Teaching Agents New Tricks via Agent Skills

A presentation and technical exploration by Brett Morgan categorizing how Agent Skills (following the agentskills.io open specification) are authored, generated, and applied in practice—using the reverse-engineering of embedded firmware for the this.is.NOISE Gamma mini synth (powered by an Electro-Smith Daisy Seed 2 DFM) as a real-world case study.

📊 Live Slide Deck

Related Resources

• Companion Hardware & Firmware Repository: github.com/domesticmouse/reverse-engineering-gamma
• Blog Post: Reverse Engineering a this.is.NOISE Gamma Synth (Medium)

---

The "Three Ways" Taxonomy

When reverse-engineering the Gamma mini synth without schematics or source code, seven custom Agent Skills emerged across three distinct authoring patterns (plus a capstone mashup):

1. Way 1: Library Review Skills

   • libdaisy-guide & daisysp-guide
   • Created by having an agent inspect C++ library source trees (libDaisy and DaisySP), synthesize architectural and real-time DSP reference reports, and package those findings into reusable skills.

2. Way 2: DFU Operational Skills

   • gamma-firmware-flash & gamma-firmware-restore
   • Created by solving USB-C dfu-util bootloader flashing requirements (BOOT_SRAM 0x24000000 vector table relocation and the ~2.5s Daisy bootloader grace window) and wrapping the workflow into deterministic Python scripts and operational runbooks.

3. Way 3: Hardware Discovery Skills

   • gamma-pinout & gamma-hardware-controls
   • Distilled from a phased series of interactive diagnostic firmwares that mapped physical pins, ADC channels, polarities, and C++ driver idioms for the 14 tactile keys, 4 potentiometers, 2 dual-axis thumbsticks, rotary encoder, and SSD1306 OLED display.

4. The Capstone Mashup Skill

   • gamma-usb-connectivity
   • Combines all three patterns into a single skill package: discovered USB pinout facts (UsbHandle::FS_EXTERNAL on D29/D30), a deadlock-immune C++ firmware driver (gamma_usb.h), and a host-side Python communication utility (gamma_usb.py).

---

Repository Structure

• slides/ — Slidev presentation deck, custom Vue components, layouts, and acts (01-primer.md through 06-experiment-and-conclusion.md).
• .agents/skills/ — Agent Skills packages used both in the Gamma mini synth case study and for authoring/maintaining this repository.
• terminal-recordings/ — Sanitized asciinema .cast recordings embedded in the presentation.
• GLOSSARY.md — Canonical project terminology and domain language.

---

Running the Slide Deck Locally

The presentation is built with Slidev (Vue 3 + Vite).

cd slides
npm install
npm run dev

To build the static bundle locally:

cd slides
npm run build

---

License & Disclaimer

Licensed under the Apache License, Version 2.0. See LICENSE for full terms and conditions.

Disclaimer: This is not an official Google product.