AI & ML

bybren-llc Bundle

Detect and repair drift between an OKF knowledge vault and the code it describes. Use after merging significant changes (schema, routes, workflows, agent config), when a staleness review is due, or when the user asks to sync, refresh, or update the knowledge vault or Obsidian vault.

futuregerald Skill

Model routing rules for codebase search. Delegates search/exploration to cheaper models (Haiku/Sonnet) while keeping Opus as the orchestrator. Invoke this skill before any codebase search or exploration task.

wojons Bundle

Enterprise-grade AI-driven development workflow engine with steering packet architecture for real-time control, plugin system, agent orchestration, monitoring, and complex workflow patterns. Build-verify loops that iterate until success.

bybren-llc Bundle

Agent assignment matrix, blocker escalation, and TDM coordination patterns. Use when assigning work to specialists, managing blockers, or coordinating multi-agent workflows.

bybren-llc Bundle

Agentic orchestration patterns for long-running tasks. Implements evidence-based delivery and Simon Willison's agent loop. Use when managing multi-step work, coordinating subagents, or orchestrating PR workflows.

peterfile Bundle

Generate and revise production-oriented 2D game maps with built-in image generation as the default visual asset source, choosing a visual model, runtime object model, collision model, art direction, and engine/export target. Use when Codex needs to create or integrate RPG maps, monster-taming maps, tactical arenas, battle backgrounds, side-scroller/parallax scenes, tilemaps, layered raster maps, clean HD hand-painted maps, pixel-inspired maps, prop packs, collision zones, walkable areas, or map previews.

citypaul Bundle

Re-explain the immediately previous LLM response when it did not land, using plain, precise UK English.

peterfile Bundle

Delegate coding, review, research, and validation work to Google Antigravity CLI (`agy`) and its subagents.

peterfile Skill

Use when operating Hermes Kanban workflows as an orchestrator or worker, including decomposition, lane routing, lifecycle ownership, and reconciliation.

peterfile Bundle

Use when operating a best-practice Project Canvas OS: project documents as the Agent knowledge layer, README.md as project contract, Obsidian Project.canvas as the human cognition map, Evidence cards as truth handles, and a zero-dependency CLI as the safe mutation/audit layer. Trigger on Project Canvas OS, Project.canvas, Obsidian Canvas project map, documentation hygiene, Heptabase-style project map, evidence-driven project state, or requests to reduce human understanding cost without losing project docs.

peterfile Skill

Use when drafting, rewriting, or launching prompts for OpenAI Codex/Codex CLI. Converts a user request into an outcome-first, scoped, verifiable Codex prompt with explicit constraints, validation, and stopping conditions.

peterfile Bundle

Use when delegating coding work to Codex CLI, Claude Code, OpenCode, Hermes subagents, or a Kanban implementation lane.

peterfile Bundle

Use when gathering, monitoring, or synthesizing external information from arXiv, feeds, YouTube, market data, LLM wikis, or academic paper workflows.

peterfile Bundle

Coordinate dense multi-agent software delivery with isolated worktrees, scoreable lanes, stacked PRs, fast validation, and truthful Slack/Linear/GitHub closeout.

citypaul Bundle

Discover and, with authorization, install agent skills from the open skills ecosystem. Use when the user explicitly asks to find or install a skill from the external skills ecosystem, asks whether an installable agent skill exists for a task, or wants to extend agent capabilities through skills.sh or a skill repository. Not for choosing among already-installed local skills or selecting software libraries, developer tools, applications, services, frameworks, or platform primitives; use evaluate-existing-solutions for technology choices.

citypaul Bundle

Get a rigorous read-only second opinion on finished work through the host's available reviewer capabilities, preferably from a different model provider. Falls back to a fresh same-provider agent only when necessary and labels the reduced independence. Every review includes a mandatory scope-fidelity check against the original requirements — unrequested additions, unrequested removals, and deleted or weakened tests. Use when asked to double-check, verify, cross-check, or get a second opinion, and before shipping high-stakes or complex work.

citypaul Skill

Decide where a learning, gotcha, or decision should live so it is not lost, and capture it there while context is fresh. Use when the user says they just discovered, learned, or found out something non-obvious, asks where a fact should be recorded, written down, or go, wants a setup gotcha, workaround, or decision kept for the next person or the next agent session, or asks whether something belongs in CLAUDE.md, a README, an ADR, tests, a glossary, or a reusable skill. Routes each fact to its durable owner and decides whether it is reusable or project-local. For the prose of the document itself see technical-writing; for recording a domain term see ubiquitous-language.

