claude-power-setup 1.0.1

package/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Shyam Sridhar
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

package/README.md

@@ -0,0 +1,275 @@
+<p align="center">
+  <img src="assets/banner.svg" alt="Claude Power Setup banner" width="100%">
+</p>
+
+<p align="center">
+  <a href="LICENSE"><img src="https://img.shields.io/badge/License-MIT-blue.svg" alt="MIT License"></a>
+  <a href="https://nodejs.org"><img src="https://img.shields.io/badge/Node.js-18%2B-green.svg" alt="Node.js 18+"></a>
+  <a href="https://docs.anthropic.com/en/docs/claude-code"><img src="https://img.shields.io/badge/Claude%20Code-Extension-purple.svg" alt="Claude Code"></a>
+  <a href="https://www.npmjs.com/package/claude-power-setup"><img src="https://img.shields.io/npm/v/claude-power-setup.svg" alt="npm version"></a>
+</p>
+
+<p align="center">
+  <strong>Multi-agent orchestration, automation pipelines, and recursive self-improvement for <a href="https://docs.anthropic.com/en/docs/claude-code">Claude Code</a>.</strong>
+</p>
+
+---
+
+Layers on top of [ECC (Everything Claude Code)](https://github.com/affaan-m/everything-claude-code). One command to install:  
+
+```bash
+npm install -g claude-power-setup
+claude-power-setup
+```
+
+<video src="https://github.com/shyamsridhar123/claude-power-setup/releases/download/v1.0.0/feature-video.mp4" width="100%" autoplay loop muted playsinline controls></video>
+
+---
+
+## Why This Exists
+
+### The Problem
+
+Claude Code is powerful out of the box, but using it at maximum efficiency requires expertise most developers don't have time to build. You end up with:
+
+- **One mode for everything** — The same approach for research, implementation, review, and debugging. Each of these is a fundamentally different cognitive task, but you're using the same prompt posture for all of them.
+- **Serial execution** — Waiting for one agent to finish before starting the next, even when the work is independent. Your 8-core machine is doing one thing at a time.
+- **No memory between sessions** — Every session starts from zero. The hard-won lesson from Tuesday's debugging marathon? Gone by Wednesday. The pattern you discovered for your project's test setup? You'll rediscover it next month.
+- **Manual quality control** — Remembering to run tests, check security, clean up slop, and review code. Humans forget. Agents don't — if you tell them to.
+
+### The Philosophy
+
+This setup is built on three beliefs about AI-assisted development:
+
+**1. Mode switching is not optional — it's fundamental.**
+
+A developer reviewing code should think differently than a developer writing code. When you're in research mode, you should be read-only and methodical. When you're orchestrating a team, you should decompose and delegate, not implement. Claude Code doesn't enforce this separation. We do — with four purpose-built context profiles that prime the agent for the task at hand.
+
+**2. The agent should get smarter over time, not just the developer.**
+
+Most AI coding tools are stateless. They make the same mistakes on day 100 that they made on day 1. The Continuous Learning system changes this: every tool use is observed, patterns are detected, instincts are created with confidence scores, and high-confidence instincts are promoted from project-scoped to global. Your Claude gets better at *your* codebases, with *your* preferences, automatically.
+
+**3. Orchestration beats implementation.**
+
+The highest-leverage thing a developer can do with AI agents is not write code through them — it's *orchestrate multiple agents in parallel*. One agent researches. Another implements. A third writes tests. A fourth reviews. They work in isolated worktrees with independent context windows, communicate through a shared task board, and produce work that no single agent session could match. This is swarm engineering, and it's the default mode for complex tasks.
+
+### What Makes This Different
+
+This isn't a collection of dotfiles or a prompt library. It's an **operating system layer** for Claude Code:
+
+- **Automation pipelines** (`cpipeline`, `crouted`, `claude-loop`) replace manual multi-step workflows with deterministic, quality-gated sequences
+- **Model routing** sends research to Opus and implementation to Sonnet — the right model for the right job
+- **Instinct-based learning** means the system evolves from observations, not from you manually curating prompts
+- **Safe packaging** — the installer is idempotent, the uninstaller is surgical, and nothing touches your existing ECC setup unless you ask
+
+The result: you spend less time managing the AI and more time thinking about the problem.
