Skip to content

further-reading.md

Latest commit

 

History

History
136 lines (101 loc) · 8.04 KB

further-reading.md

File metadata and controls

136 lines (101 loc) · 8.04 KB

Further Reading

Articles

Your codebase is NOT ready for AI (here's how to fix it)

  • Type: Article
  • Author: Matt Pocock
  • Summary: Discusses deep modules, progressive disclosure, graybox modules, and the AI-as-new-starter metaphor to prepare codebases for AI assistance.
  • Related to: L0
  • URL: https://medium.com/@mattpocockuk

Beyond Copilots: Building a Closed-Loop Agentic Testing Ecosystem with MCP and Cursor

What is Agentic Testing?

The Creator of Clawd — "I Ship Code"

  • Type: Podcast / Article
  • Creator: Gergely Orosz (The Pragmatic Engineer) with Peter Steinberger
  • Summary: Peter Steinberger (creator of OpenClaw) describes building entirely with AI agents: closing the feedback loop so agents verify their own work, investing heavily in planning before implementation, managing 5-10 parallel agents, and treating code reviews as architecture discussions. Demonstrates that agentic development requires more engineering discipline, not less.
  • Related to: L0, L1, L2, L3, L4
  • URL: https://newsletter.pragmaticengineer.com/p/the-creator-of-clawd-i-ship-code

Andrej Karpathy's LLM Coding Principles

  • Type: Behavioral specification
  • Author: Forrest Chang (forrestchang)
  • Summary: Codifies Andrej Karpathy's observations about LLM coding pitfalls into four behavioral principles: Think Before Coding, Simplicity First, Surgical Changes, and Goal-Driven Execution. Installable as a Claude Code plugin or per-project CLAUDE.md overlay. The principles reduce the most common costly mistakes LLMs make when writing code — wrong assumptions, overcomplicated abstractions, and unnecessary refactoring.
  • Related to: L2
  • URL: https://github.com/forrestchang/andrej-karpathy-skills

Claw Code — Autonomous Agent Coordination

  • Type: Tool / Framework
  • Creator: UltraWorkers
  • Summary: Rust reimplementation of a multi-agent coding coordination system. Demonstrates three-part architecture: OmX (workflow layer converting human directives into structured execution), clawhip (event and notification router keeping monitoring outside the agent's context window), and OmO (multi-agent coordination with planning, handoffs, and verification loops). Features token-based routing, a 9-lane parallel development model, and a mock parity harness for behavioral validation against a deterministic Anthropic-compatible service.
  • Related to: L1, L2, L3
  • URL: https://github.com/ultraworkers/claw-code

Compound Engineering Plugin

  • Type: Tool / Framework
  • Creator: Every Inc.
  • Summary: AI-agent plugin providing a structured development workflow built on the principle that each unit of engineering work should make subsequent units easier, not harder. The core cycle is ideate -> brainstorm -> plan -> work -> review -> compound -> repeat, with 80% planning and review, 20% execution. Knowledge is codified as "learnings" via /ce:compound so it compounds across the team over time. Available for Claude Code and convertible to 10+ other agent platforms.
  • Related to: L2, L3
  • URL: https://github.com/EveryInc/compound-engineering-plugin

REPL is dead, long live REPL

  • Type: Article
  • Author: Brandon Waselnuk (Unblocked)
  • Summary: Argues that the Read-Eval-Print Loop hasn't died with agentic coding — it's been promoted. Engineers now operate the REPL at the intent layer (read agent output, evaluate against intent, print next instruction, loop) while agents handle the code-level REPL. The article demonstrates that context curation and intent-level evaluation produce drastically better outcomes than naive code-level REPL, and that the skill shift is learnable but real: from writing code to writing instructions and reviewing diffs.
  • Related to: L1
  • URL: https://getunblocked.com/blog/repl-is-dead-long-live-repl/

Videos

Your codebase is NOT ready for AI

  • Type: Video
  • Creator: Matt Pocock
  • Summary: Covers deep modules, graybox modules, and progressive disclosure of complexity to make codebases AI-friendly.
  • Related to: L0
  • URL: https://youtu.be/uC44zFz7JSM

The WISC Framework: 90.2% Better AI Coding Results

  • Type: Video / Framework
  • Creator: Cole Medin (original), DIY Smart Code (adaptation)
  • Summary: Context engineering framework — Write, Isolate, Select, Compress. Core thesis: ~80% of agent failures come from context management, not the model. Write externalizes memory (structured commits, progress files, separate planning/implementation sessions). Isolate splits work across sub-agents (scout pattern, research delegation). Select loads only what's needed via a 3-tier context system (global rules, on-demand rules, reference docs). Compress provides focused compaction and session handoffs. The Scout Pattern used in Pattern 3.4 maps to WISC's Isolate strategy.
  • Related to: L3
  • URL: https://youtu.be/gyo0eRgsUWk
  • GitHub: https://github.com/coleam00/context-engineering-intro/tree/main/use-cases/ai-coding-wisc-framework

Tools

Superpowers (obra)

  • Type: Tool / Framework
  • Creator: Jesse Vincent (obra)
  • Summary: Skills framework for Claude Code providing automated worktree management, brainstorming, planning, TDD, code review, and verification workflows. The worktree, planning, and execution skills automate the full development lifecycle within isolated git worktrees — the pattern described in L0.6.
  • Related to: L0, L2
  • URL: https://github.com/obra/superpowers

jcodemunch

  • Type: Tool
  • Creator: Nicolo Ribaudo
  • Summary: Structured code search and symbol navigation tool that indexes repositories for AI-efficient code exploration, returning typed, summarized results instead of raw text.
  • Related to: L3
  • URL: https://github.com/nicolo-ribaudo/jcodemunch

RTK (Rust Token Killer)

  • Type: Tool
  • Creator: Community
  • Summary: CLI proxy that optimizes shell commands for token usage, providing 60-90% savings on development operations through intelligent command filtering.
  • Related to: L3
  • URL: https://github.com/rtk-ai/rtk

TOON (Token-Oriented Object Notation)

  • Type: Tool / Format
  • Creator: toon-format
  • Summary: Compact, human-readable encoding of the JSON data model designed for LLM prompts. Eliminates redundant structure in tabular data (the most common shape of MCP API responses) by declaring field names once in a header. Provides 60-65% token reduction on uniform object arrays with lossless round-trip serialization. Applied as middleware on MCP response paths to transparently optimize token usage without changing the agent's consumption model.
  • Related to: L3
  • URL: https://github.com/toon-format/toon

Claude Code (Anthropic)

  • Type: Tool
  • Creator: Anthropic
  • Summary: AI coding CLI with skills, hooks, and agentic capabilities that enhances developer productivity through intelligent code assistance and automation.
  • Related to: L2, L3
  • URL: https://claude.ai/code

Excalidraw Diagram Skill

  • Type: Tool / Skill
  • Creator: Cole Medin (coleam00)
  • Summary: Skill for generating Excalidraw diagram JSON files to visualize workflows, architectures, and concepts directly from agent sessions.
  • Related to: L0, L1
  • URL: https://github.com/coleam00/excalidraw-diagram-skill