Browse agent skills

Cursor rules for Grok 4.7: grounded coding plus design directions, screenshot review, own ideas, and direction-only questions.

Rampart Edge — client-side PII gate for AI assistants: deterministic validators + one 1.45 MB ONNX model (onnxruntime-web) that decides BLOCK or ALLOW before a message leaves the browser and marks the PII spans to redact.

A dual-face Cordis plugin for DeepSeek Harness: the host half owns a coordination SQLite database, seven model-facing tools and an HTTP surface; the web half mounts five composer dropdowns, a settings catalog and the service-level takeover dialog.

Holocron plugin for Checkly. Implements the monitoring capability against Checkly's REST API, plus exports verifyToken + AUTH_HINT for `holocron auth`.

Knowledge (RAG) for Tangen CMS — embeddings, semantic search, ask-with-citations. Optional: install only when you enable knowledge.

Database client and system schema (Drizzle + postgres.js) for the Tangen headless CMS

Type definitions and config helpers for the Tangen headless CMS

Forms for a Tangen site: a builder, public embeds, headless schema and submit routes, submissions, notifications, webhooks, spam protection and localization

Asaas (API v3) nodes for n8n: customers, payments (Pix, boleto, card), card tokenization, subscriptions (recurring billing), webhooks and a webhook trigger

Ncell account SDK and command-line interface

Describe any machine in one line. Claude builds a photoreal, physics based, interactive simulation of it in one HTML file, plus the film to post it. An open source Claude Code skill.

Describe a factory in one line. Claude builds it as a live 3D simulation you can run. Open source agent skill: real discrete event simulation, tilt-shift miniature.

Multi-LLM adversarial review council - an AXI

Luma: Luigi's Box frontend integrations (autocomplete, search, category listings, recommenders) on the Luigi's Box AI API. Pre-release.

Vercel AI SDK provider for UplinkSwitch: one key and one base URL for Claude, GPT, Gemini, Grok, DeepSeek and the open-weight models.

Command line for agent-context-pruner: evaluate decision models against recorded agent sessions.

Evaluation harness for agent-context-pruner: replays recorded sessions, labels which tool results were needed later, and measures each decision model's keep threshold.

A floating "Ask AI" button that hands the visitor's question and your site URL to ChatGPT, Claude or Google AI Mode. No API, no server, no tracking.

A claude code mod for humna-AI augmented workflow -- like tony stark

n8n node for Factline: European company checks, supplier invoice checks, sanctions and PEP screening, EU VAT and Peppol lookups from official records.

ants: Autonomous Navigation of Tiny-model Space. A small autonomous research loop: it starts on one cheap machine and scales with the compute you give it. (npm launcher: it builds a private Python environment on first run)

WeChat Official Account (微信公众号) article workflow skill: topic radar → writing → cover → themed layout → drafts box

Find bug fixes that reached some sibling implementations but not the others. A git-history scanner plus Agent Skills for Claude Code, Codex, and any SKILL.md agent.

A TypeScript client for the Checkly monitoring API

A colleague assistant on Effect: an append-only event log per chat, an interruptible agent loop, specialists, skills, and UI blocks

defineEvent() and Zod to JSON Schema extraction, so a Zod schema is the catalog entry.

Runtime-neutral HTTP client primitives for Polychat

Governed model registry mechanisms for Polychat: the pure policy evaluator, licence normalisation, static weight inspection that never executes a file, evaluation statistics, lineage and CycloneDX ML-BOM export

Feature flag mechanisms for Polychat hosts: an OpenFeature-shaped evaluation contract, deterministic bucketing from a targeting key, code-defined rules with weighted splits and targeting, a Cloudflare Flagship adapter and a layered provider that lets a da

n8n node for Amazon Bedrock knowledge bases and guardrails: retrieve, retrieve-and-generate, and apply guardrail.