citypaul Bundle

Composable multi-agent code review of any change boundary — the working tree mid-development, a branch diff, a stacked layer, or a pull request: one orchestrator fans out sub-agents, each loading exactly one installed skill as a review lens (hexagonal-architecture, domain-driven-design, structure-codebase, typescript-strict, ...), then adversarially verifies findings and synthesizes one ranked report. Invoked as /panel-review with an optional target (defaults to the current branch's work, committed or not) and lens names — sensible defaults plus project-trait auto-detection, add or remove lenses freely. Also owns the PR-readiness evidence gate (change-path classification, mutation-evidence freshness) for when a boundary is heading to a PR. Use when reviewing in-progress work, a branch, a diff, or a PR, before merging, or when the user asks to review through named skills. Built on graph-engineering (the generic machinery); for a cross-provider second opinion on finished work use double-check; for Anthropic's f

citypaul Bundle

Turn fuzzy intent into shared understanding and acceptance criteria — specification as a conversation, run one question at a time, before any story is split or planned. Use when an idea, feature request, or problem statement has no agreed rules or examples yet ("let's spec this out", "what should this actually do?", "we need acceptance criteria"). Produces an example map and acceptance criteria written back into the team's own story artifact, plus candidate glossary terms and parked questions. The agent-facilitated round is a draft for review by the accountable owner and, when risk or shared ownership warrants it, the relevant human perspectives. For decision trees with no artifact, use grill-me where installed or this skill's one-question protocol; for tightening an existing artifact, see find-gaps; for slicing agreed work, see story-splitting.

citypaul Bundle

Event sourcing patterns for functional TypeScript — persist state as an append-only log of past events and rebuild it by folding them. Use when implementing a Decider write model, an event store, projections and read models, event versioning, or snapshots. Builds on the Decider from domain-driven-design and the ports/adapters from hexagonal-architecture. Do NOT reach for it on CRUD, or when an audit log or the outbox pattern already suffices.

wojons Bundle

Build and orchestrate AI-driven development workflows (Ralph loops) that iterate until success using configurable patterns and multi-agent coordination

bahayonghang Bundle

Use when the user explicitly asks the current agent to involve Codex CLI by reviewing a plan, implementing code, revising an implementation after review, or verifying extrapolated findings. Supports "让 Codex 审一下", "交给 Codex 实现", and explicit Claude-to-Codex collaboration through portable file-backed bundles. Do not use for ordinary code review, generic multi-agent orchestration, session handoff, or any request that does not explicitly ask Codex to participate.

bahayonghang Bundle

Use when the user asks an Agent inside Herdr to start Codex for an independent review through herdr-orchestra, or invokes codex-review with established Herdr task context; 支持在 Herdr 中让 Codex 审查改动、提交、文件或计划。Do not use for generic review, direct non-Herdr Codex collaboration, automatic fixes or CHANGELOG generation, research-only discussion, or when already assigned as the leaf reviewer.

bahayonghang Bundle

Safely orchestrate Conventional Commits for staged Git changes, or all working-tree changes when the user explicitly asks to include everything. Use when asked to write a commit message, split staged or working-tree changes, organize a messy index, or draft commit text without pushing. Auto-detects commit language; explicit phrases like 用中文提交 or 请使用中文拆分提交所有的改动 override. Default headers have no [AI] tag; add it only when asked (加 AI 标记 / add [AI] tag / 不要 [AI]). Create commits with local git commit only — never GitHub web, GitHub file-commit APIs, or a platform GitHub App — so GitHub does not show "This commit was created on GitHub.com" or "Committed via Cursor Agent". 本地 git 提交; 不要在 GitHub 上创建 commit; 不要 Committed via. Handles the local commit portion of mixed requests, then hands already-authorized push, PR, or other remaining operations back to the main coordinator for the appropriate workflow.

citypaul Bundle

Compose installed agent skills into one orchestrated multi-agent run: an agent graph where each node is a sub-agent that loads exactly one skill, edges carry structured data, and stages fan work out, adversarially verify it, and synthesize one deliverable. Claude Code calls the executable form a dynamic workflow; the pattern is Anthropic's orchestrator-workers. Use when a task needs several skills applied in parallel (multi-lens review, multi-angle audit or research, migration sweeps), when the user says graph engineering, agent graph, fan out sub-agents, or orchestrate skills, or when designing any workflow where sub-agents each own one skill. For PR review specifically use panel-review (an instance of this skill); for a single second opinion on finished work use double-check; do not use for tasks one skill handles in one context.

citypaul Bundle

Writing developer-facing prose that can be skimmed first and trusted enough to finish — READMEs, guides, tutorials, reference docs, proposals, PR descriptions, release notes. Load before drafting, rewriting, restructuring, or editing any technical document, including a request to rewrite or improve a README, when a doc reads as a wall of text, when claims need receipts, or when docs must serve AI agents as well as humans. Covers reader-first structure, falsifiable claims, docs-as-behavior verification, and agent-readable reference shape. For deciding where a learning, gotcha, or decision should be recorded see expectations; for diagram choice and syntax see diagrams; for API reference semantics see api-design; for CLI help text see cli-design.