+
+---
+
+## Prerequisites
+
+| Requirement | Required | Notes |
+|-------------|----------|-------|
+| **Node.js** 18+ | Yes | Hooks and scripts depend on it |
+| **Claude Code CLI** | Yes | `npm install -g @anthropic-ai/claude-code` |
+| **Git** | Recommended | Worktrees, session save, loop runner |
+| **ECC** | Recommended | This setup extends ECC's hooks and skills |
+
+## Quick Install
+
+### npm (recommended — any platform)
+
+```bash
+npm install -g claude-power-setup
+claude-power-setup
+```
+
+### Linux / macOS / Git Bash (Windows)
+
+```bash
+git clone https://github.com/shyamsridhar123/claude-power-setup.git
+cd claude-power-setup
+bash install.sh
+```
+
+### Windows (PowerShell)
+
+```powershell
+git clone https://github.com/shyamsridhar123/claude-power-setup.git
+cd claude-power-setup
+powershell -ExecutionPolicy Bypass -File install.ps1
+```
+
+### Options
+
+```
+--dry-run      Show what would be installed without writing files
+--force        Overwrite existing files instead of skipping
+--skip-shell   Don't modify shell profile (.bashrc/.zshrc)
+--skip-ecc     Don't check for ECC installation
+```
+
+## What's Included
+
+### Context Profiles (Mode Switching)
+
+Switch between purpose-built Claude modes. Each profile primes the agent with different priorities, allowed tools, and quality gates:
+
+```bash
+source ~/.claude/bin/claude-aliases.sh
+
+cdev          # Development mode — TDD, quality gates, strategic compaction
+corchestrate  # Team lead mode — decompose, delegate, synthesize
+creview       # Code review mode — security-first, dual-model review
+cresearch     # Research mode — read-only investigation, document findings
+```
+
+### Automation Pipelines
+
+Replace multi-step manual workflows with deterministic, scriptable pipelines:
+
+```bash
+# Sequential: implement -> de-slop -> verify -> commit
+cpipeline "Implement OAuth2 login in src/auth/"
+
+# Model-routed: Opus research -> Sonnet implement -> Opus review
+crouted "Add caching layer to API endpoints"
+
+# Continuous loop with safety controls
+~/.claude/bin/claude-loop.sh "Add tests for untested functions" --max-runs 8
+
+# Quick commands
+cfix      # Build + lint + test + fix failures
+cclean    # Remove slop from uncommitted changes
+ccommit   # Auto-commit with conventional message
+caudit    # Security scan + harness audit
+```
+
+### Agent Teams (Swarm Mode)
+
+Enabled via `CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1`. Just ask naturally:
+
+```
+> Build a REST API with auth, rate limiting, and tests.
+> Please use a team of specialists for this.
+```
+
+Claude spawns teammates in isolated worktrees — each with fresh context, self-organizing via a shared task board. Up to 5 specialists working in parallel: architect, backend, frontend, tester, reviewer, docs.
+
+### Learned Instincts
+
+Four battle-tested patterns extracted from real development sessions. These load automatically and improve agent behavior without prompt engineering:
+
+| Instinct | Confidence | What It Does |
+|----------|------------|--------------|
+| Parallel agents for independent files | 0.8 | Spawn parallel agents when 3+ files don't depend on each other |
+| De-sloppify as separate pass | 0.85 | Always clean up with a different agent than the one that wrote the code |
+| Dual-review catches more | 0.9 | Two independent reviewers find ~45% more issues than one |
+| uuid ESM incompatibility | 0.95 | Use `crypto.randomUUID()` instead of uuid package in Jest/CJS projects |
+
+### Continuous Learning Observer
+
+The system watches every tool use, detects patterns in your workflow, and creates new instincts automatically:
+
+```
+Observe (every tool use) -> Detect (patterns emerge) -> Learn (create instincts)
+    -> Promote (project -> global) -> Evolve (instincts -> skills/agents)
+```
+
+Instincts have confidence scores (0.3 tentative to 0.95 certain) that evolve as evidence accumulates. High-confidence instincts are promoted from project scope to global scope automatically.
+
+## Architecture
+
+```
+YOU (developer)
+  |
+  +-- Mode Selection: cdev | corchestrate | creview | cresearch
+  |
+  +-- Orchestration: Agent Teams | Swarm Mode | Santa Loop | Multi-Model
+  |
+  +-- Loop Engine: Sequential Pipeline | Continuous Loop | Model-Routed
+  |
+  +-- Quality Layer: TDD Guide | Verify Loop | De-Slop Pass
+  |
+  +-- Memory & Learning: Session Save/Resume | Instincts (v2.1) | Strategic Compact
+  |
+  +-- Security & Hooks: PreToolUse | PostToolUse | Stop/Session hooks
+```
+
+See `~/.claude/contexts/ORCHESTRATION-REFERENCE.md` for the full reference with architecture diagrams, decision matrices, and workflow examples.
