpace）接入 ChatGPT 网页版的全流程技能。

planexeorg Skill

Use when the user wants Monte Carlo simulation of a PlanExe model — sampling from bounds to produce output distributions (mean/std/percentiles), threshold pass probabilities, and Pearson-correlation sensitivity rankings — given an extract-parameters-from-full JSON, a generate-bounds JSON, a generate-calculations Python module, and optional run settings

planexeorg Bundle

Use when the user wants to compute deterministic low/base/high scenario outputs for a PlanExe model — given an extract-parameters-from-full JSON, a generate-bounds JSON, and a generate-calculations Python module — producing a scenario result JSON with inputs, outputs, comparison spread, and warnings

jiangeplus Bundle

Company fundamentals and earnings analysis. Produces one-page tearsheets, pre-earnings previews (consensus, scenarios, catalysts), post-earnings deep dives (beat/miss, estimate revisions, thesis impact), and model updates after new actuals or guidance. TRIGGER when: the user asks for a company tear-sheet or overview, earnings analysis/preview, quarterly results interpretation, plugging earnings into a model, or refreshing estimates. DO NOT TRIGGER for: valuation-only requests (use valuation), report writing (use research-report), or position sizing (use risk-monitoring). 触发词：公司速览、tearsheet、财报分析、业绩前瞻、季报解读、业绩点评、模型更新、更新预测。

royalmilkteamaster Bundle

供執行開發任務的 Claude 或 Codex 子 Agent 按需載入。遇到非預期錯誤、錯誤行為、不穩定測試或效能退化時，讀取既有 Spec、planning 文件、CONTEXT.md 與 ADR，重現問題、驗證根因、在正確接縫建立回歸測試或記錄替代驗證、完成最小修正，再繼續原任務。不處理 TDD 刻意產生的 Red 或尚未完成造成的預期失敗。

royalmilkteamaster Bundle

快速設定目前專案共用的執行環境，以及 Developer、Reviewer A、Reviewer B 的 Agent、CLI、模型與推理強度。設定固定寫入專案的 `.milktea/agent-settings.yaml`，供之後所有 Implement 與 Brownfield Implement Task 直接使用；也可檢查或安裝 Reviewer B 使用的 Open Code Review。不派工、不實作。

royalmilkteamaster Bundle

供執行 Git 整合的臨時開發 Agent 按需載入。當 merge、rebase 或 cherry-pick 回報 CONFLICT 或 unmerged paths 時，依 Spec、Ticket、Commit、ADR 與測試理解雙方意圖，完成合併與驗證；平時不預載。

ckodex-labs Bundle

Skills Transparency Exchange v1 — symmetric publisher/consumer protocol for exposing and querying CKODEX skill-lifecycle runtime state. Use when designing, implementing, or auditing how a harness publishes skill pool state (tier, score, evidence) or how a client queries it. Covers JSON-LD domain model, JSON Schema 2020-12 contracts, OpenAPI 3.1 REST, gRPC proto3, and GraphQL SDL — all three transports normative.

ckodex-labs Bundle

Scaffold new CKODEX v1.1 skills from a blank slate, and migrate existing Agent Skills v1 skills to v1.1 in place. Use whenever the user asks to "create a new skill", "start a skill", "bootstrap a skill", "upgrade a skill to v1.1", "add synopsis/ontology/runtime to a skill", "migrate v1 → v1.1", or asks how to write a CKODEX manifest. Refuses to operate on directories that already contain conflicting files unless --force is set; emits explicit dry-run summaries before any write.