AI & ML

catcorner22 Skill

Use Hugging Face Hub via MCP server tools when the Huggingface-skills (or equivalent) MCP namespace is connected. Search models, datasets, Spaces, and papers; inspect repos; invoke Gradio Spaces. If MCP is disconnected, use the hf-cli skill instead.

catcorner22 Skill

Hugging Face CLI to estimate the required memory to load Safetensors or GGUF model weights for inference from the Hugging Face Hub. Use when you have a specific model and need its exact memory footprint. Complements `huggingface-best`, which works the other direction — given a device's memory budget, it narrows which models fit — and `huggingface-local-models` for local GGUF quantization choice.

catcorner22 Skill

Default guidance for building AI agents when no framework has been chosen, or when the target is Vercel-native. Use for generic requests to build, create, scaffold, design, architect, or implement an AI agent, agent app, tool-calling agent, durable agent, multi-agent system, or scheduled agent — where the request names no framework and the repo depends on none. Scope boundary: once a framework is named or already present, that pack owns the work — LangChain/LangGraph/Deep Agents to `ecosystem-primer`, Pydantic AI to `building-pydantic-ai-agents`, smolagents to `smolagents`; authoring the agent's prompt text to `prompt-optimizer`. Do not claim a request merely because it contains the word "agent".

catcorner22 Skill

Configure Plaud AutoFlow: automatic device-to-app transfer, auto-transcription, auto-summary with chosen template and LLM, and email delivery. Use when automating repetitive Plaud pipelines for lectures or recurring meetings. Scope boundary: one-off summary edits → `plaud-summary-templates`; export destinations → `plaud-export-integrate`.

catcorner22 Skill

Vercel Connect expert guidance — securely obtain scoped OAuth tokens for third-party services (Slack, GitHub, MCP servers, OAuth, Snowflake) on behalf of apps or users via Vercel OIDC. Use when wiring up third-party API access, connecting to MCP servers, sending Slack messages, accessing GitHub APIs, receiving webhook events from Slack/Linear/GitHub and forwarding them to your agents and apps, or building eve agent connections.

catcorner22 Skill

Hugging Face `smolagents`: choosing CodeAgent vs ToolCallingAgent, the model backends (InferenceClientModel, LiteLLMModel, TransformersModel, OpenAIModel), defining tools with `@tool` or a `Tool` subclass, the sandbox choice for the Python that CodeAgent executes, managed-agent hierarchies, and memory/step surgery. Use when the user names smolagents, CodeAgent, ToolCallingAgent, or InferenceClientModel, when code contains `from smolagents`, or when running the `smolagent` CLI. Scope boundary: this covers the smolagents library only — an agent-building request with no framework chosen belongs to `build-agents`/`eve`, Pydantic AI to `building-pydantic-ai-agents`, LangChain/LangGraph to that pack, and Hub API scripting to `huggingface-tool-builder`/`hf-cli`.

catcorner22 Skill

Corrects outdated LLM knowledge about the Vercel platform and introduces new products. Invoke manually when working on Vercel products, Fluid Compute, vercel.ts, or when platform facts may be stale.

catcorner22 Skill

Train and fine-tune transformer language models locally with the TRL CLI (SFT, DPO, GRPO, KTO, RLOO, Reward Model). For managed Hugging Face Jobs, use huggingface-llm-trainer and hf-cli instead.

catcorner22 Skill

Configure the `deepagents` harness: `create_deep_agent()` / `createDeepAgent()`, middleware selection, built-in filesystem and planning tools, the agent SKILL.md format it loads, and what the harness will and will not let you configure. Use when the user names Deep Agents or the code imports `deepagents`. Scope boundary: an agent-building request with no framework chosen belongs to `build-agents`/`eve`; raw graph code belongs to `langgraph-fundamentals`.

catcorner22 Skill

Router for the LangChain ecosystem: choose between LangChain, LangGraph, and Deep Agents, set the LANGSMITH_* env vars, find the right docs.langchain.com page, and hand off to the matching skill in this pack. Use when the user names LangChain, LangGraph, LangSmith, or Deep Agents, or when the repo already depends on `langchain*` / `langgraph*` / `deepagents` / `@langchain/*` and the layer choice is still open. Scope boundary: an agent-building request that has not chosen a framework is not this pack's — TypeScript/Vercel agents belong to `build-agents`/`eve`, and Python agents on another framework to `building-pydantic-ai-agents` or `smolagents`.

catcorner22 Bundle

Use Transformers.js to run state-of-the-art machine learning models directly in JavaScript/TypeScript. Supports NLP (text classification, translation, summarization), computer vision (image classification, object detection), audio (speech recognition, audio classification), and multimodal tasks. Works in browsers and server-side runtimes (Node.js, Bun, Deno) with WebGPU/WASM using pre-trained models from Hugging Face Hub.

