agentsclimarketplace

Context & AI engineering

25,610 rows, by stacks then stars

1,865 of these skill files have been read, by 1,328 distinct authors, and what they tell an agent counted. What Context & AI engineering authors agree on, and what they forbid

  • Stayingapi mcp agent scouting

    stayingapi/travel-skills/skills/stayingapi-mcp-agent-scouting Skill

    0★ repo

    Connect StayingAPI as a Claude/Cursor MCP connector and scout stays, compare prices and check availability conversationally — no keys pasted. Powered by the StayingAPI REST API / MCP server.

  • Stayingapi mcp

    stayingapi/travel-skills/skills/stayingapi-mcp Skill

    0★ repo

    Connect an AI agent to the StayingAPI Model Context Protocol (MCP) server for accommodation search, availability, pricing, cross-OTA price comparison and reviews. Use when the agent runtime supports MCP connectors (e.g. Claude) and you want OAuth-based access without managing API keys. Covers the OAuth 2.1 + PKCE connect flow, the seven read-only tools, and when to prefer MCP vs REST.

  • Stayingapi new inventory lead gen

    stayingapi/travel-skills/skills/stayingapi-new-inventory-lead-gen Skill

    0★ repo

    Re-run a saved search on a schedule, diff it against the last run, and sync only the newly-listed properties into a CRM or Sheet as leads. Powered by the StayingAPI REST API / MCP server.

  • Stayingapi portfolio feed

    stayingapi/travel-skills/skills/stayingapi-portfolio-feed Skill

    0★ repo

    A daily digest across a saved set of listings — price, rating and recent reviews per listing in one report. Powered by the StayingAPI REST API / MCP server.

  • Stayingapi price drop watcher

    stayingapi/travel-skills/skills/stayingapi-price-drop-watcher Skill

    0★ repo

    Watch a specific stay and get alerted the moment its total price falls to or below your threshold. Powered by the StayingAPI REST API / MCP server.

  • Stayingapi rate parity monitor

    stayingapi/travel-skills/skills/stayingapi-rate-parity-monitor Skill

    0★ repo

    Watch how your own listing's rate shows across OTAs on a schedule and flag any platform underselling the market median. Powered by the StayingAPI REST API / MCP server.

  • Stayingapi review intelligence digest

    stayingapi/travel-skills/skills/stayingapi-review-intelligence-digest Skill

    0★ repo

    Pull normalized reviews for a property on a schedule and deliver a weekly digest — average rating, recurring themes, and the latest complaints to act on. Powered by the StayingAPI REST API / MCP server.

  • Stayingapi str market scan

    stayingapi/travel-skills/skills/stayingapi-str-market-scan Skill

    0★ repo

    Scan a short-term-rental market — discover listings across platforms, price the top N for your dates, and append a normalized row-per-listing dataset to a Google Sheet. Powered by the StayingAPI REST API / MCP server.

  • Agent guardrails

    jcdavis131/cursor-agent-skills/skills/agent-guardrails Skill

    0★ repo

    Give every autonomous agent three guardrails — a per-team allowlist (what it can do), bounded loops (can't run forever), and full transcripts (auditable). Use when building an autonomous agent, a recurring agent loop, or any agent that runs without per-step human approval.

  • Deterministic core llm judgment

    jcdavis131/cursor-agent-skills/skills/deterministic-core-llm-judgment Skill

    0★ repo

    Build agents where the deterministic duties are the load-bearing backbone (always execute, reliable, auditable) and the LLM only adds an optional judgment layer on top — ranking, curation, synthesis. The agent must do its core work with the LLM layer off; the LLM only makes the work better. Use when designing or building an autonomous agent, a recurring agent loop, or any system where reliability matters more than cleverness.

  • Lane discipline

    jcdavis131/cursor-agent-skills/skills/lane-discipline Skill

    0★ repo

    In a multi-agent fleet, verify a piece of work is in your designated lane before doing it, and state the check out loud. Defer to the owning agent when the work belongs to another lane. Use whenever the repo has documented agent/team role boundaries (AGENTS.md, .cursor/rules/, .claude/TEAM.md, fleet-team rules) and the task touches a shared area.

