deep▌

Exhaustive research with configurable depth, latency, and cost trade-offs for complex topics. \n \n Three processor tiers (pro-fast, ultra-fast, ultra) ranging from 30 seconds to 25 minutes, with cost scaling from 1x to 3x baseline \n Asynchronous execution with polling: kick off research instantly, monitor progress via URL, retrieve results when ready without blocking \n Outputs formatted markdown report and JSON metadata; executive summary printed to stdout for quick overview \n Designed for e

Answers: "What's happening in this specific market? Who holds it, who's trading it?"

Orchestrate subagents, plan multi-step tasks, and require human approval for sensitive operations. \n \n Delegate work to specialized subagents via the task tool; custom subagents support isolated tool sets and system prompts, while the default \"general-purpose\" subagent inherits main agent configuration \n Plan and track complex workflows with write_todos , organizing tasks across pending, in-progress, and completed states; requires a thread_id for persistence across invocations \n Implement

Multi-source research synthesis with citation tracking, source verification, and structured reporting across 8-phase methodology. \n \n Executes parallel searches and spawns concurrent agents to gather 10+ sources quickly, with credibility scoring and triangulation across sources \n Generates comprehensive markdown reports with full bibliographies, executive summaries, and detailed findings—each claim immediately cited [N] \n Produces three output formats automatically: markdown (source), McKins

Comprehensive research assistant that synthesizes information from multiple sources with citations. \n \n Follows a systematic five-step research process: clarifying the question, identifying key aspects, gathering information, synthesizing findings, and documenting sources \n Structures output with executive summary, key findings, detailed analysis by subtopic, areas of consensus and debate, and source evaluation \n Evaluates source credibility across peer-reviewed journals, official reports, r

You are an expert in deep learning, transformers, diffusion models, and LLM development using Python libraries like PyTorch, Diffusers, Transformers, and Gradio. Follow these guidelines when writing deep learning code.

Use when:

Pluggable memory and file backends for Deep Agents with ephemeral, persistent, and hybrid routing options. \n \n Four backend types: StateBackend (thread-scoped, ephemeral), StoreBackend (cross-session persistent), FilesystemBackend (real disk access for local dev), and CompositeBackend (route different paths to different backends) \n FilesystemMiddleware provides six file operation tools: ls , read_file , write_file , edit_file , glob , grep \n CompositeBackend uses longest-prefix matching to r

Comprehensive research and discovery before building something new. Instead of jumping straight into code from training data, this skill goes wide and deep — local exploration, web research, competitor analysis, ecosystem signals, future-casting — and produces a research brief that makes the actual build 10x more productive.

Transforms surface-level reading into deep learning through systematic analysis using 10+ proven thinking frameworks. Guides users from understanding to application through structured workflows.