+
+## File Layout
+
+```
+claude-power-setup/
+├── install.sh                 # Main installer (bash, cross-platform)
+├── install.ps1                # Windows PowerShell wrapper
+├── uninstall.sh               # Safe removal (only touches what it installed)
+├── cli.js                     # npm entry point (npx claude-power-setup)
+├── package.json               # npm package config
+├── README.md                  # This file
+├── LICENSE                    # MIT
+├── assets/                    # README assets
+│   └── banner.svg             # Terminal-style banner in Anthropic colors
+├── contexts/                  # Mode profiles
+│   ├── dev.md
+│   ├── orchestrate.md
+│   ├── review.md
+│   └── research.md
+├── bin/                       # Automation scripts
+│   ├── claude-aliases.sh      # Mode aliases + pipeline functions
+│   ├── claude-loop.sh         # Continuous loop with quality gates
+│   └── claude-session-save.sh # Cross-session memory persistence
+├── instincts/                 # Learned patterns with confidence scores
+│   ├── parallel-agents-for-independent-files.md
+│   ├── de-sloppify-separate-pass.md
+│   ├── dual-review-catches-more.md
+│   └── uuid-esm-incompatibility-jest.md
+├── config/                    # Settings templates
+│   ├── env-settings.json      # Agent Teams, cost tracking, hook profile
+│   └── observer-config.json   # Continuous learning observer config
+├── reference/                 # Documentation
+│   └── ORCHESTRATION-REFERENCE.md
+└── video/                     # Remotion source for feature video
+    ├── index.ts
+    ├── Root.tsx
+    └── scenes/
+        └── FeatureVideo.tsx
+```
+
+## Uninstall
+
+```bash
+bash uninstall.sh           # Interactive removal
+bash uninstall.sh --dry-run # Preview what would be removed
+```
+
+The uninstaller only removes files it installed. It never touches ECC hooks, plugins, session data, user-created instincts, or user-modified settings. Env settings are only reverted if their values still match what was installed (user modifications are preserved).
+
+## Key Slash Commands
+
+After installation, these slash commands become available in Claude Code:
+
+| Command | Purpose |
+|---------|---------|
+| `/plan` | Plan before coding (wait for confirmation) |
+| `/tdd` | Test-driven development workflow |
+| `/verify` | Full verification loop (build + lint + test + security) |
+| `/code-review` | Comprehensive quality review |
+| `/santa-loop` | Dual-model adversarial review (both must approve) |
+| `/team-builder` | Compose and dispatch agent teams |
+| `/learn` | Extract reusable patterns from current session |
+| `/instinct-status` | View all learned instincts with confidence |
+| `/save-session` | Persist session state for later resumption |
+| `/resume-session` | Load previous session and continue |
+| `/promote` | Move project instincts to global scope |
+| `/evolve` | Cluster instincts into skills/agents |
+| `/security-scan` | Scan configuration for vulnerabilities |
+
+## Contributing
+
+Contributions welcome. If you've discovered patterns that make Claude Code more effective, consider adding them as instincts with evidence and confidence scores.