  • Prompt caching

    jcdavis131/cursor-agent-skills/skills/prompt-caching Skill

    0★ repo

    Designs and wires Anthropic prompt caching for Messages API calls — automatic vs explicit breakpoints, prefix ordering, TTL, minimum token thresholds, and cache-invalidation traps. Use when building Claude API integrations, agent loops, multi-turn chats, large tool/system prompts, or when the user mentions prompt caching, cache_control, or cache_read_input_tokens.

  • Verify subagent output

    jcdavis131/cursor-agent-skills/skills/verify-subagent-output Skill

    0★ repo

    After parallel subagents (SME reviewers, research agents, task agents) return, spot-check their output against the named contract/standard before synthesizing or marking the task done. Use whenever you dispatched subagents and they've returned artifacts — especially before treating their output as ground truth in a synthesis or a "done" report.

  • Openai agentkit

    frankxai/skills/skills/frameworks/openai-agentkit Skill

    0★ repo

    Build production-ready multi-agent systems using OpenAI AgentKit and Agents SDK with best practices for agent orchestration, handoffs, and routines

  • Mcp 2025 patterns

    frankxai/skills/skills/mcp/mcp-2025-patterns Skill

    0★ repo

    Current best practices for Model Context Protocol server design, implementation, and integration. Updated patterns for 2025 MCP ecosystem including multi-server orchestration, security, and performance.

  • Mcp architecture

    frankxai/skills/skills/mcp/mcp-architecture Skill

    0★ repo

    Design and implement Model Context Protocol servers for standardized AI-to-data integration with resources, tools, prompts, and security best practices

  • Agentic orchestration

    frankxai/skills/skills/orchestration/agentic-orchestration Skill

    0★ repo

    Patterns for multi-agent coordination, task decomposition, handoffs, and workflow orchestration. Best practices for building and managing agent systems.

  • Agent design review

    frankxai/skills/skills/reviews/agent-design-review Skill

    0★ repo

    Use when the user asks to review, grill, pressure-test, or sanity-check an AI agent or multi-agent system design before building it — triggers include "grill me on my design", "design review", "review my agent architecture", "is this architecture sound", "pressure-test this before I build". Interrogates the design against the four failure modes that kill enterprise agent initiatives and returns a go / no-go verdict with ranked fixes.

  • Codetruss

    DeliriumPulse/codetruss-plugins/skills/codetruss Skill

    0★ repo

    Operate CodeTruss local acceptance gates for coding-agent changes. Use when a developer asks to bind an agent task to allowed or denied files, configure repository verification, install or diagnose Claude Code or Codex hooks, review a working-tree or staged diff before commit, interpret or verify a signed CodeTruss receipt, repair a failed verdict, or explicitly opt into provider-backed review or receipt sync.

  • Stage gate

    k97/skills/skills/stage-gate Skill

    0★ repo

    Quality gates for the three checkpoints of a feature cycle: after planning (--plan gates the plan for unstated assumptions and unverifiable success criteria), after development (--dev reviews the diff for overcomplication and non-surgical changes), and after merge or release (--release captures lessons into rules and keeps the agent memory file lean via progressive disclosure). Agent-agnostic: targets CLAUDE.md, AGENTS.md, GEMINI.md, or Copilot instructions — Claude Code, Codex, Cursor, Copilot, Gemini CLI, Antigravity. --release also bootstraps a new repo: creates the memory file if missing, prunes init-generated bloat. Use for "review my plan", "gate this diff", "is this overcomplicated", "post-merge hygiene", "capture lessons", "CLAUDE.md is too long", "AGENTS.md is too long", "set up CLAUDE.md for this repo". Scope with --plan, --dev, --release.

  • Docsbot administration

    uglyrobot/docsbot-agent-skills/skills/docsbot-administration Skill

    0★ repo

    Use DocsBot Administration to administer DocsBot teams, bots, sources, members, integrations, Skills, reporting, and supported billing settings through an OAuth-authenticated remote MCP server. Activate when a user asks to manage DocsBot, inspect DocsBot account state, configure bots or sources, review dashboard data, or use the DocsBot Admin API through an agent.

  • Bot builder

    uglyrobot/docsbot-agent-skills/skills/docsbot-administration/references/bot-builder Skill

    0★ repo

    Use when creating, configuring, tuning, testing, or handing off a DocsBot bot through the DocsBot Admin MCP. Covers discovery, branding, bot creation, source selection, source tags, prompts, widget/embed/helpdesk/Slack deployment choices, lead capture, escalation, Skills, external MCP connectors, and final dashboard/share links.