bahayonghang Bundle

Use when a coding agent needs to run Python code, modules, one-liners, tools, tests, or standalone scripts through uv, or create and maintain PEP 723 scripts with uv init/add/remove --script. Avoid direct python/python3 shell entrypoints.

bahayonghang Bundle

Turn vague or complex agent tasks into project-aware, verifiable `/goal` commands and optional approved root `GOAL.md` handoff contracts for Claude Code, Codex, Grok Build, Oh My Pi, and Kimi Code. Use for Goal 指令, 目标指令, `/goal` prompts, 中文 Goal 模板, goal 持久化/保存/落盘, fresh-Agent or 跨会话交接, plan-to-goal interviews, bounded agent work definitions, Trellis 任务实施, review or scan remediation in one Prompt, 扫描审阅报告驱动修复, 单 Prompt 闭环返修, commit-then-archive cadence, or 终稿展示. Do not use for ordinary one-line work, pure read-only review, direct implementation without Goal authoring, memory-vault creation, or active-goal management that only needs a platform command.

bahayonghang Bundle

Use when the user explicitly asks to coordinate worker agents in Herdr, delegate tasks across Herdr panes, or collect and compare their results; 支持 Herdr 多 Agent 编排、分工、交叉核验和结果收集。Do not use for ordinary local subagents, simple pane inspection, documentation research, work outside Herdr, or when already assigned as a leaf worker.

bahayonghang Bundle

Use only when the user explicitly asks for swarm, subagents, parallel agents, dynamic workflow, multi-agent orchestration, 多智能体编排, or when the task truly needs coordinated research plus implementation plus review plus verification packets. Not for ordinary code review, planning-only work, single-line bugfixes, or routine audits unless orchestration is requested.

bahayonghang Bundle

文献深度解读助手，像研究生导师一样交互式解读 Zotero 库中的学术论文（计算机/深度学习/自动化方向，个人向）。当用户提供文献题目、DOI、PDF 或要求解读某篇论文时触发；通过 Zotero MCP 优先获取全文，按用户意图自动选择快速筛选、导师深读或研究复盘模式。多篇论文综合、对比、找研究空白或批量规范化时改用 paper-workbench。

kmshdev Skill

Use when the user needs Xcode project inspection, Apple API docs, compiler diagnostics, build logs, preview rendering, snippet execution, or targeted test runs through the official Xcode MCP bridge.

kmshdev Bundle

Non-glass SwiftUI APIs from WWDC 2025 (iOS 26, macOS 26, visionOS 26). Covers @Animatable macro, TextEditor with AttributedString/AttributedTextSelection/AttributedTextFormattingDefinition, FindContext, WebView/WebPage, UIHostingSceneDelegate, ToolbarSpacer, Slider ticks, windowResizeAnchor, dragContainer, draggable(containerItemID:), scrollEdgeEffectStyle, tabBarMinimizeBehavior. Use when building rich text editors, embedding web content, bridging UIKit scenes to SwiftUI, or configuring scroll edge effects and tab bar minimization. DO NOT use for Liquid Glass design patterns (use apple-liquid-glass-design), general Swift or pre-iOS 26 SwiftUI (use swiftui-ui-patterns).

bahayonghang Bundle

Analyze how an existing agent skill was used in past Claude Code, Grok, Codex, and Oh My Pi conversations. Use when the user asks to 分析某 skill 的使用情况, 这个 skill 用着有什么问题, 根据历史对话反馈改进 this skill, review skill sessions, or write a session-review report. Writes reports/skill-session-review and a copyable qiaomu-meta handoff prompt. Do not use for skill usage counts, unused-skill cleanup, resuming a session, optimizing CLAUDE.md, creating a new skill, or applying qiaomu-meta edits.

bahayonghang Bundle

Audit a repository and current Codex capabilities, then recommend the smallest evidence-backed read-only improvement or no change. Use for Codex setup optimization, surface selection, unapplied MCP/plugin/subagent plans, 优化 Codex 工作流, 审阅 Codex 配置, Codex 能力推荐. Exclude direct AGENTS/code-map edits, docs questions, skill audits, dynamic workflow implementation, code review, and any install/config/write.

bahayonghang Bundle

Audit and safely clean Windows dev-process trees and UWP app background-task pileups, including orphan npm/npx, leaked Playwright MCP workers, workspace dev servers, IDE services, Phone Link, Dolby Access, and backgroundTaskHost.exe. Use for Task Manager node/npm noise, Windows process buildup, 进程堆积, 清理残留开发进程, 泄漏的 Playwright MCP, 手机连接或杜比后台任务堆积. Do not use for generic performance profiling, malware response, Windows service management, app uninstall, or non-Windows cleanup.