Token导航 LogoToken导航TokenDH.com
前端设计权限需确认github未标认证来源可访问许可证需确认审计通过

claude-skill-verilogClaude 技能 verilog

Agent Skill

claude-skill-verilog 用于处理 GitHub 仓库、Issue、Pull Request 和代码协作信息,适合在 Codex、Claude、Cursor、Gemini CLI 中需要围绕仓库状态、代码变更或协作事项进行整理时使用。可结合来源仓库、安装命令和原始 README 继续核验具体用法。安装前建议确认权限范围、维护状态,以及是否会触发联网、命令执行或文件读写。

总安装

1,536

周安装

66

GitHub Stars

8

下载量

539
CodexClaudeCursorGemini CLI

安装说明

本站只整理中文说明和来源信息,不托管安装包,也不代用户安装。

GitHub

来源数

2

许可证

unknown

最后核验

2026-05-01

来源状态

来源可访问

安装方式

通过对话安装

复制提示词发给支持本地命令或 Skills 的 AI 助手,先确认命令和权限,再让它执行。

请帮我安装这个 Agent Skill:claude-skill-verilog(Claude 技能 verilog)
来源仓库:https://github.com/londey/claude-skill-verilog
仓库路径:skills/claude-skill-verilog
安装命令:
npx skills add https://github.com/londey/claude-skill-verilog --skill claude-skill-verilog
安装前请先检查当前环境是否支持对应 CLI,并向我确认将要执行的命令、安装目录、联网范围和文件读写权限;确认后再执行。

命令行安装

复制命令到本机终端执行。该命令会通过 npx skills 从第三方来源获取 Skill;本站只展示命令,不托管安装包,也不自动执行。

skills.shnpx skills
npx skills add https://github.com/londey/claude-skill-verilog --skill claude-skill-verilog

简介

Verilog/SystemVerilog 指导技能提供硬件描述语言开发规范。

  • 强制要求模块、信号与寄存器添加详细注释,采用 Qm.n 定点数标注法。
  • 适用于 FPGA 或 ASIC 设计项目,确保代码可读性与仿真一致性。
  • 需配合综合工具验证语法合规性,避免仅依赖本技能输出作为最终依据。
  • claude-skill-verilog 属于前端设计类 Skill,可作为该场景下的辅助能力补充。

SKILL.md

Verilog/SystemVerilog Guidance

Apply when working with .v, .sv, .vh, .svh files or running Verilator.

Documentation

All modules, wires, and registers require comments:

// Module: counter
// Purpose: Simple up-counter with synchronous reset
module counter #(
    parameter WIDTH = 8  // Counter bit width
) (
    input  logic             clk,      // System clock
    input  logic             rst_n,    // Active-low reset
    output logic [WIDTH-1:0] count     // Current count value
);

Fixed-Point Notation

Document all fixed-point values using TI-style Q notation:

  • Qm.n — signed: m integer bits (including sign bit), n fractional bits, total width = m + n bits.
  • UQm.n — unsigned: m integer bits, n fractional bits, total width = m + n bits.

Use Q notation in signal comments, localparam descriptions, and module-level documentation.

logic signed [15:0] scale_factor;  // Scaling multiplier, Q4.12
logic        [15:0] timer_count;   // Free-running tick counter, UQ16.0
logic signed [15:0] delta;         // Sample-to-sample difference, Q4.12

Naming Conventions

  • Active-low signals: use _n suffix (e.g., rst_n, chip_select_n)
  • Clocks: clk or clk_<domain>
  • Use descriptive names over abbreviations

always_ff: Simple Assignments Only

always_ff blocks must contain ONLY simple non-blocking assignments. No logic, no expressions — this ensures Verilator simulation matches synthesized behavior. (Exceptions: memory inference and async reset synchronizers require conditional logic — see those sections.)

// CORRECT - simple assignment
always_ff @(posedge clk) begin
    count <= count_next;
    state <= state_next;
end

// WRONG - logic in always_ff
always_ff @(posedge clk) begin
    count <= count + 1;                // Move to always_comb
    state <= enable ? RUNNING : IDLE;  // Move to always_comb
end

always_comb: All Logic Here

All combinational logic belongs in always_comb blocks:

always_comb begin
    count_next = count + 8'd1;
    state_next = enable ? RUNNING : IDLE;
end