  • Docsbot documentation search

    uglyrobot/docsbot-agent-skills/skills/docsbot-documentation-search Skill

    0★ repo

    Search and fetch a DocsBot bot's indexed documentation, knowledge base, and training sources through a per-bot remote MCP server. Activate when a user asks to use an agent to answer from, inspect, retrieve, or cite content already indexed in a specific DocsBot bot.

  • Docsbot question history

    uglyrobot/docsbot-agent-skills/skills/docsbot-question-history Skill

    0★ repo

    Search and fetch prior DocsBot bot questions, answers, and conversation history through a per-bot remote MCP server. Activate when a user asks to analyze support history, find past answers, review escalations, inspect unresolved questions, or reuse previous DocsBot conversations.

  • Vrbo search

    stayingapi/vrbo-skills/clawhub/vrbo-search Skill

    0★ repo

    Search live Vrbo stays by location, dates and occupancy, returned in one unified schema alongside every other booking platform. Use when a user wants to find Vrbo listings. Powered by the StayingAPI REST API / MCP server.

  • Vrbo search

    stayingapi/vrbo-skills/skills/vrbo-search Skill

    0★ repo

    Search live Vrbo stays by location, dates and occupancy, returned in one unified schema alongside every other booking platform. Use when a user wants to find Vrbo listings. Powered by the StayingAPI REST API / MCP server.

  • Google hotels search

    stayingapi/google-hotels-skills/clawhub/google-hotels-search Skill

    0★ repo

    Search live Google Hotels stays by location, dates and occupancy, returned in one unified schema alongside every other booking platform. Use when a user wants to find Google Hotels listings. Powered by the StayingAPI REST API / MCP server.

  • Google hotels search

    stayingapi/google-hotels-skills/skills/google-hotels-search Skill

    0★ repo

    Search live Google Hotels stays by location, dates and occupancy, returned in one unified schema alongside every other booking platform. Use when a user wants to find Google Hotels listings. Powered by the StayingAPI REST API / MCP server.

  • Website agent readiness

    paulgrape/skill-auditor/templates/website-agent-readiness Skill

    0★ repo

    Agent- and crawler-readiness for a website — stable URLs, structured data for agents, machine-readable formats, llms.txt, per-page Markdown endpoints, robots controls for AI crawlers, and well-known URIs. Use when making a site legible to AI agents and LLMs, or publishing machine-readable endpoints.

  • Claude harness refs update

    ystk-kai/claude-harness/skills/claude-harness-refs-update Skill

    no license0★ repo

    claude-harness の参照資料 (蒸留版 skills/*/references/*.md と原典 clone ~/.claude/references/ の 2 層) の鮮度チェックと更新を実行する。check-freshness → BEHIND を git pull → STALE を DISTILLING.md の手順で再蒸留。明示起動専用 (/claude-harness-refs-update で呼ぶ)。

  • Harness design

    ystk-kai/claude-harness/skills/harness-design Skill

    no license0★ repo

    LLM ハーネス・プロンプト設計の参照資料を 2 層 (このスキル内の蒸留版 references/*.md と、~/.claude/references/ の原典 clone) で使うためのスキル。プロジェクトのハーネス (CLAUDE.md / AGENTS.md / skills / slash commands / subagents / hooks / MCP 設定) を設計・作成・レビュー・監査するとき、プロンプトエンジニアリング (プロンプトの設計・改善・テクニック選定) を行うとき、エージェントの信頼性設計を検討するとき、UI 生成用の DESIGN.md を選定するときに、まず蒸留版を読み、索引が指す原典ファイルだけを深掘りする。Triggers: harness, ハーネス, CLAUDE.md, AGENTS.md, skill 作成, subagent, プロンプト改善, prompt engineering, agent design, best practice, DESIGN.md

  • Swarm plan

    grimoire-rs/grimoire-vscode/.claude/skills/swarm-plan Skill

    0★ repo

    Use for feature planning, task decomposition, multi-perspective research, or ADR scaffolding. Tier (`low | auto | high | max`) scales research depth, architect model, and review breadth. Canonical research primitive for AI config / ADR work.

