Skill UI

Provides six structured AI engineering workflows—from prompt evaluation and context budgeting to RAG architecture, security audits, evaluation harnesses, and product sense coaching—ensuring repeatable, CI/CD-friendly decisions before deployment.

This skill executes an autonomous, iterative research review loop, simulating a rigorous peer-review process. It repeatedly uses an external OpenAI-compatible LLM via API calls to score, critique, and identify weaknesses in a given research topic or document. The process continues to refine the content by implementing suggested fixes until the LLM provides a positive assessment or the maximum round limit is reached. Ideal for academic writing, technical documentation, and refining complex proposals.

Autonomous loop that submits research context to MiniMax-M2.7, parses reviewer feedback, and iterates through fixes until a positive assessment or max rounds is reached, using compact state persistence for long runs.

Converts a fuzzy research direction into a concrete, implementation-oriented method plan with iterative GPT‑5.4 review rounds, ensuring the original problem stays central while the approach remains elegant, focused, and aligned with frontier capabilities.

Behavioral X-Ray probes models through refusal, formatting, reasoning, persona, tool use, and grounding dimensions, auto-tagging each response and building a styled HTML report for red-team, compliance, and selection workflows.

Combines regex signatures, heuristic scoring, and a DeBERTa classifier to detect prompt injection attempts against LLM-based applications, flagging or blocking malicious inputs before they reach the model and fitting into validation layers, audits, or incident response workflows.

Crafts complete OpenClaw lobster personas from scratch—guiding identity tension, boundary rules, names, and avatar prompts (with optional automated portraits) via guided choices or gacha randomness for fully voiced agents.

Correlates OSINT findings across usernames, emails, social media, domains, breach records, and dark-web mentions using AI/LLM reasoning to build unified target profiles with confidence scoring and link analysis.

LLM Cost Optimizer audits and redesigns LLM pipelines to cut API spend 40-80% without quality loss by routing workloads, caching prompts, controlling lengths, and monitoring token costs.

Generates professional PowerPoint decks by planning slide structure, crafting AI images for each slide, and composing them into a consistent, visually rich presentation file.

A cultural pressure‑style skill that forces the assistant to repeatedly push past failures, think systemically, validate every change, and proactively hunt down related issues until end‑to‑end ownership is demonstrated.

This guide outlines the structure and content guidelines for an AGENTS.md file. This file serves as a comprehensive 'README for agents,' providing detailed, technical context and operational instructions (such as setup, development workflow, testing procedures, and code style) that enable advanced AI coding agents to interact with and understand a codebase effectively. It greatly complements traditional READMEs by focusing purely on machine-readable technical requirements.