A Claude Code plugin that interactively initializes projects for Claude Code. It evaluates your codebase, generates a CLAUDE.md file, creates focused documentation, and provides customized skills, hooks, and commands tailored to your project's stack.
Add the marketplace, then install the plugin:
/plugin marketplace add Tech9Apps/t9-forge
/plugin install t9-forge@tech9-claude
claude --plugin-dir /path/to/t9-forgeRun the init skill in any project:
/t9-forge:init
You can pass arguments to limit the scope:
/t9-forge:init just CLAUDE.md
/t9-forge:init full setup
/t9-forge:init hooks only
/t9-forge:init skills only
/t9-forge:init commands only
/t9-forge:init update
The init agent runs an interactive flow across five phases:
The agent explores your codebase to understand:
- Languages, frameworks, and tooling (from config files)
- Project structure and entry points
- Build, test, lint, and format commands
- Dev server configuration
- Existing Claude Code or AI tool configuration
- Git conventions
The agent asks targeted questions based on what it discovered:
- Project purpose and domain
- Architecture decisions not obvious from code
- Your preferences and conventions
It never assumes — it asks first.
The agent evaluates whether your project would benefit from superpowers — an external Claude Code plugin that adds workflow-discipline skills (brainstorming, TDD, systematic debugging, plan writing, subagent-driven development, git worktrees, verification gates).
If it fits your project, the agent recommends installation and shows you the exact /plugin command. Superpowers is not bundled into this toolkit — it's maintained upstream and installed as its own plugin.
Generates or updates:
- CLAUDE.md (under 300 lines) — build commands, architecture overview, conventions
- docs/ files (each under 300 lines) — focused documentation on architecture, API conventions, testing, etc.
All documentation uses @-references for cross-linking. Only files with real content are created.
Copies and customizes templates into your .claude/ directory:
Skills:
/commit— Guided conventional commit workflow/test— Run relevant tests based on changes/code-review— Code quality and security review
Hooks:
- Lint on edit — auto-lint after file changes
- Test on edit — auto-run related tests after file changes
- Format on edit — auto-format after file changes
- Validate bash — warn before destructive commands
Commands:
/healthcheck— Project health dashboard/logs— View recent logs/serve— Start dev server(s)
Each template is customized with your project's actual commands and conventions. The agent presents each one for your approval before writing.
- Interactive, not automatic — the agent always asks before writing
- Stack agnostic — works with any language/framework by discovering the stack organically
- Concise documentation — CLAUDE.md and docs stay under 300 lines each
- No stubs — only creates files with real content
- Customized, not generic — every command and convention is specific to your project
See CONTRIBUTING.md.