Agents, commands & prompts

A curated repository of agent commands and prompt templates, specifically for the AEC (Architecture, Engineering, Construction). While it naturally includes general-purpose coding for day-to-day development, the primary goal is to provide construction-focused agents, commands, and prompts that address domain-specific tasks (e.g., BIM/parametric workflows, Dynamo/Grasshopper analysis, technical documentation, and AEC research).

Overview

This repository provides ready-to-use building blocks for effective, agent-driven development:

14 specialized agents across architecture, languages, testing, documentation, and AEC
2 workflow commands for context generation and experimentation
4 prompt templates for precision writing and accuracy
3 skills for AEC analysis and markdown conversion

All organized into 10 focused plugins for granular installation via Claude Code marketplace, or install skills via the manifest-driven scripts.

Repository structure

.claude-plugin/
└── marketplace.json         # Plugin marketplace definition
plugins/                     # Plugin-based organization (10 plugins)
├── aec-analysis-toolkit/
│   ├── agents/
│   └── skills/
├── backend-architecture/
│   └── agents/
├── python-development/
│   └── agents/
├── dotnet-development/
│   └── agents/
├── quality-and-testing/
│   └── agents/
├── documentation-and-visualization/
│   └── agents/
├── research-and-analysis/
│   └── agents/
├── workflow-orchestration/
│   └── commands/
├── prompt-templates/
│   └── commands/
└── markdown-utilities/
    └── skills/
skills-registry.json        # Skill manifest (name → plugin path)
install-skills.sh            # Installation script (Unix/macOS/Linux)
install-skills.bat           # Installation script (Windows)
README.md
LICENSE
CLAUDE.md

Installation

Option 1: Claude Code marketplace (recommended)

Add this marketplace to Claude Code for granular plugin installation:

/plugin marketplace add sonomirco/agents-and-commands

This makes 10 focused plugins available for installation without loading anything into context. Install only what you need:

# Install specific plugins
/plugin install aec-analysis-toolkit
/plugin install python-development
/plugin install backend-architecture

# Or install multiple at once
/plugin install aec-analysis-toolkit python-development quality-and-testing

Benefits:

Install only the agents, commands, and skills you need
Minimal token usage - plugins load only their specific resources
Easy updates via /plugin update
Browse available plugins with /plugin list

Option 2: Quick install (skills only)

Install all skills with one command:

# Unix/macOS/Linux
./install-skills.sh

# Windows
install-skills.bat

Or install specific skills:

./install-skills.sh grasshopper-analyzer dynamo-analyzer

Option 3: Manual installation

Inspect the manifest to view available skills and their plugin paths:

jq '.skills[] | {name, path}' skills-registry.json

Copy the desired skill directories (paths are manifest relative) to ~/.claude/skills/:

cp -r plugins/aec-analysis-toolkit/skills/grasshopper-analyzer ~/.claude/skills/
cp -r plugins/aec-analysis-toolkit/skills/dynamo-analyzer ~/.claude/skills/
cp -r plugins/markdown-utilities/skills/markdown-to-xml ~/.claude/skills/

Usage

Agent invocation

Use agents explicitly in natural language within Claude Code or Codex. Examples:

"Use python-pro to optimize this async function."
"Have backend-architect design the service boundaries for this feature."
"Ask tdd-orchestrator to drive the red-green-refactor cycle for this change."
"Use mermaid-expert to diagram the system interactions."

Commands

Command files in commands/ outline repeatable workflows. Open a command file and follow the steps, or paste sections into your assistant to run the flow.

commands/generate-code-base-context.md — Build a concise repository context for faster onboarding and reasoning
commands/experiment-development.md — Guide exploratory spikes and small experiments with safety rails

Prompts

Prompt templates in prompts/ help set intent and constraints quickly. AEC-focused analyzers include utilities to parse Dynamo and Grasshopper files and generate diagrams or summaries.

prompts/clear-and-accurate.md — Precision-first writing template
prompts/humanise-response.md — Make responses warmer without losing accuracy
prompts/reduce-hallucination.md — Improve reliability and source awareness
prompts/dynamo-analyser/ — Extract nodes/flows from Dynamo scripts; generate flow diagrams
prompts/grasshopper-analyser/ — Analyze .gh/.ghx graphs; extract C# nodes; diagram connections

Skills

Claude Code skills live inside plugin directories. skills-registry.json is the canonical manifest used by the installation scripts; update it whenever skills are added, moved, or removed. Copy the manifest-listed directories to ~/.claude/skills/ (or run the install scripts) to make them globally available, then invoke by name in natural language or via the Skill tool.