+
+## License
+
+[MIT](LICENSE)

package/assets/banner.svg

@@ -0,0 +1,96 @@
+<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 1200 460" fill="none" role="img" aria-labelledby="title desc">
+  <title id="title">Claude Power Setup banner</title>
+  <desc id="desc">A clean dark banner with the Claude Power Setup title, installation command, and a diagram representing agents, pipelines, memory, and review workflows.</desc>
+
+  <defs>
+    <linearGradient id="canvas" x1="160" y1="24" x2="1044" y2="436" gradientUnits="userSpaceOnUse">
+      <stop offset="0%" stop-color="#171210"/>
+      <stop offset="55%" stop-color="#100D0C"/>
+      <stop offset="100%" stop-color="#0C0908"/>
+    </linearGradient>
+    <linearGradient id="frame" x1="24" y1="22" x2="1176" y2="438" gradientUnits="userSpaceOnUse">
+      <stop offset="0%" stop-color="#3A2C26"/>
+      <stop offset="100%" stop-color="#211915"/>
+    </linearGradient>
+    <linearGradient id="accent" x1="72" y1="0" x2="248" y2="0" gradientUnits="userSpaceOnUse">
+      <stop offset="0%" stop-color="#F4B08E"/>
+      <stop offset="50%" stop-color="#E58A61"/>
+      <stop offset="100%" stop-color="#C4623C"/>
+    </linearGradient>
+    <radialGradient id="glowLeft" cx="0" cy="0" r="1" gradientUnits="userSpaceOnUse" gradientTransform="translate(248 72) rotate(25) scale(360 250)">
+      <stop offset="0%" stop-color="#D87453" stop-opacity="0.18"/>
+      <stop offset="100%" stop-color="#D87453" stop-opacity="0"/>
+    </radialGradient>
+    <radialGradient id="glowRight" cx="0" cy="0" r="1" gradientUnits="userSpaceOnUse" gradientTransform="translate(1018 240) rotate(-18) scale(280 220)">
+      <stop offset="0%" stop-color="#E89A73" stop-opacity="0.22"/>
+      <stop offset="100%" stop-color="#E89A73" stop-opacity="0"/>
+    </radialGradient>
+    <filter id="panelShadow" x="-10%" y="-20%" width="140%" height="160%">
+      <feDropShadow dx="0" dy="20" stdDeviation="24" flood-color="#000000" flood-opacity="0.28"/>
+    </filter>
+  </defs>
+
+  <rect width="1200" height="460" rx="28" fill="#090807"/>
+  <rect width="1200" height="460" rx="28" fill="url(#canvas)"/>
+  <ellipse cx="248" cy="72" rx="360" ry="250" fill="url(#glowLeft)"/>
+  <ellipse cx="1018" cy="240" rx="280" ry="220" fill="url(#glowRight)"/>
+
+  <rect x="24" y="22" width="1152" height="416" rx="24" fill="#110E0C" stroke="url(#frame)" filter="url(#panelShadow)"/>
+
+  <circle cx="56" cy="52" r="6" fill="#FF5F57"/>
+  <circle cx="76" cy="52" r="6" fill="#FEBC2E"/>
+  <circle cx="96" cy="52" r="6" fill="#28C840"/>
+  <text x="600" y="57" text-anchor="middle" fill="#8F8378" font-family="ui-monospace, SFMono-Regular, SF Mono, Menlo, Consolas, monospace" font-size="13">claude-power-setup — orchestration toolkit</text>
+  <line x1="24" y1="74" x2="1176" y2="74" stroke="#241C18"/>
+
+  <rect x="72" y="100" width="264" height="30" rx="15" fill="#181311" stroke="#342823"/>
+  <circle cx="93" cy="115" r="4" fill="#EB9168"/>
+  <text x="109" y="120" fill="#CBBCAF" font-family="ui-monospace, SFMono-Regular, SF Mono, Menlo, Consolas, monospace" font-size="13" letter-spacing="1.4">CLAUDE CODE POWER LAYER</text>
+
+  <text x="72" y="184" fill="#F8F1EA" font-family="Inter, Segoe UI, Helvetica Neue, Arial, sans-serif" font-size="64" font-weight="700" letter-spacing="-1.8">Claude</text>
+  <text x="72" y="242" fill="#F8F1EA" font-family="Inter, Segoe UI, Helvetica Neue, Arial, sans-serif" font-size="60" font-weight="700" letter-spacing="-1.6">Power Setup</text>
+  <rect x="72" y="262" width="176" height="8" rx="4" fill="url(#accent)"/>
+
+  <text x="72" y="306" fill="#BFAEA1" font-family="Inter, Segoe UI, Helvetica Neue, Arial, sans-serif" font-size="21" font-weight="500">
+    <tspan x="72">Multi-agent orchestration, automation pipelines,</tspan>
+    <tspan x="72" dy="28">and recursive self-improvement for Claude Code.</tspan>
+  </text>
+
+  <rect x="72" y="364" width="352" height="40" rx="14" fill="#15110F" stroke="#3A2D27"/>
