AI & ML

wundercorp Bundle

Vector search engine for production RAG systems.

runcomfy-com Skill

Prepare datasets and manage AI Toolkit LoRA training on RunComfy GPUs. Use for reviewing training YAML, captions, dataset readiness, starting an approved job, monitoring progress and retrieving checkpoints or samples. Verify base-model compatibility and costs before training.

runcomfy-com Skill

Create videos with RunComfy AI video models including Seedance, Wan, LTX and Kling. Use for text-to-video, image-to-video, reference-to-video and video edits. Check model-specific duration, aspect ratio, audio, inputs and price before submitting one job.

utopia-v Bundle

设计或审阅 Broker 等由 Model、Harness 与 Environment 共同产生行为的 Agent 系统。用于 context、长期状态、工具、MCP、权限、数据流与恢复语义。

utopia-v Bundle

Delegate bounded work from a Codex controller to configured workers on other model providers without changing the controller's provider. Use when cross-provider delegation, a named Broker route, or a lower-cost/specialized external worker is requested. Do not use for ordinary same-agent work.

koffih Bundle

Extended autonomous work mode. The agent inspects the whole project, finds unfinished work, bugs and next steps, then implements, tests and commits for hours without waiting for approval on ordinary decisions, stopping only for destructive or irreversible ones. Use when the user will be away and wants the project to keep moving on its own.

mizoreww Skill

Inspect, add, change, remove, repair, or update Claude and Codex configuration from Mizoreww/awesome-agent-config on main. Use for configuration queries and edits, installed skills/plugins/rules/MCP, and changes to this repository's configuration templates or catalogue.

mizoreww Bundle

Adversarial code review using the opposite model's CLI. Spawns 1–3 reviewers on the opposing model (Codex sessions typically spawn Claude; Claude sessions spawn Codex) to challenge work from distinct critical lenses. Triggers: "adversarial review".

wundercorp Bundle

Train sparse autoencoders to interpret model features.

wundercorp Bundle

Constrain LLM output with grammars; guarantee valid JSON.

wundercorp Bundle

Managed vector DB for production RAG and search.

mizoreww Bundle

Adversarial code review using the opposite model. Spawns 1–3 reviewers on the opposing model (Claude spawns Codex, Codex spawns Claude) to challenge work from distinct critical lenses. Triggers: "adversarial review".

wundercorp Bundle

Use when an agent needs AgentMail CLI email inboxes.

wundercorp Bundle

Generate images, video, and audio via diffusion workflows.

wundercorp Bundle

Structured LLM outputs validated with Pydantic.

wundercorp Bundle

OBLITERATUS: abliterate LLM refusals (diff-in-means).

wundercorp Bundle

Curate LLM training data: dedupe, filter, PII redaction.

wundercorp Bundle

High-throughput LLM inference on NVIDIA GPUs.

wundercorp Skill

Pay HTTP 402 APIs via Machine Payments Protocol (MPP).

wundercorp Bundle

Outlines: structured JSON/regex/Pydantic LLM generation.

wundercorp Skill

Agent payments via Stripe Link — cards, SPT, approvals.

wundercorp Bundle

Automate Unreal Engine editor scenes, actors, and renders.

wundercorp Bundle

Set up Actual Computer (actual.inc) inference in Loki.

wundercorp Bundle

DSPy: declarative LM programs, auto-optimize prompts, RAG.

wundercorp Skill

Parallel 4-agent cleanup of recent code changes.

wundercorp Bundle

Control TouchDesigner via twozero MCP.

wundercorp Bundle

Agent RAG and long-term memory with Pinecone.

microxro Bundle

Decompose a large or multi-part build into a verified task tree, splitting it the optimal amount — wide enough to parallelize, never wider than the work actually supports — so it gets done at the best effort in the least wall-clock time, dispatch the independent pieces to subagents running on your local model, and refuse to declare the task done until every required piece has been implemented, self-tested by its own worker, re-verified by you against the merged code, and integration-checked together. Use this whenever a request involves multiple features, multiple components/files, or phrases like "build the whole thing", "make sure everything works", "don't stop until it's done", "run these in parallel" or "use multiple agents" — even if the user never says "skill" or "task tree". Also use whenever the user explicitly invokes /tree. Do NOT use it for a single small fix or one-file change — the decomposition and gate overhead only pays off on real multi-part work.

microxro Bundle

Apply practitioner judgment instead of generic-model defaults when creating or critiquing a real deliverable — code, UI, documents, plans, messages, system designs, charts. Ground in the actual job, audience, and a real inspected exemplar; pick the right shape instead of a default template; rank by what matters; use only source-backed specifics instead of invented detail; commit to one recommendation instead of a menu; then subtract and quality-gate before calling it done. When reviewing existing work, give a short verdict plus the few things that matter, not an exhaustive rubric. Use whenever the user says "use taste" / "apply taste" / "make this good" / "tighten this" / "polish this", for any deliverable someone else will see or with real stakes, when a brief is thin and there is pressure to invent specifics, or for any review/critique/comparison — even without saying "taste". Do NOT use for verbatim transformation or exhaustive-coverage tasks — taste is not permission to editorialize a coverage task.

wundercorp Skill

Embed an in-page natural-language GUI copilot in web apps.

wundercorp Bundle

Manual OAuth for remote MCP servers on headless gateways.

wundercorp Bundle

Fast BPE/WordPiece tokenization and custom vocab training.

wundercorp Bundle

Speed up long-sequence transformer training and inference.

wundercorp Skill

Agent-native web search, deep research, and enrichment.

wundercorp Skill

Delegate coding tasks to the Blackbox AI multi-model CLI.

wundercorp Skill

Delegate coding to OpenHands CLI (model-agnostic, LiteLLM).