  • Swarm review

    grimoire-rs/grimoire-vscode/.claude/skills/swarm-review Skill

    0★ repo

    Use for branch, PR, or diff review before landing on main. Tier (`low | auto | high | max`) scales breadth from single reviewer to adversarial panel with architect + SOTA-gap + Codex cross-model gate. Triggers: "review", "/swarm-review", pre-merge checks.

  • Code review graph

    SamyakJhaveri/loam/cultivation/marketplace/code-review-graph/skills/code-review-graph Skill

    0★ repo

    GraphRAG-powered code review with blast-radius analysis via Tree-sitter AST. MCP server. Recommended for 500+ file codebases.

  • Gpt researcher

    SamyakJhaveri/loam/cultivation/marketplace/gpt-researcher/skills/gpt-researcher Skill

    0★ repo

    Autonomous multi-source research agent with MCP server. Planner/executor architecture, 5 MCP tools. Requires 3 API keys (Tavily + Anthropic + embeddings). Research-flavor.

  • Align prompt

    SamyakJhaveri/loam/cultivation/marketplace/helpers/skills/align-prompt Skill

    0★ repo

    Use when you have a free-form draft prompt and want it rewritten into Claude Opus 4.6 or 4.8 conventions before pasting it into a new planning or execution session. Trigger phrases — "align prompt", "rewrite for opus", "tune prompt for claude". NOT for planning itself, code implementation, or validation. Pre-planning / pre-execution only.

  • Model route

    SamyakJhaveri/loam/cultivation/marketplace/helpers/skills/model-route Skill

    0★ repo

    Advisor for selecting optimal Claude model tier (Opus/Sonnet/Haiku) for a specific task. Use before launching an agent team, before a multi-file refactor, or when deciding whether to switch from Opus to Haiku for transactional work (commits, formatting). Analyzes reasoning depth, blast radius, domain expertise, output length, correctness cost.

  • Prompt improver

    SamyakJhaveri/loam/cultivation/marketplace/helpers/skills/prompt-improver Skill

    0★ repo

    Use when invoked automatically by the UserPromptSubmit hook after a prompt is flagged as vague. Researches context, then generates 1-6 targeted clarifying questions before executing. Do NOT invoke this manually — it is triggered by the hook in .claude/hooks/improve-prompt.py.

  • Storm research

    SamyakJhaveri/loam/cultivation/marketplace/storm-research/skills/storm-research Skill

    0★ repo

    Stanford STORM — LLM research pipeline for Wikipedia-quality articles with citations. Multi-perspective research, Co-STORM collaborative mode. Python package, no native Claude Code integration.

  • Agent team

    SamyakJhaveri/loam/cultivation/marketplace/team-deliberation/skills/agent-team Skill

    0★ repo

    Creates and launches coordinated agent teams using TeamCreate for multi-teammate tasks requiring cross-talk, shared task lists, and lifecycle management. Use when 2+ workers need to communicate findings to each other, not just report to parent. NOT for independent parallel tasks (use dispatching-parallel-agents instead).

  • Agent team

    SamyakJhaveri/loam/seed/.claude/skills/agent-team Skill

    0★ repo

    Creates and launches coordinated agent teams using TeamCreate for multi-teammate tasks requiring cross-talk, shared task lists, and lifecycle management. Use when 2+ workers need to communicate findings to each other, not just report to parent. NOT for independent parallel tasks (use dispatching-parallel-agents instead).

  • Align prompt

    SamyakJhaveri/loam/seed/.claude/skills/align-prompt Skill

    0★ repo

    Use when you have a free-form draft prompt and want it rewritten into Claude Opus 4.6 or 4.8 conventions before pasting it into a new planning or execution session. Trigger phrases — "align prompt", "rewrite for opus", "tune prompt for claude". NOT for planning itself, code implementation, or validation. Pre-planning / pre-execution only.

  • Critique swarm

    SamyakJhaveri/loam/seed/.claude/skills/critique-swarm Skill

    0★ repo

    Parallel adversarial critique with 4 specialized agents plus synthesis. Use before merging a feature branch, before committing a complex plan, or when you want independent multi-angle review of a plan or implementation. Complements /multi-review (which reviews code quality) by reviewing plan/scope quality. NOT for single-file trivial changes (use /validate), standard code review (use /multi-review), or post-implementation shipping (use /ship).