+  <text x="92" y="389" fill="#F0B08C" font-family="ui-monospace, SFMono-Regular, SF Mono, Menlo, Consolas, monospace" font-size="18">$</text>
+  <text x="116" y="389" fill="#EEE3DA" font-family="ui-monospace, SFMono-Regular, SF Mono, Menlo, Consolas, monospace" font-size="18">npx claude-power-setup</text>
+
+  <g>
+    <rect x="652" y="110" width="214" height="86" rx="22" fill="#16110F" stroke="#362923"/>
+    <text x="759" y="142" text-anchor="middle" fill="#F8EFE8" font-family="Inter, Segoe UI, Helvetica Neue, Arial, sans-serif" font-size="19" font-weight="700">Agents</text>
+    <text x="759" y="168" text-anchor="middle" fill="#A59286" font-family="Inter, Segoe UI, Helvetica Neue, Arial, sans-serif" font-size="14">Parallel specialists for</text>
+    <text x="759" y="188" text-anchor="middle" fill="#A59286" font-family="Inter, Segoe UI, Helvetica Neue, Arial, sans-serif" font-size="14">research, build, and docs</text>
+  </g>
+
+  <g>
+    <rect x="936" y="110" width="214" height="86" rx="22" fill="#16110F" stroke="#362923"/>
+    <text x="1043" y="142" text-anchor="middle" fill="#F8EFE8" font-family="Inter, Segoe UI, Helvetica Neue, Arial, sans-serif" font-size="19" font-weight="700">Pipelines</text>
+    <text x="1043" y="168" text-anchor="middle" fill="#A59286" font-family="Inter, Segoe UI, Helvetica Neue, Arial, sans-serif" font-size="14">Deterministic flows</text>
+    <text x="1043" y="188" text-anchor="middle" fill="#A59286" font-family="Inter, Segoe UI, Helvetica Neue, Arial, sans-serif" font-size="14">with quality gates</text>
+  </g>
+
+  <g>
+    <rect x="652" y="302" width="214" height="86" rx="22" fill="#16110F" stroke="#362923"/>
+    <text x="759" y="334" text-anchor="middle" fill="#F8EFE8" font-family="Inter, Segoe UI, Helvetica Neue, Arial, sans-serif" font-size="19" font-weight="700">Memory</text>
+    <text x="759" y="360" text-anchor="middle" fill="#A59286" font-family="Inter, Segoe UI, Helvetica Neue, Arial, sans-serif" font-size="14">Promote instincts from</text>
+    <text x="759" y="380" text-anchor="middle" fill="#A59286" font-family="Inter, Segoe UI, Helvetica Neue, Arial, sans-serif" font-size="14">project scope to global</text>
+  </g>
+
+  <g>
+    <rect x="936" y="302" width="214" height="86" rx="22" fill="#16110F" stroke="#362923"/>
+    <text x="1043" y="334" text-anchor="middle" fill="#F8EFE8" font-family="Inter, Segoe UI, Helvetica Neue, Arial, sans-serif" font-size="19" font-weight="700">Review</text>
+    <text x="1043" y="360" text-anchor="middle" fill="#A59286" font-family="Inter, Segoe UI, Helvetica Neue, Arial, sans-serif" font-size="14">Security, verification,</text>
+    <text x="1043" y="380" text-anchor="middle" fill="#A59286" font-family="Inter, Segoe UI, Helvetica Neue, Arial, sans-serif" font-size="14">and de-slop passes</text>
+  </g>
+
+  <g>
+    <rect x="816" y="214" width="170" height="72" rx="22" fill="#201511" stroke="#6B422F"/>
+    <text x="901" y="242" text-anchor="middle" fill="#F5C6AC" font-family="ui-monospace, SFMono-Regular, SF Mono, Menlo, Consolas, monospace" font-size="15" font-weight="600" letter-spacing="1.2">ORCHESTRATE</text>
+    <text x="901" y="264" text-anchor="middle" fill="#C8B4A7" font-family="Inter, Segoe UI, Helvetica Neue, Arial, sans-serif" font-size="13">modes • teams • loops</text>
+  </g>
+</svg>

package/bin/claude-aliases.sh

@@ -0,0 +1,71 @@
+#!/bin/bash
+# ~/.claude/bin/claude-aliases.sh
+# Source this in your .bashrc or .bash_profile:
+#   source ~/.claude/bin/claude-aliases.sh
+
+# ── Mode-Switched Claude Sessions ──────────────────────────
+alias c='claude'
+alias cdev='claude --system-prompt "$(cat ~/.claude/contexts/dev.md)"'
+alias corchestrate='claude --system-prompt "$(cat ~/.claude/contexts/orchestrate.md)"'
+alias creview='claude --system-prompt "$(cat ~/.claude/contexts/review.md)"'
+alias cresearch='claude --system-prompt "$(cat ~/.claude/contexts/research.md)"'
+
+# ── Quick Non-Interactive Pipelines ────────────────────────
+alias cfix='claude -p "Run build + lint + tests. Fix any failures. Do not add features."'
