Skill UI

This skill provides advanced secondary workflows for Ideogram, enabling comprehensive image manipulation beyond simple text-to-image generation. Key functions include Magic Fill (inpainting/editing), Remix (style transfer), Upscaling, describing image content (image-to-text), and re-framing. Ideal for professional image refinement, style transfer, and content enhancement.

This skill automates the research refinement process by entering an autonomous, multi-round review loop. It leverages any OpenAI-compatible LLM to score, critique, and guide necessary fixes until the work meets a high, expert-defined standard. It manages state persistence across rounds, simulating the rigorous process of professional peer review.

This skill implements an autonomous, multi-round research review loop, simulating the rigorous peer-review process of top-tier conferences. It iteratively reviews, identifies weaknesses, suggests fixes, and re-reviews a research document using the MiniMax API. The system manages state persistence across multiple rounds, ensuring continuous improvement until a 'ready' assessment is achieved or the maximum round limit is reached. Ideal for academic writing, thesis refinement, or comprehensive project documentation.

Orchestrate a complete idea discovery workflow that goes from a broad research direction to validated, pilot-tested ideas. Chains literature survey, idea generation, novelty checking, critical review, and research refinement into a single automated pipeline, producing a ranked idea report, a final proposal, and an experiment plan.

Turn vague research directions into concise, problem-anchored method plans. Uses iterative AI review to focus contributions, ensure frontier leverage, and produce implementation-ready proposals suitable for top venues. Ideal when users need help refining research ideas, decomposing problems, or polishing approaches.

Provides patterns and techniques for building robust evaluation loops for AI agents. This skill guides the implementation of self-critique, evaluator-optimizer pipelines, and test-driven refinement workflows. Use it to significantly enhance the accuracy and quality of complex outputs like code, reports, and analyses by implementing structured, iterative improvement cycles (Generate → Evaluate → Critique → Refine).

This skill provides a comprehensive framework for optimizing, improving, and debugging Large Language Model (LLM) prompts. It leverages production trace data, system evaluations, and human annotations to extract prompts from spans. Users can run a data-driven optimization loop, ensuring the AI responses meet desired quality standards. Ideal for prompt tuning, system prompt refinement, and improving overall AI output quality.

Interactive workflow that interrogates scope, deliverables, constraints, and success criteria while guiding users to refine task prompts; copies the final markdown with Joyride clipboard commands and never writes code.

This skill autonomously optimizes artifacts like code or models using Hypothesis Tree Refinement. It runs iterative experiments with dev/test evaluators to prevent overfitting, coordinating subagents in isolated worktrees. Ideal for improving training recipes, agents, or pipelines without manual supervision.

A comprehensive guide for positioning AI/ML manuscripts for top-tier general AI conferences like AAAI. It helps authors refine their scope, frame their contribution, provide necessary evidence (e.g., ablations, baselines), and strategically compare their work against related research and similar venues (ICML, IJCAI). Ideal for pre-submission manuscript refinement.

This skill processes verbose, stream-of-consciousness voice dictations (from tools like macOS Dictation or Wispr Flow) and transforms them into highly structured, token-efficient prompts. It removes filler words, repetitions, and tangents, organizing the core intent into standard sections (Context, Objective, Constraints) suitable for advanced LLMs like Claude Code.

This workflow implements a bounded, iterative refinement loop for AI agents. It utilizes a required markdown rubric to grade the agent’s output in an isolated context. Based on the grader's verdict, it recommends the next move (sharpen, re-run, or promote). Once the agent passes the rubric, held-back evaluation cases are run in parallel to ensure generalization before final deployment.