Engineer in headphones at a desk with an FPGA board, facing a monitor where editor, build log, waveforms and timing report are joined in a glowing loop
FPGAAGENTIC AITOOLS 4 min read

FPGA Agentic AI Resources

Resources for agentic AI in FPGA design: agent skills including hdldiagZero, the open-source tools an agent can drive from the command line, and the standards behind skills and tool access.

By Leonardo Capossio
Contact us Subscribe

A running list of things that make an agentic loop useful on FPGA work: skills that teach an agent a specific job, tools it can drive from a command line, and the standards underneath. It is a companion to Agentic AI for FPGA Design: How the Loop Works, and it grows as we find or build something worth listing. Nothing here is sponsored, and we use what we list.

Skills

A skill is a folder with a SKILL.md file in it: a name, a description, and instructions for one job, optionally with scripts and reference files alongside. The agent loads only the name and description at startup and reads the full instructions when a task matches, so a set of skills costs almost nothing until one is needed.[1] The format started at Anthropic, is published as an open standard, and is now read by a long list of agents rather than one.[2]

For FPGA work this matters because the useful knowledge is procedural. Which report to open after synthesis, what a warning actually means, how your project names its files: that is exactly what a skill holds, and exactly what you would otherwise retype into a prompt every session.

hdldiagZero bard0 design

Turns an HDL, RTL or SoC architecture description into a clean SVG block diagram. It walks the hierarchy from the top module, classifies the interfaces it finds, colours blocks by clock domain, marks clock domain crossings, and leaves out the clock, reset, JTAG and debug signals that make a diagram unreadable. The geometry is validated, so lines do not run through blocks.

The diagram nobody drew, or the one that stopped matching the code three revisions ago. Ask the agent to draw the top level and you get an SVG you can put in a review, a specification or a handover.

Install it in Claude Code as a plugin, or run python install.py from the repository for other runtimes. Python 3.10 or newer, standard library only, MIT licensed.

github.com/lcapossio/hdldiagZero

Tools the loop drives

The loop needs tools that run from a command line and report in text. These are the open-source ones the companion repository uses or builds on.

Icarus Verilog

Verilog and SystemVerilog simulator. Compiles and runs a testbench from two commands and writes whatever the testbench prints to standard output.

steveicarus.github.io/iverilog

GHDL

VHDL simulator, and a synthesis front end that can hand a design to Yosys. The VHDL half of the companion repository runs on it.

github.com/ghdl/ghdl

Yosys

Open-source synthesis. Gives the agent a vendor-independent check that a design elaborates, infers no latches, and costs roughly what you expected in cells and flops, without waiting for a vendor flow.

github.com/YosysHQ/yosys

OSS CAD Suite

A single binary bundle of the open-source flow: Yosys, GHDL, Icarus Verilog, nextpnr, the formal tools and the programmers, for Linux, macOS and Windows. The fastest way to get an agent a working toolchain on a fresh machine.

github.com/YosysHQ/oss-cad-suite-build

cocotb

Testbenches written in Python, driving VHDL or SystemVerilog through the simulator you already use. Worth knowing about because agents write Python more reliably than they write SystemVerilog testbenches.

cocotb.org

Standards worth knowing

Agent Skills

The open format behind the skills above: a folder, a SKILL.md, optional scripts and references. Write a skill once and it is read by any agent that supports the standard.

agentskills.io

Model Context Protocol

The open standard for connecting an agent to an external system, so a tool or a data source can be exposed once and used by any client that speaks it. Relevant when the thing you want the agent to reach is a server or a database rather than a command.

modelcontextprotocol.io

The agents themselves

Installing and signing in to Claude Code, OpenAI's Codex CLI or Cursor is covered separately, in Setting Up Claude Code and Codex CLI for the FPGA Loop, along with the vendors' own guides and videos.

Something missing? If you use a skill, tool or write-up that earns its place on this list, tell us and we will look at it. We list what we have used, so a suggestion is a starting point rather than a listing.

References

2 sources
  1. Anthropic, "Agent Skills" in the Claude Code documentation: skill structure, where skills are installed, and how they load.
  2. "Agent Skills": the open standard, its specification, and the clients that support it.
Engineer at a desk facing a curved monitor showing an FPGA design flow as a loop of code editor, build log, waveform viewer and timing report, with the title FPGA Agentic Loop
Related article Agentic AI for FPGA Design: How the Loop Works An introduction to agentic AI for FPGA design: what separates an agent from a chatbot, what the plan, act, observe, iterate loop looks like on RTL, and what it takes to set one up.
An engineer at a workstation using an MCP-based Vivado chatbot, its answer shown beside Vivado waveforms, a schematic, and a timing report, with an FPGA board on the desk
Related article Using Vivado's New AI Chatbot: A Practical Guide A practical guide to AMD's Vivado chatbot: what to ask, where it helps, where it can mislead, and how to verify its answers against docs and tool reports.

Our FPGA and AI training sets this loop up on your own tools and designs, with the checks and limits in place.

Discuss your project with our engineering team

Comments

Loading comments...

Leave a Comment

By posting a comment, you agree that we may store your submitted details for moderation and site operation. See the privacy policy.

We use analytics and marketing tools to analyze traffic and improve our services. By clicking "Accept", you agree to our use of these tools.