  • Handoff

    SamyakJhaveri/loam/seed/.claude/skills/handoff Skill

    0★ repo

    Use when ending a session with unfinished work, switching context, or preparing for another agent to continue. Write or update a handoff document so the next agent with fresh context can continue this work. NOT for: session notes, commit messages, or PR descriptions — only structured handoffs for agent continuity.

  • Plan review invoke

    SamyakJhaveri/loam/seed/.claude/skills/plan-review-invoke Skill

    0★ repo

    Invoke the plan-reviewer agent using the canonical reference prompt from seed/plan-reviewer-design.md. Use when reviewing a plan in a fresh Claude Code session (after the plan was created in a prior session) and you want adversarial review (checklist + Elegance Gate + handoff plan) without manually copy-pasting the reference prompt. Accepts the plan path as an argument. NOT for: same-session plan reviews (just invoke the agent directly with the plan in context), reviewing implementations that already shipped (use /multi-review or /session-critique), or spec audits (the agent has a separate Spec Audit mode invoked differently).

  • Template sync

    SamyakJhaveri/loam/seed/.claude/skills/template-sync Skill

    0★ repo

    Sync Claude Code assets between this project and the project-template buffer. Use when promoting a generally-useful agent/skill/hook/rule from this project back to the template, when pulling a template-side update down into this project, or when checking what has diverged. Subcommands - status, diff, pull, promote, trial, sync-from-buffer. Promotion is always opt-in - never automatic.

  • Eval grader

    SamyakJhaveri/loam/seed/_research/skills/eval-grader Skill

    0★ repo

    Evaluation result grading and classification workflow. Use when classifying eval results into pass/fail categories, diagnosing failure patterns in batch results, computing pass rates with proper exclusions, or preparing result tables for papers. Handles the grading pipeline: load results, apply exclusions, classify failure modes, compute statistics, generate summary tables.

  • Eval run

    SamyakJhaveri/loam/seed/_research/skills/eval-run Skill

    0★ repo

    Launch a model evaluation batch. Use when running interactive/foreground eval runs — parameter collection, pre-flight checks, execution, and post-run analysis. NOT for: long-running batches that need tmux isolation (use overnight-eval instead).

  • Overnight eval

    SamyakJhaveri/loam/seed/_research/skills/overnight-eval Skill

    0★ repo

    Long-running evaluation campaign with tmux isolation. Use when running any eval batch exceeding ~30 minutes, canonical benchmark runs, or when the user will be away during execution. Pre-flight verification, tmux launch, monitoring, post-flight analysis. NOT for: short interactive runs (use eval-run instead).

  • Post eval

    SamyakJhaveri/loam/seed/_research/skills/post-eval Skill

    0★ repo

    Post-batch analysis pipeline that runs after an eval batch completes. Use after /eval-run or /overnight-eval finishes. Verifies results, runs analysis scripts, refreshes dashboard, writes summary. Does NOT launch new eval runs.

  • Campaign planning

    ats4321/claude-engineering-skills/skills/campaign-planning Skill

    0★ repo

    Decompose and execute large multi-step engineering efforts — multi-file features, migrations, and multi-agent workstreams. Auto-load when a task spans many files/days/agents, when planning a migration or rollout, when coordinating parallel subagents or engineers, when a big effort has stalled half-done, or when delegating slices of work with expectations of unattended execution. Covers risk-first slicing, per-slice done-criteria, dependency ordering, checkpoints that keep the system working, parallelization rules, delegation contracts, and pre-planned failure recovery. NOT for single-slice tasks (just do them) and NOT for the design itself (system-design).

  • Evaluation frameworks

    ats4321/claude-engineering-skills/skills/evaluation-frameworks Skill

    0★ repo

    Measure whether an LLM system actually works. Auto-load when building or improving evals, golden sets, or benchmarks for an LLM feature; when comparing prompts, models, or parameters; when someone claims an AI change "seems better"; when calibrating an LLM-as-judge; or before shipping any prompt/model change to an LLM system. An eval is to an LLM system what a test is to code. NOT for writing conventional code tests (validation-and-testing) and NOT for fixing malformed outputs (llm-integration-reliability).

