Prompts

Capture the real shape of an autonomous trading desk — named roles, shared files, one brief, human approval — without cloning a seven-bot factory.

No-hype, step-by-step process for actually shipping a working agent.

Clean modular repository structure optimized for Claude Code with skills, hooks, and docs.

Decide the shape of a workflow before you add more agents. Start as a chain. Reshape only where the wait is fake.

Conceptual framework for building durable, teachable agent systems that survive across sessions.

The one piece left in the antpalkin recap: a persistent hypothesis graph and negative-results memory so the next strategy does not forget what already failed.

Turn a codebase into a dark-theme HTML diagram with clickable flows, plus a JSON graph agents can read.

Framework for building recursive agentic systems that generate and verify their own work (originally applied to hedge-fund style loops).

Ready-made specialized agent personas organized by function for multi-agent teams.

Explicit rules for how an agent should plan, use subagents, self-improve, verify, and manage tasks.

One-page checklist that prevents the most common Claude misuse and gets you to high-leverage usage fast.

Practical guide to using the Cowork surface instead of plain Chat, with concrete use cases.

Clear decision matrix for which Claude surface to use so you stop over-prompting or under-structuring.

10-step practical path to go from basic chat user to high-leverage Claude operator in seven days.

Curated, high-signal skills that people actually run, grouped by function.

Official Anthropic plugin that analyzes your project and recommends the right automations (hooks, skills, MCP servers, subagents).

A Claude skill that writes one paste-ready prompt for a named tool. Use it when you are writing or fixing a prompt, not as a course.

Forces Claude Code to give action-first, numbered, zero-fluff answers.

Forces investigation + clear plan before any code is written. Treats the agent like a contractor who bills for rework.

A short list of hard rules that prevent the most common, predictable mistakes language models make when writing code.

Eliminate half-finished work, unnecessary questions, and slow sequential execution in coding agents.

Stop an agent from guessing, overbuilding, touching extra code, or agreeing with a bad idea.

Combined high-standard operating manual from Boris Cherny and Andrej Karpathy for coding agents.

Structured template that treats context as a first-class engineering artifact instead of an afterthought.

The useful core of Prompt Master: extract intent, then write a short prompt. Every word has to earn its place.

Set of ready-to-use role-based system prompts originally designed for Perplexity Computer / agentic use.

Precise, labeled template for high-performance Claude prompts that forces clarity and structure.

The walkable method from David Roberts, not the Calico course. Product URL in, 30-second talking-head ad out.

Official v2 run files. Load the skill once, then one of these two prompts. Not a second Taste item.

Turn Claude into a pixel-perfect design replication engine that can recreate any UI exactly.

Device-to-device file sync with no cloud account. A lookup when Dropbox / iCloud / OneDrive is the wrong shape.

Classic example of breaking a complex system down to raw material cost vs finished cost (Musk / SpaceX).

Comprehensive system for Obsidian + Claude as a personal knowledge OS using PARA, Zettelkasten, and AI roles.

Curated starting points for free or low-cost AI learning paths, with emphasis on Claude-related and practical courses.

Structured monthly roadmap from coding foundations to specialized AI engineering roles.

High-level behavioral skills that compound over time instead of one-off prompts.

Two-stage workflow that protects expensive tokens by using cheap models for exploration and strong models only for precise execution.

The token rules sprint #10 did not have. Most spend is the agent rereading history, not what you typed.

End-to-end example of building a full life OS (tasks, finance, habits, nutrition, calendar, journal, second brain) with voice control.

Semi-roadmap for client implementations. Does not change what good looks like. Not a 21st sprint item.

Practical 7-step modern workflow for shipping a real website quickly with Claude Code.

Full system prompt that scaffolds a production-ready iOS app infrastructure with modern patterns.

High-leverage prompt that forces planning + production-quality output in one shot for apps, dashboards, and tools.

Drop-in prompt that lets agents run for hours with almost zero hand-holding. Set the goal and walk away.

Comprehensive set of 90 ready-to-use commands organized by intent for speed and consistency.

Visual mapping of classic tools → 2026 AI-native replacements.

Reference for a free CLI that gives an agent eyes on X, YouTube, GitHub, Reddit, LinkedIn, and other sites without paying platform API fees.

Optional agent-owned inbox when a practice address cannot live on Colton’s Gmail. Not a replacement for Sieve.

The working catalog of open-source agent apps and skills. Browse here before cloning a one-off repo or inventing a ninth pattern.

A four-agent stack that picks a direction, checks it, sizes risk, and sends an order. Wants a wallet key.

Official x.ai index for how people actually run Grok Bot. Beats the X tip/course pile. Not a Week 1 swap.

Turn agent-written HTML into an MP4. Same HTML should render the same video.

A searchable index of open models and artifacts, sold as torrents for weights. Use as a map. Do not treat it as the source of truth.

An offline talking avatar with voice interrupt and optional screen watch.

A finance workspace where specialist agents argue a strategy out loud. Not a live book.

One prompt bar that routes to many image and video models. Community wrapper. Keys still billed.

A desktop terminal for filings, charts, and public market data. Open edition is AGPL. Enterprise is paid and can route live.

Quick mapping of paid tools to free (or cheaper) alternatives.

Read-only-default audit of paid ad accounts across the big platforms.

One self-hosted chat window over Claude, Gemini, DeepSeek, and other keys. Data stays on your box.

An inbox that can read mail and draft replies while the messages stay on your routing.

A headless browser built so an agent is harder to fingerprint. Returns an accessibility tree instead of a huge HTML dump.

Complete layered breakdown of a production prediction-market bot that scans 50+ markets per cycle.

Safe, structured multi-step process for creating custom Pine Script indicators via TradingView MCP.

Exact step-by-step install and configuration for live TradingView data into Claude via MCP.

Extreme custom instructions that force pure tool behavior — zero opinion, zero preamble, pure execution.

Make AI output sound exactly like you by encoding writing rules, banned phrases, and real samples.