grasshopper-analyzer — Analyze Grasshopper (.ghx/.xml) files

Location: plugins/aec-analysis-toolkit/skills/grasshopper-analyzer/ (see manifest)
Input: Path to .ghx or .xml file (binary .gh not supported)
Output: Comprehensive Markdown report with workflow topology, custom scripts, algorithmic analysis, and C# migration guidance
Usage: "Use grasshopper-analyzer to analyze this definition"

dynamo-analyzer — Analyze Dynamo (.dyn) graphs

Location: plugins/aec-analysis-toolkit/skills/dynamo-analyzer/ (see manifest)
Input: Path to .dyn file (Dynamo 2.x JSON)
Output: Comprehensive Markdown report with workflow topology, Python/DesignScript code, Revit API patterns, and C# migration guidance
Usage: "Analyze this Dynamo graph with dynamo-analyzer"

markdown-to-xml — Convert markdown articles to XML format

Location: plugins/markdown-utilities/skills/markdown-to-xml/ (see manifest)
Input: Article name from Obsidian vault
Output: XML-formatted article with preserved formatting (bold, blockquotes, headers)
Usage: "Convert this article to XML"

Plan validation workflow

Use plan-validator to validate and iteratively improve plans using Codex feedback loops:

Claude drafts a plan → plan-validator crafts a Codex prompt and executes it
Codex streams output to the CLI; Claude reads it, updates the plan, and repeats until satisfied
Typical command (invoked by the agent): codex --full-auto exec --sandbox workspace-write "[prompt]"

Long-running validations can exceed 2 minutes. To prevent timeouts, set extended bash timeouts for Claude Code:

In ~/.claude/settings.json:

"env": {
  "BASH_DEFAULT_TIMEOUT_MS": "1800000",  // 30 minutes
  "BASH_MAX_TIMEOUT_MS": "7200000"      // 2 hours
}

Or export environment variables before starting the session:

export BASH_DEFAULT_TIMEOUT_MS=800000   # ~13 minutes
export BASH_MAX_TIMEOUT_MS=7200000      # 2 hours

Alternative: Run Codex as an MCP server and call it from Claude Code. This shifts execution to the MCP tool runtime and mitigates CLI timeout constraints for long operations.

Example Codex prompt used by plan-validator:

Review and validate the following implementation plan for the <feature/topic> in this repository.

Context:
- Repo root: $WORKSPACE
- Relevant dirs: src/, server/, scripts/

Please analyze:
1) Missing steps or dependencies
2) Risks and edge cases
3) Conflicts with current code
4) Concrete improvements (be specific)

If helpful, read files to verify assumptions. Then return:
- A concise feedback summary
- A numbered list of fixes/improvements
- Any commands, file changes, or tests to add

Plan:
<paste the full plan here>

MCP setup (optional):

You can expose Codex as an MCP server and register it with Claude Code.
Example config snippet (adjust to your environment):

{
  "mcpServers": {
    "codex": {
      "command": "codex",
      "args": ["mcp-server"],
      "env": {
        "BASH_DEFAULT_TIMEOUT_MS": "1800000",
        "BASH_MAX_TIMEOUT_MS": "7200000"
      }
    }
  }
}

Available plugins

The 14 agents, 2 commands, 4 prompts, and 3 skills are organized into 10 focused plugins optimized for granular installation:

Plugin	Description	Agents	Commands	Skills
aec-analysis-toolkit	Dynamo/Grasshopper analysis with C# migration guidance	2	0	2
backend-architecture	API design, microservices, security patterns	3	0	0
python-development	Modern Python 3.12+, async, FastAPI/Django	2	0	0
dotnet-development	C#/.NET with testing and performance	1	0	0
quality-and-testing	TDD, plan validation, Codex integration	2	0	0
documentation-and-visualization	Diagrams, prompts, technical writing	2	0	0
research-and-analysis	Codebase research, document parsing	2	0	0
workflow-orchestration	Context generation, experimental development	0	2	0
prompt-templates	Precision writing, accuracy templates	0	3	0
markdown-utilities	Markdown to XML conversion	0	0	1

Agent categories

Architecture & system design

backend-architect (opus) — Design RESTful APIs, service boundaries, schemas
backend-security-coder (opus) — Secure backend patterns and hardening
api-documenter (sonnet) — OpenAPI and developer docs

Programming languages

python-pro (sonnet) — Modern Python 3.12+, async, performance
python-developer (sonnet) — Practical Python, Django/FastAPI
csharp-pro (sonnet) — Modern C#/.NET with testing and performance

Quality & testing

tdd-orchestrator (opus) — Red/green/refactor workflow leadership
plan-validator (opus) — Validates and iteratively upgrades plans via Codex feedback loops

Documentation & visualization

mermaid-expert (sonnet) — Diagrams for systems and flows
prompt-engineer (opus) — Design and refine system prompts

Research & analysis

codebase-researcher (opus) — Study and summarize codebases
document-parser-searcher (haiku) — Parse and search large document sets

AEC domain

aec-research-consultant (opus) — Industry research and analysis
grasshopper-analyzer (sonnet) — Analyze Grasshopper files and extract C# nodes

Model configuration

Agents specify a target Claude model when appropriate:

haiku: 1
sonnet: 6
opus: 7

Use lighter models for quick, deterministic tasks; sonnet for day-to-day engineering; opus for complex reasoning, architecture, and rigorous analysis.

