{"id":"6d44504f-7506-4c77-ac45-c173d377cb96","slug":"claude-reimagined","kind":"mcp","name":"claude-reimagined","summary":"One-command bootstrap for a fully wired Claude Code workstation — installs Claude Code, RTK, context-mode, code-review-graph, caveman, hooks, MCP servers, statusline, and 39+ curated skills. macOS + L","owner":"kunaltulsidasani","repoUrl":"https://github.com/kunaltulsidasani/claude-reimagined","license":"MIT","licenseTier":"permissive","trust":"community","stars":8,"match":null,"fewStars":false,"format":"mcp","createdAt":"2026-04-25T14:45:45.000Z","addedAt":"2026-10-14T05:57:22.803Z","logoUrl":"https://avatars.githubusercontent.com/u/93653643?u=47cd50806e9ec96be25027eb02cbe44cf90923cf&v=4","linkOnly":false,"collection":null,"skillPath":null,"categories":["developer-tools","memory-knowledge","cloud-infra"],"tags":["bash-script","bootstrap","claude","claude-code","cli","context-engineering","developer-tools","hooks"],"runtime":"binary","local":true,"cloudCapable":false,"cloudCapability":"unknown","cloudReadyOnPlatform":false,"needsApiKey":false,"reviewed":false,"syncedAt":"2026-10-26T22:30:29.548Z","sample":false,"url":"https://conduits.dev/tools/claude-reimagined","language":"Shell","transports":["stdio"],"install":{},"requirements":{"env":[],"upstreamMayCharge":false},"tools":null,"readme":"# claude-reimagined\n\n> A one-command bootstrap that turns a fresh machine into a fully wired Claude Code workstation — CLI, plugins, hooks, MCP servers, statusline, and a 40+ skill library, all configured to work together out of the box.\n\nIf you've ever spent an afternoon stitching together [Claude Code](https://github.com/anthropics/claude-code), [RTK](https://github.com/rtk-ai/rtk), [context-mode](https://github.com/mksglu/context-mode), [code-review-graph](https://pypi.org/project/code-review-graph/), [caveman](https://github.com/JuliusBrussee/caveman), custom hooks, and a skill router — this repo is that afternoon, automated.\n\n---\n\n## Table of Contents\n\n- [Install](#install)\n- [Repository Structure](#repository-structure)\n- [What You Get](#what-you-get)\n- [code-review-graph Per-Repo Setup](#code-review-graph-per-repo-setup)\n- [Hooks Deep-Dive](#hooks-deep-dive)\n- [Settings](#settings)\n- [Skills Library](#skills-library)\n- [Where Files Land](#where-files-land)\n- [Rollback](#rollback)\n- [Requirements](#requirements)\n- [Troubleshooting](#troubleshooting)\n- [Contributing](#contributing)\n- [License](#license)\n\n---\n\n## Install\n\nThree steps. macOS or Linux.\n\n```bash\ngit clone https://github.com/kunaltulsidasani/claude-reimagined.git\ncd claude-reimagined\n./bootstrap.sh\n```\n\nThe bootstrapper will:\n\n1. Detect your OS and check for missing dependencies (`curl`, `git`, `python3`, `npm`, `jq`, `pipx`).\n2. Walk through each component, explain what it does, and ask before installing — **default is yes**, just hit Enter to accept. Type `n` to skip.\n3. Print a verification summary at the end.\n\n### Other modes\n\n```bash\n./bootstrap.sh --dry-run                              # preview only — no changes\n./bootstrap.sh --yes                                  # auto-approve every prompt (no confirmations)\n./bootstrap.sh --force                                # reinstall components already present\n./bootstrap.sh --clean --yes                          # wipe ~/.claude first, then fresh install\n./bootstrap.sh --skip rtk,caveman                     # skip specific components\n./bootstrap.sh --skills-only python-pro,golang-pro    # install only chosen skills\n```\n\n---\n\n## Repository Structure\n\n```\nclaude-reimagined/\n├── bootstrap.sh                  # entry point — orchestrates every install script\n├── statusline.sh                 # Claude Code statusline renderer\n├── Makefile                      # `make install`, `make test`, `make verify`\n├── CLAUDE.md                     # project-level Claude Code instructions\n├── LICENSE                       # MIT\n│\n├── configs/\n│   └── settings.json             # reference settings.json (user-managed copy)\n│\n├── hooks/\n│   ├── pre-compact.sh            # PreCompact hook — survives auto-compaction\n│   └── subagent-model-router.sh  # PreToolUse[Agent] hook — routes models per subagent\n│\n├── lib/\n│   └── common.sh                 # shared bash helpers (logging, prompts, results)\n│\n├── scripts/\n│   ├── install_deps.sh             # curl, git, python3, npm, jq, pipx\n│   ├── install_claude_code.sh      # Claude Code CLI\n│   ├── install_rtk.sh              # RTK token-killer proxy\n│   ├── install_context_mode.sh     # context-mode MCP plugin\n│   ├── install_code_review_graph.sh# code-review-graph MCP server\n│   ├── install_caveman.sh          # caveman terse-mode plugin\n│   ├── install_statusline.sh       # statusline registration\n│   ├── install_subagent_router.sh  # subagent model-routing hook\n│   ├── install_pre_compact.sh      # pre-compact hook\n│   ├── install_skills.sh           # 39 skills from registry.yaml via sparse clone\n│   └── verify_installation.sh      # post-install verification\n│\n├── skills/\n│   └── registry.yaml             # curated skill registry (id → repo:path, tier)\n│\n├── tests/\n│   ├── run_tests.sh              # zero-dep bash test runner\n│   ├── lib/helpers.sh            # test fixtures and assertions\n│   ├── unit/                     # unit tests — common.sh, hooks, routers\n│   └── integration/              # E2E tests for every install script\n│\n└── logs/                         # ignored — populated at runtime\n```\n\n---\n\n## What You Get\n\nEach component solves a specific pain point with vanilla Claude Code. You can skip any of them with `--skip <id>`.\n\n\n| Component             | Upstream                                                               | Why you want it                                                                                                                                                                                                                                                  |\n| --------------------- | ---------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |\n| **claude-code**       | [anthropics/claude-code](https://github.com/anthropics/claude-code)    | The CLI everything else plugs into.                                                                                                                                                                                                                              |\n| **rtk**               | [rtk-ai/rtk](https://github.com/rtk-ai/rtk)                            | Shell proxy that rewrites `git`/`ls`/`grep`/etc. through a token-aware filter. **60–90%** less context burned on shell output. A hook rewrites `git status` → `rtk git status` transparently.                                                                    |\n| **code-review-graph** | [PyPI: code-review-graph](https://pypi.org/project/code-review-graph/) | Tree-sitter knowledge graph of your codebase: functions, classes, calls, imports, tests. Updates incrementally on file change. Claude queries the graph instead of re-reading files. *Per-repo setup required — see [below](#code-review-graph-per-repo-setup).* |\n| **context-mode**      | [mksglu/context-mode](https://github.com/mksglu/context-mode)          | MCP plugin. Routes large command output (build logs, test runs, JSON dumps) into a sandbox FTS5 index. A `pytest -v` eats ~200 tokens instead of 50k.                                                                                                            |\n| **caveman**           | [JuliusBrussee/caveman](https://github.com/JuliusBrussee/caveman)      | Plugin that injects a terse-style system prompt. ~75% fewer output tokens with no loss of correctness.                                                                                                                                                           |\n| **statusline**        | (this repo)                                                            | Renders model, repo, branch, context %, cost, session time, 5h/7d burn rate, and active mode tag into your terminal statusline. *([preview](#statusline-preview))*                                                                                               |\n| **subagent-router**   | (this repo)                                                            | `PreToolUse[Agent]` hook that pins Haiku for cheap lookups, Sonnet for general work, Opus only when explicitly chosen. Saves cost on every spawned agent automatically.                                                                                          |\n| **pre-compact**       | (this repo)                                                            | `PreCompact` hook that detects your stack and writes preservation instructions before auto-compaction. Mid-task state, error messages, and blocked work survive every compact.                                                                                   |\n| **skills**            | various — see [registry](skills/registry.yaml)                         | 39 curated domain skills installed into `~/.claude/skills/` via sparse clone.                                                                                                                                                                                    |\n\n\n---\n\n## Statusline Preview\n\n![statusline](./statusLine.png)\n\nShows model, repo, branch, context %, cost, session time, 5h/7d burn rate, and active mode tag.\n\n---\n\n## code-review-graph Per-Repo Setup\n\nThe bootstrapper installs the `code-review-graph` binary globally, but the **MCP integration is per-repo** — you opt in for each project where you want it.\n\n```bash\ncd /path/to/your/repo\ncode-review-graph install\n```\n\nThat command:\n\n1. Adds an MCP server entry to the project's `.mcp.json` (or creates one).\n2. Sets up file-watcher hooks (via Claude Code's `PostToolUse` hook system) so the graph updates incrementally on every `Edit`, `Write`, or `MultiEdit`.\n3. Builds the initial graph by parsing the repo with Tree-sitter (one-time, takes seconds to a minute depending on repo size).\n\n### Verify it worked\n\n```bash\ncode-review-graph status      # shows nodes/edges/files counts\nls .code-review-graph/         # graph DB lives here, gitignored\ncat .mcp.json                  # MCP server entry should be present\n```\n\nIn Claude Code, the graph tools (`detect_changes`, `get_review_context`, `query_graph`, `semantic_search_nodes`, etc.) will appear in the tool list once you restart the session. See `[CLAUDE.md](CLAUDE.md)` in this repo for an example of instructing Claude to prefer graph tools over Grep.\n\n### Update or rebuild\n\n```bash\ncode-review-graph build       # incremental rebuild (called automatically by hooks)\ncode-review-graph build --full # full rebuild from scratch\n```\n\n### Add `.code-review-graph/` to gitignore\n\n```\necho '.code-review-graph/' >> .gitignore\n```\n\nThe graph database is local-machine state — don't commit it.\n\n---\n\n## Hooks Deep-Dive\n\n### `pre-compact` — survive context compaction\n\n- **File:** `hooks/pre-compact.sh` → `~/.claude/hooks/pre-compact.sh`\n- **Fires:** immediately before Claude auto-compacts (triggered at 80% context via `CLAUDE_AUTOCOMPACT_PCT_OVERRIDE`).\n- **Does:** detects your stack, schema files, API directories, and git state, then writes preservation instructions to stdout. Claude's compaction summarizer reads these and decides what to keep.\n\n**Stack detection covers:** JS/TS (Next.js, NestJS, React, Node), Go, Python (FastAPI, Django, Flask), Rust, Flutter, Java/Maven, Kotlin/Gradle, Ruby/Rails, Swift, C#/.NET. Each stack has its own preservation rules — test command, build command, schema file, API dirs.\n\n**The compacted summary is guaranteed to contain:** active task and next step, exact error messages, every file touched, non-obvious decisions and their reasons, pending/blocked work, discovered env constraints.\n\n### `subagent-model-router` — auto-pick the cheapest capable model\n\n- **File:** `hooks/subagent-model-router.sh` → `~/.claude/hooks/subagent-model-router.sh`\n- **Fires:** before every `Agent` tool call.\n- **Does:** inspects `subagent_type` and prompt complexity, rewrites the tool input via `updatedInput` to pin a specific model.\n\n\n| Subagent type                                                    | Routed model          |\n| ---------------------------------------------------------------- | --------------------- |\n| `Explore`, `statusline-setup`, `claude-code-guide`               | Haiku                 |\n| `general-purpose` with simple lookup prompt                      | Haiku                 |\n| `general-purpose` with complex prompt (implement/debug/design/…) | Sonnet                |\n| `Plan`, `superpowers:code-reviewer`                              | Sonnet (floor)        |\n| everything else                                                  | inherits parent model |\n\n\n---\n\n## Settings\n\n`~/.claude/settings.json` is **user-managed**. The bootstrapper verifies it exists at the end. A reference copy lives in `[configs/settings.json](configs/settings.json)`.\n\n\n| Setting                           | Value        | Why                                                                                  |\n| --------------------------------- | ------------ | ------------------------------------------------------------------------------------ |\n| `CLAUDE_AUTOCOMPACT_PCT_OVERRIDE` | `80`         | Trigger auto-compact at 80% instead of 95%, giving the pre-compact hook room to run. |\n| `skillListingMaxDescChars`        | `300`        | Cap skill descriptions in the listing — keeps the skill router prompt lean.          |\n| `skillListingBudgetFraction`      | `0.005`      | Limit skill listing to 0.5% of context.                                              |\n| `spinnerTipsEnabled`              | `false`      | Disable loading tips.                                                                |\n| `effortLevel`                     | `medium`     | Default thinking effort; override per-task with `/effort`.                           |\n| `promptSuggestionEnabled`         | `false`      | Disable inline suggestions.                                                          |\n| `tui`                             | `fullscreen` | Full-screen TUI mode.                                                                |\n| `prefersReducedMotion`            | `true`       | Disable TUI animations.                                                              |\n| `showThinkingSummaries`           | `false`      | Hide extended thinking summaries.                                                    |\n| `autoScrollEnabled`               | `false`      | Output stays readable mid-task.                                                      |\n\n\nHooks registered in `settings.json`:\n\n- `PreCompact` → `pre-compact.sh`\n- `PreToolUse[Agent]` → `subagent-model-router.sh`\n\n---\n\n## Skills Library\n\n39 skills from community repos, sparse-cloned into `~/.claude/skills/<id>/`. Use `--skills-only <ids>` to pick a subset.\n\nEach entry in `[skills/registry.yaml](skills/registry.yaml)` is tagged with a tier:\n\n\n| Tier       | Anatomy                                 | Example                                      |\n| ---------- | --------------------------------------- | -------------------------------------------- |\n| **gold**   | `SKILL.md` + `scripts/` + `reference/`  | `code-reviewer`, `mcp-builder`, `claude-api` |\n| **silver** | `SKILL.md` + helper scripts/refs inline | `tdd`, `debugging`                           |\n| **bronze** | `SKILL.md` only (knowledge-only skill)  | most language and infra skills               |\n\n\n### Gold tier (full anatomy with executable helpers)\n\n\n| Skill           | Source                                                                                  |\n| --------------- | --------------------------------------------------------------------------------------- |\n| `mcp-builder`   | [anthropics/skills](https://github.com/anthropics/skills/tree/main/skills/mcp-builder)  |\n| `claude-api`    | [anthropics/skills](https://github.com/anthropics/skills/tree/main/skills/claude-api)   |\n| `code-reviewer` | [awesome-skills/code-review-skill](https://github.com/awesome-skills/code-review-skill) |\n\n\n### Silver tier (helpers inline)\n\n\n| Skill       | Source                                                                                           |\n| ----------- | ------------------------------------------------------------------------------------------------ |\n| `tdd`       | [obra/superpowers](https://github.com/obra/superpowers/tree/main/skills/test-driven-development) |\n| `debugging` | [obra/superpowers](https://github.com/obra/superpowers/tree/main/skills/systematic-debugging)    |\n\n\n### Bronze tier (knowledge skills)\n\n\n| Category       | Skills                                                                                                                                                                  |\n| -------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------- |\n| Languages      | `python-pro`, `typescript-pro`, `golang-pro`, `rust-pro`, `java-architect`, `kotlin-specialist`, `swift-expert`, `csharp-developer`, `cpp-pro`, `php-pro`, `ruby-rails` |\n| Frontend       | `react`, `vue-expert`, `angular-architect`, `nextjs-developer`                                                                                                          |\n| Backend        | `nestjs-expert`, `django-expert`, `fastapi-expert`, `laravel-specialist`                                                                                                |\n| Databases      | `postgres-pro`, `redis-pro`, `mongodb-pro`, `database-designer`                                                                                                         |\n| Infrastructure | `aws-solution-architect`, `terraform`, `kubernetes-ops`, `docker`, `ci-cd`                                                                                              |\n| Testing        | `unit-tests`                                                                                                                                                            |\n| API            | `api-designer`, `openapi-docs`                                                                                                                                          |\n| Engineering    | `security-review`, `git-workflow`                                                                                                                                       |\n\n\nUpstream sources (full repo list):\n\n- [anthropics/skills](https://github.com/anthropics/skills)\n- [obra/superpowers](https://github.com/obra/superpowers)\n- [awesome-skills/code-review-skill](https://github.com/awesome-skills/code-review-skill)\n- [Jeffallan/claude-skills](https://github.com/Jeffallan/claude-skills)\n- [Mindrally/skills](https://github.com/Mindrally/skills)\n- [alirezarezvani/claude-skills](https://github.com/alirezarezvani/claude-skills)\n- [seokheejang/claude-ops-skills](https://github.com/seokheejang/claude-ops-skills)\n- [SpillwaveSolutions/document-specialist-skill](https://github.com/SpillwaveSolutions/document-specialist-skill)\n- [clear-solutions/unit-tests-skills](https://github.com/clear-solutions/unit-tests-skills)\n\n---\n\n## Where Files Land\n\n\n| Path                                       | Contents                                |\n| ------------------------------------------ | --------------------------------------- |\n| `~/.claude/settings.json`                  | Claude Code settings — **user-managed** |\n| `~/.claude/CLAUDE.md`                      | Global instructions — **user-managed**  |\n| `~/.claude/statusline.sh`                  | Statusline script                       |\n| `~/.claude/hooks/subagent-model-router.sh` | Subagent model routing hook             |\n| `~/.claude/hooks/pre-compact.sh`           | Pre-compact instructions hook           |\n| `~/.claude/skills/<id>/`                   | Skill directories from registry         |\n| `~/.claude.json`                           | MCP server registrations (context-mode) |\n| `~/.claude-bootstrap/logs/results.tsv`     | Per-component install results           |\n| `~/.claude-bootstrap/logs/`                | Full log output per run                 |\n\n\n---\n\n## Rollback\n\nEvery file modified by a bootstrap script is backed up first:\n\n```\n~/.claude/settings.json.bak.20240101_120000\n```\n\nFormat: `<original-path>.bak.YYYYMMDD_HHMMSS`. To restore, copy the `.bak.*` file back over the original.\n\nFor a clean reinstall: `./bootstrap.sh --clean --yes` wipes `~/.claude` first.\n\nInstall results land in `~/.claude-bootstrap/logs/results.tsv` with status `OK`, `FAILED`, `SKIPPED`, or `DECLINED`.\n\n---\n\n## Requirements\n\n- macOS 12+ or Linux (x86_64 or arm64)\n- bash 4.0+\n- Internet access for downloads\n\nThe bootstrapper installs missing deps automatically (Homebrew on macOS; apt-get/dnf/yum on Linux).\n\n- **Hard deps:** `curl`, `git`, `python3`, `npm`\n- **Soft deps:** `jq`, `pipx`, `uv` (for code-review-graph)\n\n---\n\n## Troubleshooting\n\n`**[FAIL] CLAUDE.md` or `[FAIL] settings`** — these files are user-managed. Create them at `~/.claude/CLAUDE.md` and `~/.claude/settings.json`. The bootstrapper only verifies their existence; it does not install them.\n\n`**rtk gain` fails** — you may have the wrong `rtk` binary on your PATH (name collision with `reachingforthejack/rtk` Rust Type Kit). Run `which rtk` to check, then uninstall the wrong one (`cargo uninstall rtk` or `brew uninstall rtk`).\n\n**Component fails mid-bootstrap** — non-critical components don't abort the run. Check `~/.claude-bootstrap/logs/results.tsv` and the per-script logs in the same directory.\n\n**Skill missing `scripts/` or `reference/` after install** — fixed in the latest install_skills.sh. Re-run with `--force --skills-only <skill-id>`.\n\n---\n\n## Contributing\n\nSee [CONTRIBUTING.md](CONTRIBUTING.md) for development setup, testing, and how to add new components or skills.\n\nQuick reference:\n\n```bash\nmake install     # run bootstrap.sh interactively\nmake test        # run the full test suite (~140 tests)\nmake verify      # run verify_installation.sh\n```\n\n---\n\n## License\n\n[MIT](LICENSE)"}