catcorner22 Skill

Use when the user asks about finding the best, top, or recommended model for a task, wants to know what AI model to use, or wants to compare models by benchmark scores. Triggers on: "best model for X", "what model should I use for", "top models for [task]", "which model runs on my laptop/machine/device", "recommend a model for", "what LLM should I use for", "compare models for", "what's state of the art for", or any question about choosing an AI model for a specific use case. Always use this skill when the user wants model recommendations or comparisons, even if they don't explicitly mention HuggingFace or benchmarks.

catcorner22 Skill

Router for memory AI-transfer techniques: stratified sessions, tablebase cache, taxonomy retrieval, cross-session bridges. Use for long-horizon context and RAG design. Scope boundary: full catalog → `ai-transfer-ecosystem-primer`.

catcorner22 Bundle

Build Gradio web UIs and demos in Python. Use when creating or editing Gradio apps, components, event listeners, layouts, or chatbots.

catcorner22 Skill

Look up and read Hugging Face paper pages in markdown, and use the papers API for structured metadata such as authors, linked models/datasets/spaces, Github repo and project page. Use when the user shares a Hugging Face paper page URL, an arXiv URL or ID, or asks to summarize, explain, or analyze an AI research paper.

catcorner22 Skill

Middleware on a LangChain `create_agent` / `createAgent` agent: `HumanInTheLoopMiddleware` to approve dangerous tool calls, writing custom middleware hooks, `Command` resume patterns, and structured output with Pydantic/Zod. Use when the agent being extended is a LangChain agent. Scope boundary: structured output on the Vercel AI SDK (`Output.object`) belongs to `ai-sdk`; interrupts in a hand-written LangGraph graph belong to `langgraph-human-in-the-loop`.

catcorner22 Skill

Post-generation debrief: intended vs actual vs gap vs prescription. Log for recurring pattern fixes. Use after agent tasks complete. Scope boundary: five whys on hard failures → `five-whys-failure-recovery`.

catcorner22 Skill

Hallucination suppression via structural claim/evidence pairing: every factual claim must balance with a supporting source before output delivers. Use when verifying AI outputs, building RAG gates, or when the user asks for claim checking, evidence balance, or hallucination suppression. Scope boundary: human workspace prep → `workspace-mise-en-place`; source citation style → `citation-literacy`.

catcorner22 Skill

Layer memory by session strata plus disturbance markers; retrieve with integrity confidence. Use for long-horizon agents. Scope boundary: flat RAG → `library-taxonomy-retrieval`; cross-session bridges → `wildlife-corridor-bridging`.

catcorner22 Skill

Measure sentence-length tide (high/mid/low) and even the rhythm when variance is extreme. Use on long prose that feels rushed or swampy. Scope boundary — token budget cuts → `prompt-optimizer`; zoom depth → `cartographic-zoom`.

catcorner22 Bundle

Audit a multi-vendor agent-skill library for routing pathology — the defect classes that appear only when many packs from different vendors share one router: vendor steering, SKILL.md/overlay.yaml drift, silently disabled rules, unreachable guards, greedy descriptions, cross-pack territory claims, shadowed and cyclic routes, dangling references. Use when adding a skill pack, merging two libraries, or when skills fire on the wrong thing. Ships a runnable analyzer at scripts/audit_skill_library.py. Scope boundary: this audits a SKILL LIBRARY's routing metadata, not application code — for authoring a single new skill use skill-creator, and for reviewing ordinary source changes use code-review.

catcorner22 Bundle

Track and visualize ML training experiments with Trackio. Use when logging metrics during training (Python API), firing alerts for training diagnostics, or retrieving/analyzing logged metrics (CLI). Supports real-time dashboard visualization, alerts with webhooks, HF Space syncing, and JSON output for automation.

catcorner22 Bundle

AI demos and GPU compute with Gradio Spaces and Hugging Face Spaces ZeroGPU. Use when writing or reviewing code that uses `@spaces.GPU`, configuring `python_version` or `requirements.txt` for a ZeroGPU Space, or handling ZeroGPU-specific code constraints — pickle-based process isolation, `gr.State` semantics across the worker boundary, no `torch.compile` (use AoTI instead), CUDA wheel-only **builds** (no `nvcc` at build; runtime has `nvcc` for AoTI), large vs xlarge sizing, and dynamic duration callables. Make sure to use this skill whenever the user mentions ZeroGPU, `@spaces.GPU`, or the `spaces` Python package, or hits ZeroGPU-specific code errors like `PicklingError` across the worker boundary, `illegal duration`, or `flash-attn` wheel-build failures — even when the user does not explicitly ask for ZeroGPU coding guidance. Trigger on `import spaces` or `@spaces.GPU` in code.