+alias cclean='claude -p "Review all uncommitted changes. Remove console.logs, dead code, commented-out code, test slop. Run tests after cleanup."'
+alias ccommit='claude -p "Create a conventional commit for all staged changes. Summarize what changed in 1-2 sentences."'
+alias caudit='claude -p "Run /security-scan and /harness-audit. Report findings."'
+
+# ── Sequential Pipeline ───────────────────────────────────
+# Usage: cpipeline "Implement OAuth2 login in src/auth/"
+cpipeline() {
+  local spec="$1"
+  echo "=== Step 1: Implement ==="
+  claude -p "Read the codebase. ${spec}. Write tests first (TDD). Do NOT create documentation files."
+  echo "=== Step 2: De-sloppify ==="
+  claude -p "Review all uncommitted changes. Remove tests that verify language behavior, redundant type checks, console.logs, commented code. Keep business logic tests. Run test suite."
+  echo "=== Step 3: Verify ==="
+  claude -p "Run full build, lint, type check, tests. Fix any failures. Do not add features."
+  echo "=== Step 4: Commit ==="
+  claude -p "Create a conventional commit for all changes. Use a clear, descriptive message."
+  echo "=== Pipeline complete ==="
+}
+
+# ── Model-Routed Pipeline ─────────────────────────────────
+# Usage: crouted "Add caching layer to API endpoints"
+crouted() {
+  local spec="$1"
+  echo "=== Research (Opus) ==="
+  claude -p --model opus "Analyze the codebase architecture. Plan how to: ${spec}. Write plan to .claude/plans/current.md"
+  echo "=== Implement (Sonnet) ==="
+  claude -p "Read .claude/plans/current.md. Implement the plan with TDD."
+  echo "=== De-sloppify ==="
+  claude -p "Cleanup pass on all uncommitted changes. Remove slop, run tests."
+  echo "=== Review (Opus) ==="
+  claude -p --model opus "Review all uncommitted changes for security, correctness, performance. Write findings to .claude/plans/review.md"
+  echo "=== Fix + Commit ==="
+  claude -p "Read .claude/plans/review.md. Fix all critical/high issues. Run tests. Commit."
+}
+
+# ── Worktree Helpers ──────────────────────────────────────
+# Usage: cwt feature-auth "Implement JWT authentication"
+cwt() {
+  local name="$1"
+  local task="$2"
+  git worktree add -b "feat/${name}" "../${PWD##*/}-${name}" HEAD
+  echo "Worktree created: ../$(basename $PWD)-${name}"
+  echo "Run: cd ../$(basename $PWD)-${name} && claude"
+  if [ -n "$task" ]; then
+    (cd "../${PWD##*/}-${name}" && claude -p "$task")
+  fi
+}
+
+# Clean up worktree after merge
+cwt-clean() {
+  local name="$1"
+  git worktree remove "../${PWD##*/}-${name}" 2>/dev/null
+  git branch -d "feat/${name}" 2>/dev/null
+  echo "Cleaned: ${name}"
+}
+
+echo "[claude-aliases] Loaded: cdev, corchestrate, creview, cresearch, cpipeline, crouted, cwt"

package/bin/claude-loop.sh

@@ -0,0 +1,84 @@
+#!/bin/bash
+# ~/.claude/bin/claude-loop.sh
+# Continuous development loop with safety controls
+#
+# Usage:
+#   claude-loop "Add tests for untested functions" --max-runs 5
+#   claude-loop "Fix all linter errors" --max-runs 10 --mode fast
+#   claude-loop "Improve coverage to 80%" --max-runs 8 --model sonnet
+
+set -euo pipefail
+
+PROMPT="${1:?Usage: claude-loop \"prompt\" [--max-runs N] [--mode safe|fast] [--model MODEL]}"
+shift
+
+MAX_RUNS=5
+MODE="safe"
+MODEL=""
+NOTES_FILE="SHARED_TASK_NOTES.md"
+
+while [[ $# -gt 0 ]]; do
+  case "$1" in
+    --max-runs) MAX_RUNS="$2"; shift 2 ;;
+    --mode) MODE="$2"; shift 2 ;;
+    --model) MODEL="--model $2"; shift 2 ;;
+    *) echo "Unknown flag: $1"; exit 1 ;;
+  esac
+done
+
+echo "=== Claude Loop ==="
+echo "Prompt: ${PROMPT}"
+echo "Max runs: ${MAX_RUNS}"
+echo "Mode: ${MODE}"
+echo "Model: ${MODEL:-default}"
+echo ""
+
+# Initialize shared notes
+if [ ! -f "$NOTES_FILE" ]; then
+  cat > "$NOTES_FILE" << 'EOF'
+## Progress
+(auto-updated by loop)
+
+## Next Steps
+(auto-updated by loop)
+
+## Issues Found
+(auto-updated by loop)
+EOF
+fi
+
+QUALITY_GATE=""
+if [ "$MODE" = "safe" ]; then
+  QUALITY_GATE='After implementation, run the full build + lint + test suite. Fix any failures before completing.'
