# Context Forge > Auto harness engineering meta-repo: discuss β†’ curate β†’ bootstrap a context-engineered project for Claude Code, Cursor, Copilot, and other AI coding agents. - **Type:** MCP server - **Install:** `agentstack add mcp-glowelephant-context-forge` - **Verified:** Yes β€” security-reviewed for prompt injection and unsafe behavior - **Seller:** [glowElephant](https://agentstack.voostack.com/s/glowelephant) - **Installs:** 0 - **Category:** [Integrations](https://agentstack.voostack.com/c/integrations) - **Latest version:** 0.1.0 - **License:** MIT - **Upstream author:** [glowElephant](https://github.com/glowElephant) - **Source:** https://github.com/glowElephant/context-forge - **Website:** https://github.com/glowElephant/context-forge ## Install ```sh agentstack add mcp-glowelephant-context-forge ``` Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent. ## About # context-forge > **Auto harness engineering for AI coding agents.** > Discuss your project, curate battle-tested know-how, and bootstrap a fully context-engineered repository β€” for Claude Code, Cursor, Copilot, and beyond. [πŸ‡°πŸ‡· ν•œκ΅­μ–΄](./README.ko.md) Β· [Catalog](./catalog) Β· [Discussions](https://github.com/glowElephant/context-forge/discussions) Β· [Contributing](./docs/contributing.md) --- ## What is context-forge? **context-forge** is a meta repository that turns scattered AI-agent know-how β€” `CLAUDE.md` patterns, `AGENTS.md` standards, `.cursorrules`, agent skills, MCP servers, spec-driven workflows, prompt libraries, and convention guides β€” into a **single curated catalog** that you can apply to a brand-new project in one go. You describe what you're building (web app, Unity game, MCP server, CLI tool…) through a guided discussion. context-forge picks the matching know-how from the catalog, creates a new GitHub repository, and writes everything in: `CLAUDE.md`, `AGENTS.md`, `README.md`, `docs/spec.md`, `.claude/skills/`, `.claude/agents/`, `.claude/rules/`, boilerplate, and more. Think of it as **`create-react-app` for AI-augmented codebases** β€” except instead of a fixed template, it's a living catalog crowd-curated from the best agentic-engineering repos on GitHub. ## Why? The current state of AI coding agents (Claude Code, Cursor, Copilot, Codex, Aider, Windsurf, …) demands serious context engineering before useful work begins. Every project re-solves the same boilerplate: - Which `CLAUDE.md` patterns apply here? - Which skills / subagents / hooks should be installed? - Which spec-driven workflow fits this team? - Which MCP servers do we need? - Which language conventions, code-review checklists, commit standards? context-forge collapses that hours-long ritual into a guided 5-minute conversation. Then you `cd` into the new repo and start coding. ## How it works ``` [Forked know-how repos] glowElephant/... ↓ [Catalog] ───────► /context-forge slash command ↓ β”‚ [Discussion] β”‚ Brief fixed questions ↓ β”‚ Free-form goal/constraint chat [Match + Confirm] β—„β”€β”€β”€β”€β”€β”€β”˜ ↓ [gh repo create user/] β†’ New harness-engineered repo β”œβ”€β”€ CLAUDE.md / AGENTS.md β”œβ”€β”€ README.md β”œβ”€β”€ docs/spec.md β”œβ”€β”€ docs/know-how/ β”œβ”€β”€ .claude/skills,agents,rules/ └── boilerplate ``` ## Catalog scope (v1) The v1 catalog is intentionally **narrow** β€” focused on the AI-agent harness itself: | Category | Examples | |----------|----------| | Claude Code core | `anthropics/skills`, `obra/superpowers`, `garrytan/gstack` | | Cursor / Copilot rules | `PatrickJS/awesome-cursorrules`, `github/awesome-copilot` | | Spec-driven development | `github/spec-kit`, `Fission-AI/OpenSpec` | | Multi-agent orchestration | `wshobson/agents`, `ruvnet/ruflo` | | MCP infrastructure | `modelcontextprotocol/servers`, `punkpeye/awesome-mcp-servers` | | Standards | `agentsmd/agents.md`, `google-labs-code/design.md` | | Prompt engineering | `dair-ai/Prompt-Engineering-Guide`, `f/prompts.chat` | v2 will broaden to domain conventions (game engines, mobile, web, backend, ML, security, documentation). See [`docs/specs/2026-04-28-fork-candidates.md`](./docs/specs/2026-04-28-fork-candidates.md) for the full curated list. ## Status βœ… **v1.1.0 released** ([release notes](https://github.com/glowElephant/context-forge/releases/tag/v1.1.0)) - [x] Design specification - [x] Candidate repositories curated (67 v1 + 210 v2) - [x] Forking the v1 catalog sources (82 active after first monthly adoption) - [x] `/context-forge` slash command implementation - [x] **Catalog frontmatter for every source (82/82)** β€” manual v1 seed (15) + auto-seeded via `scripts/seed_frontmatter.py` (67) - [x] Weekly fork sync + link-check ([`docs/fork-sync-setup.md`](./docs/fork-sync-setup.md)) - [x] Monthly borderline rescore + quarterly full rescore (5-factor rubric) - [x] Monthly auto-discovery job β†’ Issue triage (`automation.py discover` + `monthly-rediscovery.yml`) - [x] Topic tag `context-forge-source` on all forks ([browse](https://github.com/glowElephant?tab=repositories&q=topic%3Acontext-forge-source)) - [x] End-to-end