⚒️ Forge

An AI coding toolkit that fits into how you already work.

8 slash commands · 9 skills · 3 agents · works with any LLM

/forge:review

[CRITICAL] security — Hardcoded API key in source
File: src/api/client.ts:14
Problem: API key assigned to const — committed to version control
Fix: Move to process.env.API_KEY, add to .gitignore

[HIGH] correctness — Off-by-one in pagination
File: src/lib/paginate.ts:42
Problem: Loop runs to items.length instead of items.length - 1
Fix: Change loop condition to i < items.length - 1

Decision: BLOCK
CRITICAL: 1  HIGH: 1  MEDIUM: 2  LOW: 0

What is Forge?

Tired of writing the same review prompts over and over? Forge is a Claude Code plugin that gives you proper commands for the things you do every day: code review, security audits, TDD, architecture planning, docs, and more. Each one has a consistent output format and a clear ship/block verdict.

It works with whatever model you're running — Claude, GPT-4o, Gemini, or a local model like Llama or Mistral. You don't have to change anything.

Install

One-click via claude.com: Go to claude.com/plugins, search Forge, hit Install in Claude Code.

Official marketplace (coming soon):

/plugin install forge@claude-plugins-official

Via GitHub:

claude plugin install Prashantxo/Forge

Via Claude Plugin Hub:

npx claudepluginhub install forge

Manual:

git clone https://github.com/Prashantxo/Forge
cd Forge
claude plugin validate

Commands

Command	What it does
`/forge:review`	Reviews your diff or a PR. Returns severity-rated findings and a ship verdict.
`/forge:secure`	OWASP Top 10 audit. Covers secrets, injection, auth, supply chain.
`/forge:arch`	Reviews existing architecture or designs a new one from your description.
`/forge:tdd`	Writes failing tests first, waits for your approval, then implements.
`/forge:docs`	Generates README, API reference, architecture doc, or runbook.
`/forge:refine`	Simplifies changed code without touching behavior.
`/forge:eval`	Defines pass/fail criteria before you touch any LLM feature.
`/forge:standards`	Checks naming, types, error handling, and observability patterns.

Skills

Skills load automatically based on what you're doing. No manual activation.

Skill	Activates when
`forge:review`	Reviewing PRs or changed files
`forge:secure`	Code touches auth, inputs, APIs, payments, or secrets
`forge:arch`	Designing or evaluating system architecture
`forge:tdd`	Writing new features or fixing bugs
`forge:agentic`	Building agents, prompts, MCP servers, evals, or RAG pipelines
`forge:docs`	Generating documentation of any kind
`forge:refine`	Cleaning up recently changed code
`forge:eval`	Shipping or changing any LLM-powered feature
`forge:standards`	Code quality review or setting up a new module

Agents

Agent	Best with	Does
`forge-reviewer`	Any model	Code review across 6 dimensions
`forge-architect`	MAX tier preferred	Architecture diagrams, ADRs, verdicts
`forge-analyst`	Any model	Deep research with adversarial verification

Agents inherit whatever model you're running. No model names hardcoded anywhere.

Works across model tiers

Forge detects what your model can do and adjusts depth accordingly. Override anytime with FORGE_TIER=max|standard|fast.

Tier	When it kicks in	What you get
MAX	Frontier reasoning models	Full depth, all phases, detailed findings
STANDARD	Mid-range models (default)	Full phases, concise output
FAST	Lightweight / local models	3 phases, bullets, CRITICAL + HIGH only

⚒️ Forge

An AI coding toolkit that fits into how you already work.

8 slash commands · 9 skills · 3 agents · works with any LLM

/forge:review

[CRITICAL] security — Hardcoded API key in source
File: src/api/client.ts:14
Problem: API key assigned to const — committed to version control
Fix: Move to process.env.API_KEY, add to .gitignore

[HIGH] correctness — Off-by-one in pagination
File: src/lib/paginate.ts:42
Problem: Loop runs to items.length instead of items.length - 1
Fix: Change loop condition to i < items.length - 1

Decision: BLOCK
CRITICAL: 1  HIGH: 1  MEDIUM: 2  LOW: 0

What is Forge?

It works with whatever model you're running — Claude, GPT-4o, Gemini, or a local model like Llama or Mistral. You don't have to change anything.

Install

One-click via claude.com: Go to claude.com/plugins, search Forge, hit Install in Claude Code.

Official marketplace (coming soon):

/plugin install forge@claude-plugins-official

Via GitHub:

claude plugin install Prashantxo/Forge

Via Claude Plugin Hub:

npx claudepluginhub install forge

Manual:

git clone https://github.com/Prashantxo/Forge
cd Forge
claude plugin validate

Commands

Command	What it does
`/forge:review`	Reviews your diff or a PR. Returns severity-rated findings and a ship verdict.
`/forge:secure`	OWASP Top 10 audit. Covers secrets, injection, auth, supply chain.
`/forge:arch`	Reviews existing architecture or designs a new one from your description.
`/forge:tdd`	Writes failing tests first, waits for your approval, then implements.
`/forge:docs`	Generates README, API reference, architecture doc, or runbook.
`/forge:refine`	Simplifies changed code without touching behavior.
`/forge:eval`	Defines pass/fail criteria before you touch any LLM feature.
`/forge:standards`	Checks naming, types, error handling, and observability patterns.

Skills

Skills load automatically based on what you're doing. No manual activation.

Skill	Activates when
`forge:review`	Reviewing PRs or changed files
`forge:secure`	Code touches auth, inputs, APIs, payments, or secrets
`forge:arch`	Designing or evaluating system architecture
`forge:tdd`	Writing new features or fixing bugs
`forge:agentic`	Building agents, prompts, MCP servers, evals, or RAG pipelines
`forge:docs`	Generating documentation of any kind
`forge:refine`	Cleaning up recently changed code
`forge:eval`	Shipping or changing any LLM-powered feature
`forge:standards`	Code quality review or setting up a new module

Agents

Agent	Best with	Does
`forge-reviewer`	Any model	Code review across 6 dimensions
`forge-architect`	MAX tier preferred	Architecture diagrams, ADRs, verdicts
`forge-analyst`	Any model	Deep research with adversarial verification

Agents inherit whatever model you're running. No model names hardcoded anywhere.

Works across model tiers

Forge detects what your model can do and adjusts depth accordingly. Override anytime with FORGE_TIER=max|standard|fast.

Tier	When it kicks in	What you get
MAX	Frontier reasoning models	Full depth, all phases, detailed findings
STANDARD	Mid-range models (default)	Full phases, concise output
FAST	Lightweight / local models	3 phases, bullets, CRITICAL + HIGH only

Forge

Popularity

What's Inside

Confidence

README

⚒️ Forge

What is Forge?

Install

Commands

Skills

Agents

Works across model tiers

Similar Plugins

ecc

fullstack-dev-skills

prompts.chat

context7-plugin

⚒️ Forge

What is Forge?

Install

Commands

Skills

Agents

Works across model tiers

Popularity

Health & Quality

Similar Plugins

ecc

fullstack-dev-skills

prompts.chat

context7-plugin

claude-md-management

ponytail