191 prompts
Multi-step prompts for planning, tool choice, state tracking, and action boundaries.
As a Verifier Engineering Strategist, you design, audit, and reject verifier systems that convert model outputs (final answers, intermediate steps, tool calls, agent trajectories) into trustworthy signals for downstream systems like RL trainers or evaluators. Treat verifiers as first-class engineering artifacts with failure modes, calibration curves, and adversarial surfaces.
This prompt defines an agentic CAD and hardware design engineer capable of translating natural language requirements into validated CAD artifacts (parts, assemblies, enclosures, fixtures, robot models), treating Python/build123d source as the single source of truth and STEP/STP as primary outputs. Suitable for parametric solid modeling, robotics description formats (URDF/SDF/SRDF), and source-controlled mechanical design automation.
Design AI agent systems by explicitly separating cognitive functions across model weights, context window, and externalized artifacts (memory, skills, protocols, harness) to enhance testability, auditability, and maintainability.
This prompt provides a battle-tested architecture for building cross-platform desktop apps that feel truly native in appearance, interaction, and performance. Based on Raycast's 2.0 rewrite, it balances code sharing with platform fidelity, ideal for utility apps requiring high UI consistency and OS integration.
A long-horizon agent that treats the filesystem as durable working memory and the context window as volatile cache, using three core files (task_plan.md, findings.md, progress.md) to enable recoverable multi-step execution and error tracking.
Optimize existing AI coding-agent harnesses (e.g., Claude Code, Codex CLI, Cursor) to achieve consistent, measurable, production-grade outcomes through cross-harness parity, memory persistence, security, and continuous learning.
This prompt guides the transformation of open-source GUI applications into stateful, machine-readable CLI tools operable by AI agents without a display, using real backend software for rendering and export.
Design production-grade cybersecurity skills following the agentskills.io standard to transform generic AI agents into capable security analysts, with cross-mapping to five industry frameworks and executable workflows.
Designs a multi-role tooling system for AI coding agents that operates like a software factory, emphasizing role specialization, mutual review, and explicit invocation to improve code quality and delivery reliability.
Design a coordination system where humans provide direction via lightweight chat messages and autonomous coding agents ("claws") self-coordinate to plan, build, test, recover, and push code without micromanagement, emphasizing externalization of meta-tasks and context-window purity.
Design a unified virtual filesystem layer enabling AI agents to interact with heterogeneous backends (S3, Google Drive, GitHub, etc.) using standard Unix-like tools, abstracting away multiple APIs into a single familiar filesystem interface.
Design a fully on-device voice input/output architecture supporting multiple TTS engines, zero-shot voice cloning, global dictation, agent voice output, and post-processing, ensuring voice data never leaves the device unless explicitly authorized by the user.
Designs generator/evaluator harness patterns enabling parallel LLM sub-agents to collaboratively build large, coherent software artifacts (e.g., compilers, interpreters, runtimes) with deterministic quality gates, bounded coordination cost, and failure isolation.
Design a zero-human multi-agent company operating system with org structure, task allocation, budget control, governance, and audit trails for autonomous, goal-driven execution under financial constraints.
A fully autonomous machine learning experimentation agent that runs closed-loop experiments on a fixed codebase without human intervention, iteratively modifying training code, running short-budget trials, and optimizing a single ground-truth metric.
Determine optimal timing for requesting user clarification in long-horizon AI agents based on information type and execution progress to maximize value and avoid harm.
An interactive, stateful research partner for mathematicians working on open-ended problems, supporting the entire lifecycle of mathematical discovery: ideation, literature search, computational exploration, conjecture formation, theorem proving, and theory building. This prompt emphasizes exploratory, iterative collaboration over simple problem-solving.
Design cost-efficient, distribution-shift-robust routing policies to dynamically assign queries between reasoning and non-reasoning LLM judges under a fixed compute budget, optimizing accuracy-cost trade-offs.
A local-first, agent-agnostic design production system that generates complete visual artifacts without cloud lock-in, emphasizing structured pipelines, brand consistency, multimodal output, and five-dimensional self-critique.
Design a two-layer permission classifier for agents to operate efficiently on low-risk actions while preserving human approval for high-risk operations, eliminating confirmation fatigue without compromising safety.
This prompt diagnoses whether a reasoning model's chain-of-thought (CoT) is substantive (genuinely changes the final answer) or theatrical (decorative output around a pre-decided answer), and designs routing policies to allocate CoT budget only where needed.
A strategic job-search system that treats career moves as capital allocation decisions, helping users identify high-value opportunities, evaluate them rigorously, and execute with precision.
Design and deploy scalable parallel prompt-learning systems that achieve significant speedup over serial Automatic Prompt Optimization (APO) methods without sacrificing quality.
Design a self-wiring, entity-centric knowledge brain for a personal AI agent that ingests multi-source content, enriches entities, performs hybrid retrieval, and maintains itself autonomously during idle cycles.