Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/ronaldjdev/forge/llms.txt

Use this file to discover all available pages before exploring further.

Forge is not a plugin or extension in the traditional sense — it is a skill: a self-contained intelligence bundle installed directly into an AI agent’s configuration directory. When an agent starts a session, it loads the Forge SKILL.md as part of its system context, giving it architecture instructions, command routing, boot sequences, and runtime enforcement hooks. Each supported agent receives a copy of the full skill (scripts, profiles, reference docs, and templates), customized with agent-specific paths so every script invocation resolves correctly at runtime.

​
Supported Agents

Forge deploys into five AI agents simultaneously, each with a hook model tailored to its extension system:
AgentInstall DirectoryHook TypeHookEffect
OpenCode.opencode/skills/forge/SKILL.md system—Architecture instructions loaded at startup
Claude Code.claude/skills/forge/PostToolUseforgeSentinelReports violations after each Edit / Write / MultiEdit
Cursor.cursor/skills/forge/preToolUseforgeSmithCan DENY writes that introduce CRITICAL or ERROR violations
Codex CLI.agents/skills/forge/PostToolUseforgeSentinelReports violations after Edit / Write / apply_patch
Gemini.gemini/skills/forge/SKILL.md—Architecture instructions loaded at startup
All agents that expose a runtime hook share the same underlying violation-detection logic via forgeSentinel-lib.mjs. The difference is only when enforcement fires: Cursor’s forgeSmith acts before a write, while the forgeSentinel adapters on Claude Code and Codex CLI act after.

​
Installation

Install Forge into one or more agents using npx. The interactive wizard auto-detects which agents are present in your project: 
npx @ronaldjdevfs/forge install
The --global flag installs Forge into ~/.config/opencode/skills/forge/ so it is available in every OpenCode session on your machine, not just the current project.
The install flags and the agents they target:
FlagTarget AgentActive Hook
--opencodeOpenCodeforgeSentinel (via SKILL.md)
--cursorCursorforgeSmith (preToolUse)
--claudeClaude CodeforgeSentinel (PostToolUse)
--codexCodex CLIforgeSentinel (PostToolUse)
--geminiGemini Code Assist—
--allAll of the above—
--globalOpenCode (global)forgeSentinel

​
Path Rendering

Every file in the Forge skill — SKILL.md, all reference docs, all scripts that call other scripts — uses the placeholder {{AGENT_PATH}} wherever the agent-specific root directory is needed. During installation, cli.js calls renderSkillPaths(), which recursively walks the installed skill directory and replaces every occurrence of {{AGENT_PATH}} with the agent’s actual path:
Agent{{AGENT_PATH}} resolves to
OpenCode (local).opencode/skills/forge
OpenCode (global)~/.config/opencode/skills/forge
Claude Code.claude/skills/forge
Cursor.cursor/skills/forge
Codex CLI.agents/skills/forge
Gemini.gemini/skills/forge
This means a single source template produces correct, agent-specific installations. The boot sequence in SKILL.md, for example, becomes: 
node .claude/skills/forge/scripts/context.mjs
node .claude/skills/forge/scripts/graph.mjs --json
# ... etc.
after Claude Code installation, and: 
node .cursor/skills/forge/scripts/context.mjs
node .cursor/skills/forge/scripts/graph.mjs --json
# ... etc.
after Cursor installation — sourced from the same template.

​
Skill Directory Structure

Every agent receives the same directory layout under its own root. Here is the structure as installed for Claude Code (.claude/skills/forge/): 
.claude/skills/forge/
├── SKILL.md                     ← Orchestrator: boot sequence, command routing, execution flow
├── scripts/
│   ├── context.mjs              ← Detects stack, platform, features, shared, infra, graph
│   ├── armorer.mjs              ← Ownership: orphans, duplicates, misplaced components
│   ├── profile.mjs              ← Matches stack against 10 known profiles
│   ├── graph.mjs                ← Architecture graph: 6 node types, R1-R9 rules, risk score
│   ├── chain.mjs                ← Multi-layer dependency graph, topological sort
│   ├── detect.mjs               ← Rule validation R1-R12, inline ignores, --fix
│   ├── inspect.mjs              ← Full audit orchestrator (180 pts, 10 categories)
│   ├── architecture.mjs         ← Generates / updates ARCHITECTURE.md
│   ├── forgeSentinel.mjs        ← PostToolUse hook adapter (Claude / Codex / OpenCode)
│   ├── forgeSentinel-lib.mjs    ← Shared violation-detection logic
│   ├── forgeSmith.mjs           ← preToolUse write gate (Cursor only)
│   ├── forgeSmith-admin.mjs     ← Hook management: on / off / status
│   └── ...                      ← (assay, smelt, relocate, rollback, and more)
├── reference/
│   ├── principles.md
│   ├── patterns.md
│   ├── di-strategies.md
│   ├── hooks.md
│   └── ...                      ← (one .md per command)
├── profiles/
│   ├── express-mongodb.md
│   ├── express-prisma.md
│   ├── express-drizzle.md
│   ├── fastify-mongodb.md
│   ├── fastify-prisma.md
│   ├── nestjs-mongodb.md
│   └── ...                      ← (10 profiles total)
└── templates/
    ├── feature/                  ← 19 TypeScript templates for feature scaffolding
    ├── platform/                 ← 6 templates for platform components
    ├── shared/                   ← 4 templates for shared (errors, contracts, types, utils)
    └── infra/                    ← 4 templates for infra (prisma, mongodb, redis, mail)
The OpenCode install also generates /commands/forge-*.md slash-commands so you can invoke every Forge command directly from the OpenCode command palette (e.g., /forge-cast, /forge-inspect, /forge-quench).

​
Learn More

Hooks

How forgeSentinel and forgeSmith enforce architecture rules in real time

Profiles

10 pre-built tech profiles — Express, Fastify, NestJS × MongoDB, Postgres, Prisma, Drizzle

Quickstart

Install Forge and run your first audit in under five minutes

Build docs developers (and LLMs) love

Get started for freeTalk to us