smoke test β€” Phase 0~6 dogfood-validated 2026-05-27 (2 bugs found & fixed: YAML quote leak in auto-seeded category, `git branch -M main` missing for Windows). Fresh-session test in real Claude Code still recommended for full Skill-tool invocation flow. - [ ] Automated catalog extraction from forked sources (v1.5) - [ ] v2 domain expansion (game/mobile/web/backend/ML/security) Track progress in [`docs/specs/`](./docs/specs) and [`docs/plans/`](./docs/plans). ## Quick start ### Prerequisites | # | Requirement | Check | |---|---|---| | 1 | `gh` CLI authenticated | `gh auth status` | | 2 | context-forge cloned locally | `git clone https://github.com/glowElephant/context-forge ~/code/context-forge` | | 3 | `CONTEXT_FORGE_PATH` env var pointing at the clone (**optional** β€” skill asks once per session if missing) | see below | The slash command itself runs `validate-catalog.sh` at startup and refuses to proceed if anything is wrong, so the catalog state is verified for you. ### Set `CONTEXT_FORGE_PATH` **macOS / Linux:** ```bash echo 'export CONTEXT_FORGE_PATH=~/code/context-forge' >> ~/.bashrc # or ~/.zshrc source ~/.bashrc ``` **Windows (PowerShell, persist across all shells):** ```powershell [Environment]::SetEnvironmentVariable('CONTEXT_FORGE_PATH', 'C:\Git\context-forge', 'User') # μƒˆ 터미널/μ„Έμ…˜λΆ€ν„° 적용 ``` **Claude Code only (works across Mac/Linux/Windows, syncs with `.claude/settings.json`):** ```json { "env": { "CONTEXT_FORGE_PATH": "C:\\Git\\context-forge" } } ``` ### Run In any directory: ``` /context-forge ``` Answer a few questions about your project. context-forge will: 1. Match catalog know-how (82+ sources) to your stack and goals 2. Create a new GitHub repository 3. Populate it with curated `CLAUDE.md`, skills, rules, docs 4. Hand off β€” `cd && claude` to start coding > v1.1.0 β€” weekly fork sync + monthly auto-discovery active. End-to-end smoke test still pending; report issues via [Discussions](https://github.com/glowElephant/context-forge/discussions). ## Contributing context-forge is a **community catalog**. Add your favorite repository or your own: - πŸ’¬ **GitHub Discussions** β€” propose a source repo to fork ([open a discussion](https://github.com/glowElephant/context-forge/discussions)) - πŸ›  **Pull request** β€” directly add an entry to `sources/index.json` - πŸ“‹ **Scoring rubric** β€” see [`docs/scoring.md`](./docs/scoring.md) All sources are evaluated on **popularity, activity, real-world reviews, content quality, and maintainer reliability** before promotion to Tier 1 (auto-fork). ## Why "harness engineering"? The agentic coding community has converged on the term **harness** for the layer that wraps an AI model β€” system prompt, tools, memory, hooks, context. **Harness engineering** is the discipline of designing that layer well. context-forge is the meta tool that does this for you, automatically, before you write a single line of code. ## Related work context-forge stands on the shoulders of these projects: - [obra/superpowers](https://github.com/obra/superpowers) β€” agentic skills framework - [garrytan/gstack](https://github.com/garrytan/gstack) β€” Claude Code 23-tool stack - [github/spec-kit](https://github.com/github/spec-kit) β€” spec-driven development - [agentsmd/agents.md](https://github.com/agentsmd/agents.md) β€” `AGENTS.md` open standard - [coleam00/context-engineering-intro](https://github.com/coleam00/context-engineering-intro) β€” context engineering primer ## License MIT Β© glowElephant --- **Keywords**: claude code Β· claude skills Β· agent skills Β· cursor rules Β· copilot instructions Β· agents.md Β· claude.md Β· context engineering Β· spec-driven development Β· mcp servers Β· harness engineering Β· agentic coding Β· ai coding assistant Β· prompt engineering Β· meta repository Β· auto bootstrap ## Source & license This open-source MCP server is cataloged on AgentStack and links to its original source β€” we do not rehost the code. - **Author:** [glowElephant](https://github.com/glowElephant) - **Source:** [glowElephant/context-forge](https://github.com/glowElephant/context-forge) - **License:** MIT - **Homepage:** https://github.com/glowElephant/context-forge Install and usage instructions live in the source repository linked above. ## Pricing - **Free** β€” Free ## Security capabilities Automated source analysis of v0.1.0 β€” what this tool can access: - **Network access:** no - **Filesystem access:** no - **Shell / process execution:** no - **Environment & secrets:** no - **Dynamic code execution:** no *"Yes" means the capability is present in the source β€” more access means more to trust, not that it is unsafe.* ## Versions - **0.1.0** β€” security scan: passed β€” Imported from the upstream source. ## Links - Listing page: https://agentstack.voostack.com/l/mcp-glowelephant-context-forge - Seller: https://agentstack.voostack.com/s/glowelephant - Browse the marketplace: https://agentstack.voostack.com/browse --- Listed on AgentStack β€” the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.