Ironforge
A Claude Code plugin that orchestrates six development tools and conditional LSPs into a composable, skill-based development workflow.
What It Does
Ironforge installs and coordinates the following tools:
Permanent Tools (all projects)
| Tool | Role |
|---|
| BMAD | Specialized agents: @bmad-agent-analyst, @bmad-agent-pm, @bmad-agent-sm, @bmad-agent-dev, @bmad-brainstorming |
| Superpowers | TDD enforcement, brainstorming discipline, sub-agents |
| Sudocode | Git-native persistent memory: specs, issues, statuses (.sudocode/) |
| Context7 | Up-to-date, version-specific library documentation injected into prompts |
| security-guidance | Security prevention hook during code writing (automatic) |
| agency-agents | 144 specialized agents for review, QA, architecture, and more |
Conditional LSPs (detected per session)
| LSP | Activated when |
|---|
| rust-analyzer | Cargo.toml detected |
| TypeScript LSP | package.json or tsconfig.json detected |
Installation
One command installs Ironforge and all dependencies:
bash <(curl -sSL https://raw.githubusercontent.com/RomainDECOSTER/ironforge/main/install.sh)
This automatically:
- Adds all required marketplaces
- Installs Ironforge, BMAD, Superpowers, Sudocode, Context7, and security-guidance
- Installs agency-agents into
~/.claude/agents/
- Initializes Sudocode in the current project (if in a git repo)
After installation, verify everything is configured:
/ironforge:setup
Workflow Modes
Ironforge supports three modes. Start with /ironforge:start — it reads the project context, proposes the right mode, and tells you what to invoke next.
| Mode | When | Entry point |
|---|
| FAST | Bug fix, small change (< 1h) | Superpowers /quick-spec |
| FEATURE | New feature, new component (1h–1 day) | /ironforge:start |
| FULL | New module, major refactor (1 day+) | /ironforge:start |
Skills
Skills are composable — invoke any skill independently when artefacts already exist.
| Command | Description |
|---|
/ironforge:start | Reads project context, proposes mode and entry point, waits for confirmation |
/ironforge:bmad-analyze | Runs BMAD phases (brief → PRD → architecture) with human gates at each step |
/ironforge:bmad-to-sudocode | Converts existing BMAD docs into Sudocode specs and issues |
/ironforge:implement [issue-id] | TDD implementation of Sudocode issues (RED → GREEN → REFACTOR → COMMIT) |
/ironforge:review | Code review (@engineering-code-reviewer) + QA sign-off (@testing-reality-checker) |
/ironforge:full-workflow <task> | Orchestrator: chains all skills for a project with no existing artefacts |
/ironforge:setup | Verify and install all dependencies |
Typical Flows
Starting fresh:
/ironforge:start Build a JWT authentication system
Already have BMAD docs, need Sudocode issues:
/ironforge:bmad-to-sudocode
Resuming implementation on a specific issue:
/ironforge:implement i-0042
Ready to merge:
/ironforge:review
Using with GitHub Copilot CLI
Ironforge ships a parallel copilot/ subtree for GitHub Copilot CLI. The same nine skills are
available with prose-based agent delegation instead of @agent-name shorthand.
Install:
bash <(curl -sSL https://raw.githubusercontent.com/RomainDECOSTER/ironforge/main/install-copilot.sh)
This installs Ironforge (copilot/ subtree), BMAD Method, and agency-agents automatically.
The following dependencies require a separate manual install via copilot plugin install or
their respective CLI install commands:
- Superpowers
- Sudocode
- Context7
- security-guidance
After installation, verify everything is configured:
/ironforge:setup
Limitations:
- Agent delegation uses prose instructions — no
@agent shorthand
- MCP tool prefix for Graphify (
mcp__graphify__<tool_name>) may need verification against
your Copilot MCP configuration
graphify install writes its hook to the Claude Code settings.json path; verify it is
registered in your Copilot settings after running /ironforge:graph-init
BMAD Artefact Paths
Standardized paths expected by all Ironforge skills:
docs/
briefs/ → docs/briefs/brief.md
prd/ → docs/prd/prd.md
arch/ → docs/arch/architecture.md
.sudocode/ → specs and issues (FEATURE and FULL modes)
Context7
Context7 activates automatically whenever a library or framework is detected during design or implementation. It injects current, version-specific documentation into the prompt.
Manual trigger:
use context7
Security
security-guidance runs as a PreToolUse hook before every Edit or Write operation. It warns about command injection, XSS, SQL injection, and other OWASP vulnerabilities. Fully automatic.
