rafayels-engineering

rafayels-engineering

Rafayel's engineering toolkit for Claude Code, OpenCode, and now Pi.

The main/global install is now intentionally bootstrap-only: it exposes /re:new_repo and /re:existing_repo, then those commands install and customize the prompts and skills a specific repository needs.

This repo stays compatible with Claude/OpenCode project-local resources while also shipping a native Pi package with extensions for the tools that matter most:

• memory
• project-config
• browser / Playwright-style automation
• Figma access

Compatibility

Harness	Status	Entry point
Claude Code	Supported	.claude-plugin/plugin.json
OpenCode	Supported	.opencode/
Pi	Supported	package.json + extensions/

What changed for Pi

This repository now includes a Pi package manifest and native Pi extensions under extensions/.

Native Pi extensions

• extensions/memory.ts
• extensions/project-config.ts
• extensions/playwright.ts
• extensions/figma.ts
• extensions/ask-user-question.ts
• extensions/issue-tracker.ts
• extensions/toon.ts (global Toon preprocessor for tool output)

These extensions are additive. They do not replace the existing Claude/OpenCode plugin files.

New workflow primitives

This package now also includes two Pi-native workflow tools:

• ask_user_question — interactive single-question user input for workflow pauses and clarifications
• issue_tracker — project-local structured issue tracking backed by JSON files in .pi/issues/

Installation

Use the installer:

./install.sh

The installer lets you choose:

• target agent: Pi / Claude Code / OpenCode / all
• project-local vs global where relevant

See also: INSTALL.md

Pi usage

Install globally:

pi install /path/to/rafayels-engineering

Install into the current project only:

pi install -l /path/to/rafayels-engineering

Both install modes expose only the bootstrap prompts by default:

• /re:new_repo
• /re:existing_repo

Run one of those commands inside or near a repository to copy and customize project-local workflow prompts and skills from references/project-resources/.

Toon preprocessing (Pi)

extensions/toon.ts now runs a transparent Pi-native preprocessing pipeline for Bash tools:

1. It tries to rewrite each command through rtk rewrite first (when available).
2. It always applies Toon-style JSON compression to tool output in Pi when available.

This is enabled by default and works without special env toggles.

Modes are mainly for control/rollback:

export PI_TOOL_PREPROCESSOR=auto   # (default) rewrite with RTK when possible, then encode output with Toon
export PI_TOOL_PREPROCESSOR=rtk    # force RTK rewrite attempt first, then Toon encoding
export PI_TOOL_PREPROCESSOR=toon   # skip RTK, only do Toon output encoding
export PI_TOOL_PREPROCESSOR=off    # disable preprocessing entirely

# Optional binary/script overrides
export TOON_BIN=/opt/homebrew/bin/toon
export RTK_BIN=/opt/homebrew/bin/rtk
export TOON_DETECT_SCRIPT=/path/to/custom/toon-detect.sh

Memory setup

Initialize memory after /re:new_repo or /re:existing_repo installs the memory skill into a project (memory is optional and not part of the minimal core set):

# Claude/Pi project-local skill install
python3 .claude/skills/memory/scripts/memory.py init

# OpenCode project-local skill install
python3 .opencode/skills/memory/scripts/memory.py init

If your machine needs a specific interpreter:

export PYTHON_FOR_RAFAYELS_ENGINEERING=/opt/homebrew/bin/python3.12

Migration notes

Claude / OpenCode users

Nothing is removed:

• existing plugin manifests remain
• codex-related files remain
• MCP-based Claude flows remain available

Pi users

Pi uses native extensions instead of relying on Claude MCP wiring.

Examples:

• memory becomes Pi tools like memory_query and memory_write
• project config becomes Pi tools like get_config_value and init_config
• browser automation is exposed as Pi tools and also compatibility aliases for existing prompt content
• Figma is exposed through native HTTP-based Pi tools

Browser compatibility layer

To reduce prompt churn, the Pi Playwright bridge also registers compatibility aliases matching existing prompt usage, such as:

• mcp__plugin_compound-engineering_pw__browser_navigate
• mcp__plugin_compound-engineering_pw__browser_snapshot
• mcp__plugin_compound-engineering_pw__browser_take_screenshot
• mcp__plugin_compound-engineering_pw__browser_console_messages

That means existing command text can keep working while the underlying execution is Pi-native.

Figma support

Pi-native Figma tools currently include:

• figma_get_node
• figma_get_node_from_url
• figma_get_file
• figma_get_image

Requires:

export FIGMA_API_KEY=...

Project config support