+fi
+
+for i in $(seq 1 "$MAX_RUNS"); do
+  echo ""
+  echo "━━━ Iteration ${i}/${MAX_RUNS} ━━━"
+
+  ITERATION_PROMPT="You are on iteration ${i} of ${MAX_RUNS} of a continuous development loop.
+
+Read ${NOTES_FILE} for context from previous iterations.
+
+TASK: ${PROMPT}
+
+${QUALITY_GATE}
+
+After completing your work:
+1. Update ${NOTES_FILE} with what you accomplished and what's left
+2. If the task is fully complete, include the line: LOOP_COMPLETE
+3. Commit your changes with a conventional commit message"
+
+  claude -p $MODEL "$ITERATION_PROMPT"
+
+  # Check for completion signal
+  if grep -q "LOOP_COMPLETE" "$NOTES_FILE" 2>/dev/null; then
+    echo ""
+    echo "=== Loop completed at iteration ${i} (task signaled complete) ==="
+    break
+  fi
+done
+
+echo ""
+echo "=== Loop finished after ${i} iterations ==="
+echo "Notes: $(cat "$NOTES_FILE" | head -20)"

package/bin/claude-session-save.sh

@@ -0,0 +1,61 @@
+#!/bin/bash
+# ~/.claude/bin/claude-session-save.sh
+# Save current session state for cross-session continuity
+# Called automatically by Stop hook, or manually via /save-session
+#
+# Saves to: ~/.claude/session-data/YYYY-MM-DD-HHMMSS-{project}.md
+
+set -euo pipefail
+
+PROJECT_NAME="${1:-$(basename "$(pwd)")}"
+TIMESTAMP="$(date +%Y-%m-%d-%H%M%S)"
+SESSION_DIR="$HOME/.claude/session-data"
+SESSION_FILE="${SESSION_DIR}/${TIMESTAMP}-${PROJECT_NAME}.md"
+
+mkdir -p "$SESSION_DIR"
+
+# Get git context
+BRANCH="$(git rev-parse --abbrev-ref HEAD 2>/dev/null || echo 'no-git')"
+LAST_COMMITS="$(git log --oneline -5 2>/dev/null || echo 'none')"
+UNCOMMITTED="$(git diff --stat HEAD 2>/dev/null || echo 'none')"
+UNTRACKED="$(git ls-files --others --exclude-standard 2>/dev/null | head -10 || echo 'none')"
+
+cat > "$SESSION_FILE" << EOF
+# Session: ${PROJECT_NAME}
+**Date:** ${TIMESTAMP}
+**Branch:** ${BRANCH}
+**Working Directory:** $(pwd)
+
+## Last 5 Commits
+\`\`\`
+${LAST_COMMITS}
+\`\`\`
+
+## Uncommitted Changes
+\`\`\`
+${UNCOMMITTED}
+\`\`\`
+
+## Untracked Files
+\`\`\`
+${UNTRACKED}
+\`\`\`
+
+## What Was Accomplished
+<!-- Fill in or let Claude update -->
+
+## What's Left To Do
+<!-- Fill in or let Claude update -->
+
+## Approaches That Worked
+<!-- Fill in with evidence -->
+
+## Approaches That Failed
+<!-- Fill in so we don't repeat them -->
+
+## Key Files Modified
+<!-- List the important files touched this session -->
+
+EOF
+
+echo "Session saved: ${SESSION_FILE}"