Formatting

  • One statement per line — never chain multiple statements or assignments on a single line
  • One declaration per line
  • Explicit bit widths on all literals
  • Start files with ` default_nettype none ``
  • Always use begin/end blocks for if, else, case items (prevents bugs when adding code later)
  • Prefer to keep modules under ~500 lines; if a module grows significantly larger, consider refactoring into smaller sub-modules
`default_nettype none

module example (
    input  logic        clk,
    input  logic        rst_n,
    input  logic [7:0]  data_in,
    output logic [7:0]  data_out
);

    logic [7:0] data_reg;    // Registered data
    logic [7:0] data_next;   // Next state value
    logic       valid;       // Data valid flag

    localparam logic [7:0] INIT_VAL = 8'd0;

endmodule

`default_nettype wire

Yosys Synthesis Compatibility

Synthesizable RTL should work with both Verilator (lint/simulation) and Yosys (synthesis for open-source FPGA flows). Yosys supports a subset of SystemVerilog via read_verilog -sv. Code that passes Verilator may still fail Yosys synthesis.

Constructs to avoid in synthesizable RTL:

AvoidUse instead
return <expr>; in functionsfunction_name = <expr>; (Verilog-2005 style)
interface / modportExplicit port lists
unique case / priority casePlain case with default
Multi-dimensional packed arrays in portsFlatten to single vectors
// CORRECT - Yosys-compatible function
function automatic logic [7:0] add_saturate(input logic [7:0] a, input logic [7:0] b);
    logic [8:0] sum;
    sum = {1'b0, a} + {1'b0, b};
    add_saturate = sum[8] ? 8'hFF : sum[7:0];
endfunction

// WRONG - return statement (Yosys rejects this)
function automatic logic [7:0] add_saturate(input logic [7:0] a, input logic [7:0] b);
    logic [8:0] sum;
    sum = {1'b0, a} + {1'b0, b};
    return sum[8] ? 8'hFF : sum[7:0];
endfunction

Always verify with the actual synthesis flow (e.g. yosys -p "synth_ecp5...", yosys -p "synth_ice40...", or the project's build target), not only verilator --lint-only, when using SystemVerilog features. Lint-clean does not imply synthesizable.

Testing with Verilator

Every module requires a testbench. Build and run with Verilator:

# Build testbench
verilator --binary -Wall module_tb.sv module.sv

# Run simulation
./obj_dir/Vmodule_tb

Testbench structure:

module counter_tb;
    logic       clk = 1'b0;  // System clock
    logic       rst_n;       // Active-low reset
    logic [7:0] count;       // DUT output

    counter dut (
        .clk(clk),
        .rst_n(rst_n),
        .count(count)
    );

    always begin
        #5 clk = ~clk;
    end

    initial begin
        rst_n = 1'b0;
        #20 rst_n = 1'b1;
        #100;
        $display("Test complete, count=%d", count);
        $finish;
    end
endmodule

Verilator Linting

Run linting on all files and fix all warnings:

verilator --lint-only -Wall module.sv
  • Fix all warnings — do not suppress with pragmas
  • Key warnings: WIDTH (bit-width mismatch), UNUSED, UNDRIVEN

Verilator Simulation Flags

Recommended flags for simulation builds:

verilator --binary \
    -Wall \
    -Wno-fatal \
    -j 0 \
    --assert \
    --timing \
    --trace-fst \
    --trace-structs \
    --main-top-name "-" \
    --x-assign unique \
    --x-initial unique \
    module_tb.sv module.sv
FlagPurpose
-WallEnable all warnings
-Wno-fatalDon't exit on warnings (allows full report)
-j 0Fully parallelized compilation
--assertEnable SystemVerilog assertions
--timingEnable timing constructs
--trace-fstDump waveforms as FST (compressed)
--trace-structsHuman-readable struct dumps
--main-top-name "-"Remove extra TOP module wrapper
--x-assign uniqueReplace X with random constant per-build
--x-initial uniqueRandomly initialize uninitialized variables

Module Instantiation

  • One module per file, filename matches module name
  • Always use named port connections (never positional)
// CORRECT - named connections
counter #(
    .WIDTH(16)
) u_counter (
    .clk    (clk),
    .rst_n  (rst_n),
    .count  (count_value)
);

// WRONG - positional connections
counter u_counter (clk, rst_n, count_value);

Avoiding Latches

Latches are inferred when signals aren't assigned in all paths. Prevent with:

  • Default assignments at start of always_comb
  • Cover all cases including default
always_comb begin
    // Default assignments first
    data_next = data_reg;
    valid_next = 1'b0;

    case (state)
        IDLE: begin
            data_next = 8'd0;
        end
        LOAD: begin
            data_next = data_in;
        end
        default: begin
            data_next = data_reg;
        end
    endcase
end

Reset Handling

Use synchronous resets when possible. For external async resets, synchronize first.

Note: Async reset synchronizers require conditional logic in always_ff for the reset condition — this is a necessary exception similar to memory inference.