  • Llm integration reliability

    ats4321/claude-engineering-skills/skills/llm-integration-reliability Skill

    0★ repo

    Engineer reliable systems around unreliable LLM outputs. Auto-load when code calls an LLM (Ollama, Anthropic/Claude, OpenAI, any provider SDK), parses model output as JSON, builds agent/ReAct loops, implements LLM grading/review/judging, or when debugging malformed model responses, infinite retry loops, hallucinated facts, or flaky LLM-backed pipelines. Covers strict output contracts with enum validation, layered JSON extraction fallbacks, retry strike limits, ground-truth injection, generation/evaluation separation, measurement/interpretation split, tool-error-as-observation, iteration hard caps, fatal-vs-recoverable error taxonomy, and concurrency caps.

  • Llm system design

    ats4321/claude-engineering-skills/skills/llm-system-design Skill

    0★ repo

    Decide WHETHER and WHERE an LLM belongs in a system, and shape the pipeline around it. Auto-load when designing, architecting, or proposing any LLM-backed feature or system — "should we use an LLM for this", choosing between local (Ollama) and hosted (Anthropic/OpenAI) models, selecting model size, choosing pipeline shape (single call, chain, fan-out, agent loop), budgeting cost/latency per call, or designing the failure story for an AI feature. NOT for debugging malformed outputs or parsing (that is llm-integration-reliability) and NOT for building the agent loop itself (that is agent-engineering).

  • Prompt and context engineering

    ats4321/claude-engineering-skills/skills/prompt-and-context-engineering Skill

    0★ repo

    The craft of what goes into a model — prompt structure, few-shot selection, context budgets, and prompt versioning. Auto-load when writing, restructuring, or tuning any prompt or system prompt; when selecting few-shot examples; when deciding what context/evidence/retrieval to include or exclude from a model call; when a prompt has grown unwieldy or contradictory; or when chunking long inputs for a model. NOT for output parsing/retries (llm-integration-reliability), NOT for whether an LLM belongs at all (llm-system-design), and NOT for measuring prompt quality (evaluation-frameworks).

  • Context engineering

    kulapoo/opencode-agent-harness/.opencode/skills/context-engineering Skill

    0★ repo

    Optimizes agent context setup. Use when starting a new session, when agent output quality degrades, when switching between tasks, or when you need to configure rules files and context for a project.

  • Using agent skills

    kulapoo/opencode-agent-harness/.opencode/skills/using-agent-skills Skill

    0★ repo

    Discovers and invokes agent skills. Use when starting a session or when you need to discover which skill applies to the current task. This is the meta-skill that governs how all other skills are discovered and invoked.

  • Project commander zh

    ProfesseurHaipeng/codex-project-commander/skills/project-commander-zh Skill

    0★ repo

    在 Codex、Claude Code、OpenCode、Kimi Code、Hermes 等 Agent Skills 宿主中建立“总指挥”项目治理;在具备持久命名项目任务工具的 Codex 界面中进一步组织侧边栏员工窗口。支持 OpenAI、MiniMax、DeepSeek、豆包等兼容模型提供商,具备部门、Token 监管、任务账本、完成监听、连续派工、运行与交付模式、非攻击式 APP 验收及低打扰 API Key 处理。用户说“我的总指挥”“总指挥”“指挥官”“项目指挥官”“启动指挥官”或“启动总指挥”时使用。绝不能用子智能体冒充持久员工窗口,也不能把模型 API 冒充项目任务系统。

  • Project commander

    ProfesseurHaipeng/codex-project-commander/skills/project-commander Skill

    0★ repo

    Establish commander-led project governance in Agent Skills hosts including Codex, Claude Code, OpenCode, Kimi Code, and Hermes; on Codex surfaces with persistent named project-task tools, also organize sidebar employee windows. Supports compatible OpenAI, MiniMax, DeepSeek, Doubao, and other model providers with departments, Token Governance, a durable ledger, completion monitoring, continuous dispatch, delivery modes, non-adversarial app validation, and low-friction API-key handling. Use for “my project commander”, “project commander”, “commander”, “be the commander”, or “start commander”. Never substitute subagents for persistent employee windows or confuse a model API with a project-task system.

  • Emoting on 8x8

    srfinch17/peckworks-skills-laboratory/skills/emoting-on-8x8 Skill

    0★ repo

    Design legible expressions and custom animations for the ESP32-S3 8x8 LED matrix (Claude's expression channel, matrix_express / matrix_animate / the CANNED library in mcp_server/expressions.ts). Use whenever creating, redesigning, or debugging a glyph or animation that has to read on the physical 64-pixel panel, including when an expression "doesn't read," looks like one flat color, flickers, or freezes the board.