catcorner22 Skill

Router for adaptive AI-transfer techniques: compute triage, AAR debriefs, mid-flight feedback, wayfinding telemetry, annealing phases. Use when tuning dynamic agent behavior. Scope boundary: full catalog → `ai-transfer-ecosystem-primer`.

catcorner22 Skill

Require three independent retrieval paths per factual claim; score agreement 3/3, 2/3, or contested. Use when single-source RAG is insufficient. Scope boundary: double-entry gate → `double-entry-claims`.

catcorner22 Skill

Run prompt on models with different strengths; fusion pass extracts best elements selectively — not averaging. Use when models complement (reasoning + phrasing + facts). Scope boundary: single-model routing → `emergency-triage-compute`.

catcorner22 Skill

Create LangChain agents with `create_agent` / `createAgent`: model and tool wiring, the agent loop, and where middleware attaches. Use when the code imports `langchain`, `langchain.agents`, or `@langchain/core`, or the user names LangChain agents or `create_agent`. Scope boundary: writing the middleware itself and structured output belong to `langchain-middleware`; hand-written graphs to `langgraph-fundamentals`; an agent request that has not chosen a framework to `build-agents`/`eve`.

catcorner22 Skill

Mid-generation monitoring: user activity, corrections typing, deadlines, confidence — adjust or abort mid-stream. Use for long agent runs and streaming workflows. Scope boundary: post-hoc AAR → `after-action-review`.

catcorner22 Skill

Declare reasoning DAG before prose: premises, claims, dependencies. Flag downstream if upstream fails. Use for multi-step arguments and agent plans. Scope boundary: dual-axis → `score-study-dual-axis`.

catcorner22 Bundle

Creates, optimizes, and iteratively refines agent prompts, system prompts, developer prompts, and reusable prompt templates. Use when asked to improve a prompt, optimize a system prompt, rewrite an agent prompt, tune prompt wording, make a prompt more reliable, port prompts between OpenAI, Claude, or Gemini, or build prompt evals. Scope boundary — this skill owns agent PROMPT authoring and optimization: the instruction text itself, its layering, its examples, and the evals that score it. Agent ARCHITECTURE and scaffolding belong elsewhere: generic "build/create/design/scaffold an agent" requests go to `build-agents`, and eve project structure, runtime, channels, and tooling go to `eve`. Do not claim a request just because it mentions an agent; claim it when the artifact under edit is a prompt.

catcorner22 Skill

Router for college and academic work in this library: writing, citations, study planning, source attribution, lecture notes, and presentations. Use when the user is a student, mentions college/school/coursework, or asks how to approach an assignment, paper, exam prep, or lab. Scope boundary: CS/software implementation → `coding-ecosystem-primer`; AI/ML engineering → huggingface/langchain packs; GitHub PR workflow for group code projects → `cursor-team-kit`; general execution posture → `proactive-agency`.

catcorner22 Skill

Router for refinement AI-transfer techniques: three-axis grading, debate voting, model blending, faceting passes, localization QA. Use for final polish and multi-model fusion. Scope boundary: full catalog → `ai-transfer-ecosystem-primer`.

catcorner22 Skill

Graft a specialist scion onto a safe rootstock output: keep the trusted base, attach expert additions at a marked join. Use when mixing a general answer with a domain specialist pass. Scope boundary — multi-model fusion → `wine-blending-fusion`; foreign genre skeleton → `cross-pollination-structure`.

catcorner22 Skill

Instrument consumption behavior (scroll-back, re-prompt, abandon) to restructure future outputs. Use for productized AI interfaces with telemetry. Scope boundary: static layout → `stage-blocking-layout`.

catcorner22 Skill

Subagent delegation, task planning, and approval gates inside the `deepagents` harness: SubAgentMiddleware and the `task` tool, TodoListMiddleware and `write_todos`, and HITL interrupts on subagent tool calls. Use when a Deep Agent must delegate to a named subagent, plan multi-step work, or gate a tool call on human approval. Scope boundary: this covers orchestration inside the `deepagents` harness only — harness setup and middleware selection belong to `deep-agents-core`, approval gates on a plain LangChain agent to `langchain-middleware`, `interrupt()` in a hand-written graph to `langgraph-human-in-the-loop`, and a multi-agent request that has not chosen a framework to `build-agents`/`eve`.

catcorner22 Skill

Source-first generation: retrieve evidence per claim before prose, inline attribution, strip unattributable claims. Use for factual writing and RAG outputs. Scope boundary: bibliography formatting → `citation-literacy`; triple-path check → `survey-triangulation`.