// Synchronous reset (preferred)
logic [7:0] count;       // Counter register
logic [7:0] count_next;  // Next counter value

always_comb begin
    count_next = rst_n ? (count + 8'd1) : 8'd0;
end

always_ff @(posedge clk) begin
    count <= count_next;
end

// Reset synchronizer for external async reset
logic [1:0] rst_sync;       // Synchronizer flip-flops
logic [1:0] rst_sync_next;  // Next synchronizer value

always_comb begin
    rst_sync_next = {rst_sync[0], 1'b1};
end

always_ff @(posedge clk or negedge rst_async_n) begin
    if (!rst_async_n) begin
        rst_sync <= 2'b00;
    end else begin
        rst_sync <= rst_sync_next;
    end
end
assign rst_n = rst_sync[1];

FSM Patterns

Separate state register from next-state logic. Use enums for state encoding.

typedef enum logic [1:0] {
    IDLE,
    RUN,
    DONE
} state_t;

state_t state;       // Current state register
state_t state_next;  // Next state value

// Next-state logic (combinational)
always_comb begin
    state_next = state;
    case (state)
        IDLE: begin
            if (start) begin
                state_next = RUN;
            end
        end
        RUN: begin
            if (finish) begin
                state_next = DONE;
            end
        end
        DONE: begin
            state_next = IDLE;
        end
        default: begin
            state_next = IDLE;
        end
    endcase
end

// State register (sequential)
always_ff @(posedge clk) begin
    state <= state_next;
end

Clock Domain Crossing (CDC)

Single-bit signals: use 2-FF synchronizer. Multi-bit: use gray coding or handshake.

// 2-FF synchronizer for single-bit CDC
logic [1:0] sync_reg;       // Synchronizer flip-flops
logic [1:0] sync_reg_next;  // Next synchronizer value
logic       signal_sync;    // Synchronized output

always_comb begin
    sync_reg_next = {sync_reg[0], signal_src};
end

always_ff @(posedge clk_dst) begin
    sync_reg <= sync_reg_next;
end
assign signal_sync = sync_reg[1];

// Gray code for multi-bit counters crossing domains
function automatic logic [WIDTH-1:0] bin2gray(input logic [WIDTH-1:0] bin);
    bin2gray = bin ^ (bin >> 1);
endfunction

Memory Inference

Use standard patterns for RAM/ROM inference by synthesis tools.

Note: Memory patterns are an exception to the "simple assignments only" rule for always_ff. Synthesis tools require these specific patterns to correctly infer RAM/ROM primitives.

// Single-port RAM
logic [DATA_WIDTH-1:0] mem [0:DEPTH-1];  // Memory array

always_ff @(posedge clk) begin
    if (we) begin
        mem[addr] <= wdata;
    end
    rdata <= mem[addr];
end

// ROM (initialized memory)
logic [7:0] rom [0:255];  // ROM array
initial $readmemh("rom_data.hex", rom);

always_ff @(posedge clk) begin
    rdata <= rom[addr];
end

Assertions (SVA)

Use assertions for verification. They're enabled with --assert in Verilator.

// Immediate assertion
always_comb begin
    assert (count < MAX_COUNT) else $error("Count overflow");
end

// Concurrent assertions
property p_valid_handshake;
    @(posedge clk) disable iff (!rst_n)
    valid |-> ##[1:3] ready;
endproperty

assert property (p_valid_handshake)
    else $error("Handshake timeout");

// Cover property (for functional coverage)
cover property (@(posedge clk) state == DONE);

适合场景

01

用户想查找某类 Agent Skill 时

02

需要根据任务场景推荐可安装能力包时

03

需要对比不同来源的安装命令和来源信息时

能力概览

能力 1

按任务关键词查找相关 Skills

能力 2

展示可复制的安装命令

能力 3

保留来源站点、仓库和原始说明,方便继续核验

能力 4

展示第三方安全扫描或审计结果

安装后应在对应宿主中按原始 README 的触发条件使用;具体调用方式请以来源页面和 README 为准。

平台分布

Codex

34.14%
按下载量换算184

Claude

30.13%
按下载量换算162

Cursor

17.88%
按下载量换算96

Gemini CLI

9.16%
按下载量换算49

安全审计

Gen Agent Trust Hub

通过

Socket

通过

Snyk

通过

权限和风险

权限需确认

当前来源未能明确判断权限范围,默认进入异常复核队列。

安装前确认

本站仅展示第三方公开信息,不托管安装包,不提供自动安装或运行环境。安装前应自行审查源码、依赖和命令行为。当前只有一个来源,正式发布前建议补源仓库或其他目录站核验。

来源信息

继续浏览同类 Skills