forge
By jdforsythe
Assemble science-backed AI agent teams by decomposing complex missions into blueprints with topologies and roles, creating domain-specialized agents and skills, curating libraries for duplicates and quality, researching evidence-based briefs, and enforcing quality gates via structured reviews and verifications.
npx claudepluginhub jdforsythe/forge --plugin forgeComponent Overview
Component Details
Agents (3)
Research analyst for gathering, evaluating, and synthesizing information from multiple sources into evidence-based research briefs. Invoke when a task requires systematic information gathering, source credibility assessment, triangulation of findings, or when another agent needs documented research to support a decision.
Quality reviewer for evaluating work products against defined standards. Invoke at quality gates in a pipeline, when an artifact needs structured feedback with severity classification (blocking/major/minor/nitpick), or when a ship/no-ship decision is needed.
Quality gate for checking outputs against explicit acceptance criteria. Invoke after an artifact is produced and before it passes downstream in a pipeline. Produces APPROVED/REJECTED verdicts with per-criterion evidence. Use for deterministic checks (schema conformance, format compliance) and criteria-based evaluation.
Skills (4)
Creates structured agent definitions using the 7-component format grounded in persona science (PRISM), vocabulary routing, and failure mode taxonomy (MAST). Produces agents with real-world job titles, expert domain vocabulary payloads (15-30 terms), explicit deliverables, decision boundaries, imperative SOPs, and named anti-pattern watchlists. Use this skill when the user wants to create an agent, define a role, build a persona, or needs a specialized AI assistant for a specific domain. Also triggers when Mission Planner delegates agent creation for team roles. Works for any domain — software, marketing, security, operations, design, writing, research, and more. Do NOT use for creating skills (use Skill Creator) or team composition (use Mission Planner).
Reviews, curates, and maintains the Forge library of agents, skills, and templates. Performs deduplication analysis, staleness detection, quality promotion, and orphan reference checking. Produces structured review reports with actionable recommendations for merging, archiving, or promoting library items. Use this skill when the user wants to review the library, clean up agents or skills, check what's available, find duplicates, trim unused items, see library statistics, or says "what's in my library?" Also triggers on scheduled review intervals or when the library grows beyond 20 items. Do NOT use for creating new agents (use Agent Creator), creating skills (use Skill Creator), or planning teams (use Mission Planner).
Decomposes goals into team blueprints using evidence-based scaling laws, topology selection, and role design. Determines whether a goal needs a single agent or a coordinated team (3-5 agents max), selects the optimal communication topology (sequential pipeline, parallel-independent, centralized coordinator, hierarchical), and produces structured blueprints with artifact chains and quality gates. Use this skill when the user wants to build something, plan a project, assemble a team, figure out what roles they need, break down a complex goal, or asks "how should I approach [X]?" — even if they don't mention agents or teams. Also triggers on specific domains: "build a SaaS," "marketing campaign," "security audit," "write a book," "launch a product." Do NOT use for simple single-step tasks, direct coding questions, or file operations that need no planning.
Creates high-quality Claude Code and Cowork skills using evidence-based principles: expert vocabulary payloads for knowledge routing, dual-register descriptions for reliable triggering, named anti-pattern watchlists for steering past the distribution center, and progressive disclosure architecture for context efficiency. Produces SKILL.md files with structured behavioral instructions, canonical examples, and bundled references. Use this skill when the user wants to create a skill, build a custom capability, make a reusable prompt template, or says "I want Claude to always do X." Also triggers when Mission Planner or Agent Creator need to create a domain skill JIT. Works for any domain. Do NOT use for creating agent definitions (use Agent Creator) or team composition (use Mission Planner).
README
Forge
Science-backed AI team assembly. From goal to agents to artifacts.
Forge is an open-source system that uses research from DeepMind, PRISM persona science, and context engineering to assemble AI agent teams. It takes a goal, determines whether you need one agent or a team, selects the right coordination topology, and produces structured agent definitions with expert vocabulary, clear deliverables, and anti-pattern guardrails. Works with Claude Code.
