samkeen

PR Decomposition Agent Prompt

Agent Instructions

You are a senior staff engineer acting as a PR decomposition facilitator. You've been given a diff that is far too large to review meaningfully. Your job is to work with the PR creator through a structured interview to:

1. Understand the core motivation — the single problem that triggered this work
2. Identify what's essential vs. what got bundled in along the way
3. Surface opportunities for simplification — places where the same intent could be achieved with less code, less complexity, or a more incremental approach
4. Make a deliberate decision about every piece of intent — ship it, spec it, track it, simplify it, or drop it

samkeen

 * ▐▛███▜▌ *   Claude Code v2.0.76
* ▝▜█████▛▘ *  Opus 4.5 · Claude Max
 *  ▘▘ ▝▝  *   ~/Projects/revenue-experiments/expenses-sbucks

> use your ralph task setup skill to gather requirements for gh issue 4 
  ⎿  ⧉ Selected 1 lines from README.md in Visual Studio Code

⏺ Bash(gh issue view 4)
  ⎿  title:	Add receipt upload UI  

samkeen

Dev Journal: January 3-5, 2026

You had a productive 3-day stretch focused on three main areas: advancing fast-llms-txt (your FastAPI OpenAPI-to-markdown library), polishing chat-rag as an educational tool, and significant business/marketing work for AlteredCraft.

---

Friday, January 3

fast-llms-txt - Schema documentation improvements

• Worked through nested object extraction for arrays (e.g., Array[HistoryEvent]) to improve markdown output quality
• Added a premise section to the README explaining why OpenAPI specs, while excellent for machines, need LLM-optimized formats

samkeen

---

Dev Journal: January 3-6, 2026

A productive 4-day stretch focused on three main projects. The highlight was building out the dev-journal skill for Claude Explorer—creating an API-powered workflow that lets an AI agent journal your development activity. You also shipped several UX improvements to Chat-RAG and refined the markdown output for fast-llms-txt.

---

Monday, January 6

Claude Explorer - Dev journal skill and API refinements

• Crafted a system prompt for an LLM agent to generate developer journals from the Claude Explorer API

samkeen

description	research_date	claude_code_version	sources
Documentation of the `~/.claude/` directory structure for dev journal integration.	2026-01-01	2.0.x	Local filesystem analysis + [Official Docs](https://code.claude.com/docs)

Claude Code Data Structures

Overview

samkeen

╭─── Claude Code v2.0.55 ──────────────────────────────────────────────────────╮
│                            │ Tips for getting started                        │
│        Welcome back!       │ Run /init to create a CLAUDE.md file with inst… │
│                            │ ─────────────────────────────────────────────── │
│           ▐▛███▜▌          │ Recent activity                                 │
│          ▝▜█████▛▘         │ No recent activity                              │
│            ▘▘ ▝▝           │                                                 │
│                            │                                                 │
│   Opus 4.5 · Claude API    │                                                 │

samkeen

# Example create PR skill

## Layout
```
skills/create-pr
├── references
│   └── pr-best-practices.md
└── SKILL.md
```

samkeen

Can you add an export trace button to the ui

I'll help you add an export trace button to the UI. Let me first explore the codebase structure to understand the current implementation.

Task
Explore UI codebase structure
Bash
find /home/user/mcp_visualizer -type f -name "*.tsx" -o -name "*.ts" -o -name "*.jsx" -o -name "*.js" | head -30
/home/user/mcp_visualizer/mcp-inspector-app/app/test-module-2/page.tsx
/home/user/mcp_visualizer/mcp-inspector-app/app/demo/page.tsx

samkeen

Review Efficiency Prompt Template

Purpose: This template helps you engineer AI-generated artifacts that are optimized for efficient human review. Based on the principles from "Engineering AI Generated Artifacts for Human Review" (AlteredCraft).

The Core Insight: When AI generates artifacts, the marginal cost of including full context, rationale, and alternatives is minimal—but the review time saved is massive. This template embeds those requirements into your prompts.

Our Approach to Context

Advanced context tools exist (MCP servers, vector databases, ...), but this template takes a simpler path: leveraging AI coding agents' native ability to access file paths and URLs you provide.

samkeen

Of course. Here is the updated constitution with those three principles integrated.

---

Core Philosophy & Architecture 🏛️

• Follow SOLID Principles: Adhere strictly to the five SOLID principles (Single Responsibility, Open/Closed, Liskov Substitution, Interface Segregation, Dependency Inversion) to create a modular, scalable, and maintainable architecture.
• Explicit is Better than Implicit: Code should be clear and self-evident. Avoid clever tricks or "magic" that hide the underlying logic. The flow of data and control should be easy to follow.
• Simplicity and YAGNI: Prioritize simplicity and adhere to the "You Ain't Gonna Need It" principle. Do not add functionality or complexity until it is actively required.
• Manage State Predictably: Application state must be handled in a centralized and predictable manner. Avoid scattering state logic throughout the codebase. State changes should be explicit and traceable, minimizing side effects and making the application easier to debug.