Contributing

To add new agents, commands, prompts, or skills:

Keep everything generic: All contributions must be generic and reusable. Carefully eliminate any specific references to:
- Personal projects or proprietary codebases
- Company names or internal tools
- Specific file paths or directory structures
- Personal workflows or Obsidian vaults
- API endpoints or internal services
File naming: Use lowercase, hyphen-separated file names
File placement: Place in the appropriate plugin directory (e.g., plugins/<plugin>/agents/, plugins/<plugin>/commands/, plugins/<plugin>/skills/)
Content quality: Keep concise, actionable, and production-focused
Frontmatter: For agents, include at least name, description, and optionally model

Generic examples

❌ Bad (specific): "Fetch data from the CompanyX API at https://internal.companyx.com/api" ✅ Good (generic): "Fetch data from the API endpoint"

❌ Bad (specific): "Read files from /Users/john/Documents/MyProject/" ✅ Good (generic): "Read files from the project directory"

❌ Bad (specific): "Connect to our MongoDB instance at prod-db-01" ✅ Good (generic): "Connect to the database"

Templates

For agent templates, check the Claude Code documentation

For skill templates, check the Claude Code skills documentation

Sharing and distribution

Option 1: Claude Code marketplace (recommended)

This repository is configured as a Claude Code plugin marketplace. Users simply add the marketplace:

/plugin marketplace add yourusername/agents-and-commands

Then install specific plugins:

/plugin install aec-analysis-toolkit python-development

Benefits for users:

Browse 10 focused plugins with /plugin list
Install only what they need
Minimal token usage (plugins load only their resources)
Easy updates with /plugin update

Option 2: Direct installation (skills only)

Users clone and run installation scripts:

git clone https://github.com/sonomirco/agents-and-commands.git
cd agents-and-commands
./install-skills.sh

Option 3: Individual resources

Copy specific agents, commands, prompts, or skills:

Copy agents/agent-name.md to .claude/agents/
Copy manifest-listed skill directories (e.g., plugins/aec-analysis-toolkit/skills/grasshopper-analyzer/) to ~/.claude/skills/
Copy commands or prompts as needed

Creating your own plugin marketplace

Fork this repository
Update .claude-plugin/marketplace.json:
- Change owner name, email, URL
- Update repository URLs (replace sonomirco with your GitHub username)
- Customize plugin descriptions
Add/remove plugins to match your content
Ensure all paths in marketplace.json are correct
Push to GitHub
Share with: /plugin marketplace add your-github-username/your-repo

Marketplace structure:

{
  "name": "your-marketplace-name",
  "owner": { "name": "...", "email": "...", "url": "..." },
  "metadata": { "description": "...", "version": "1.0.0" },
  "plugins": [
    {
      "name": "plugin-name",
      "description": "...",
      "agents": ["./plugins/sample-plugin/agents/agent.md"],
      "commands": ["./plugins/sample-plugin/commands/command.md"],
      "skills": ["./plugins/sample-plugin/skills/skill/SKILL.md"]
    }
  ]
}

License

Apache License 2.0 — see LICENSE for details.

Resources

Claude Code documentation: https://docs.anthropic.com/en/docs/claude-code
Subagents: https://docs.anthropic.com/en/docs/claude-code/sub-agents
Claude Code on GitHub: https://github.com/anthropics/claude-code
Awesome Claude Code: https://github.com/hesreallyhim/awesome-claude-code
Claude Code Templates: https://www.aitmpl.com
Awesome Codex: https://github.com/KarelDO/awesome-codex

Name		Name	Last commit message	Last commit date
Latest commit History 23 Commits
.claude-plugin		.claude-plugin
plugins		plugins
prompts		prompts
.gitignore		.gitignore
AGENTS.md		AGENTS.md
CLAUDE.md		CLAUDE.md
LICENSE		LICENSE
README.md		README.md
install-skills.bat		install-skills.bat
install-skills.sh		install-skills.sh
skills-registry.json		skills-registry.json

Uh oh!

License

Uh oh!

sonomirco/agents-and-commands

Folders and files

Latest commit

History

Repository files navigation

Agents, commands & prompts

Overview

Repository structure

Installation

Option 1: Claude Code marketplace (recommended)

Option 2: Quick install (skills only)

Option 3: Manual installation

Usage

Agent invocation

Commands

Prompts

Skills

Plan validation workflow

Available plugins

Agent categories

Architecture & system design

Programming languages

Quality & testing

Documentation & visualization

Research & analysis

AEC domain

Model configuration

Contributing

Generic examples

Templates

Sharing and distribution

Option 1: Claude Code marketplace (recommended)

Option 2: Direct installation (skills only)

Option 3: Individual resources

Creating your own plugin marketplace

License

Resources

About

Topics

Resources

License

Code of conduct

Uh oh!

Stars

Watchers

Forks

Languages