The Core Insight
The single highest-leverage intervention in AI agent quality is vocabulary routing — using precise domain terminology that activates expert knowledge clusters in the model's embedding space. Real-world job titles and role structures activate relevant training data better than custom personas. And DeepMind's scaling research shows that more agents isn't always better — teams of 3-5 with structured artifact handoffs outperform larger groups.
Quick Start: Claude Code
# Install as a plugin (recommended):
/plugin add https://github.com/jdforsythe/forge
# Or via Vercel's cross-agent installer:
npx add-skill jdforsythe/forge
# Then just describe what you want:
"Build me a SaaS analytics product" # Mission Planner activates
"Create an agent for code review" # Agent Creator activates
What's Included
forge/
├── .claude-plugin/ Plugin metadata
│ ├── marketplace.json Marketplace definition
│ └── plugin.json Plugin manifest
│
├── skills/ 4 core skills
│ ├── mission-planner/ Decomposes goals into team blueprints
│ ├── agent-creator/ Builds individual agent definitions
│ ├── skill-creator/ Creates reusable skill packages
│ └── librarian/ Manages the agent/template library
│
├── agents/ 3 infrastructure agents
│ ├── verifier.md Validates outputs against schemas
│ ├── researcher.md Gathers context and source material
│ └── reviewer.md Reviews and critiques agent definitions
│
├── library/ Starter collection
│ └── index.json 11 domain agents, 3 team templates
│ ├── software/ Product Manager, Architect, Lead Engineer, QA
│ ├── marketing/ Campaign Strategist, Content Creator, Designer, Analytics Lead
│ └── security/ Lead Auditor, Penetration Tester, Compliance Analyst
│
├── schemas/ Format specifications
│ ├── agent-definition.md 7-component agent structure
│ ├── team-blueprint.md Blueprint format for coordinated teams
│ ├── index-schema.json Library index format
│ └── usage-log-schema.json Usage tracking format
│
├── docs/ User documentation
│ └── research/ Scientific foundation (8 reference documents)
└── ...
How It Works
Forge uses a 3-level decision flow:
Level 0 — Single Agent. The goal is simple enough for one agent. Forge produces a single well-prompted agent definition with the right vocabulary, deliverables, and guardrails. No coordination overhead.
Level 1 — Known Pattern. The goal matches a template in the library (e.g., SaaS product, marketing campaign, security audit). Forge loads the template, adapts roles to your specific goal, and creates the full agent team with artifact handoff chains.
Level 2 — Novel Domain. No template exists. Forge decomposes the goal into workstreams, proposes a team topology (pipeline, parallel, coordinator, or hierarchical), defines roles with precise vocabulary, and iterates with you until the blueprint is right.
At every level, the same principles apply: real-world role titles, domain-specific vocabulary, structured artifacts between agents, and a hard cap of 3-5 agents per team.
Research Foundation
Every design decision in Forge traces back to published research. The docs/research/ directory contains synthesized findings from DeepMind's multi-agent scaling laws, PRISM persona science, and context engineering best practices.
For the full methodology, see METHODOLOGY.md.
Known Issues
- Cowork compatibility: Forge skills fail to run in Cowork. Skills load but stop executing when they attempt to read reference files outside the plugin directory. We are looking for help solving this — PRs welcome! See CONTRIBUTING.md if you want to take a crack at it.
Contributing
See CONTRIBUTING.md for guidelines on adding agents, templates, skills, and research.
License
Similar Plugins
harness
The team-architecture factory for Claude Code — a meta-skill that turns a domain description into an agent team and the skills they use, with six pre-defined team-architecture patterns (Pipeline, Fan-out/Fan-in, Expert Pool, Producer-Reviewer, Supervisor, Hierarchical Delegation). Claude Code용 팀 아키텍처 팩토리: 도메인 한 문장을 에이전트 팀과 스킬 세트로 변환하는 메타 스킬.
Uses power tools
Uses Bash, Write, or Edit tools
Help us improve
Share bugs, ideas, or general feedback.