supermind-claude 2.1.0 → 4.0.1

package/.claude-plugin/plugin.json

@@ -0,0 +1,21 @@
+{
+  "name": "supermind",
+  "description": "Unified skill engine combining execution infrastructure with behavioral discipline for autonomous Claude Code development",
+  "version": "0.0.0",
+  "author": {
+    "name": "Steven Spivak"
+  },
+  "homepage": "https://github.com/steven-3/supermind",
+  "repository": "https://github.com/steven-3/supermind",
+  "license": "MIT",
+  "keywords": [
+    "skills",
+    "hooks",
+    "tdd",
+    "debugging",
+    "code-review",
+    "worktrees",
+    "autonomous",
+    "developer-tools"
+  ]
+}

package/README.md

@@ -1,47 +1,37 @@
 # Supermind
 
-Complete, opinionated Claude Code setup — hooks, skills, status line, MCP servers, and living documentation.
+A unified skill engine for Claude Code — execution infrastructure meets behavioral discipline. Two modes, 15 skills, 8 hooks, zero dependencies.
 
-## Quick Install
+## Install
 
 ```bash
-npm install -g supermind-claude
-supermind-claude
+npx supermind-claude
 ```
 
-Or run without installing: `npx supermind-claude`
+Or install globally: `npm install -g supermind-claude && supermind-claude`
 
-## What Gets Installed
+Then run `/supermind-init` in any project to generate CLAUDE.md, ARCHITECTURE.md, and DESIGN.md.
 
-| Component | Location | Purpose |
-|-----------|----------|---------|
-| Hooks | ~/.claude/hooks/ | Session persistence, bash permissions, status line, cost tracking |
-| Skills | ~/.claude/skills/ | /supermind-init, /supermind-living-docs |
-| Plugins | settings.json | Superpowers, frontend-design, claude-md-management, ui-ux-pro-max |
-| Settings | settings.json | Thinking mode, effort level, hook registration |
-| Templates | ~/.claude/templates/ | CLAUDE.md project template |
+## Two Modes
 
-## Project Setup
+**`/quick`** — Single-executor path for small tasks (bug fixes, renames, config changes). Fast and stateless.
 
-After installing, run `/supermind-init` in any project to:
-1. Create or merge CLAUDE.md with project-specific config
-2. Generate ARCHITECTURE.md (and DESIGN.md for UI projects)
-3. Check setup health and discover relevant tools
+**`/project`** — Full six-phase lifecycle for features and refactors: discuss, research, plan, execute (parallel waves), verify, ship. Coordinates fresh-context subagents with injected methodology skills.
 
-## Living Documentation
+**`/supermind`** auto-detects which mode fits your task and routes accordingly.
 
-- **Auto-read**: Session-start hook reads ARCHITECTURE.md and DESIGN.md at every conversation start
-- **Manual sync**: Run `/supermind-living-docs` to update docs after code changes
+## What's Included
 
-## Status Line
-
-Two-line terminal display showing: user, model, path, git branch, context usage, thinking level, active agents, and session cost.
-
-## MCP Servers
-
-Choose during setup:
-- **Docker** (AIRIS gateway): Single endpoint routing to context7, playwright, serena, tavily, chrome-devtools, shadcn
-- **Direct**: Individual servers via npx/uvx
+| Component | Count | Purpose |
+|-----------|-------|---------|
+| Skills | 15 | Complexity router, TDD, debugging, brainstorming, code review, planning, execution, worktrees, and more |
+| Hooks | 8 | Bash permissions (blocklist), session persistence, cost tracking, status line, context monitoring |
+| Agents | 1 | Code reviewer for structured review in verify phase |
+| Executor engine | 3 modules | Task packets, wave parallelism, agent prompt templates (cli/lib/) |
+| Safety layer | 1 hook | Blocklist-based command classification with gate override logging |
+| Living docs | 2 skills | Project onboarding (/supermind-init) and doc sync (/supermind-living-docs) |
+| Vendor skills | CLI | Install third-party skills from GitHub with hash-locked integrity |
+| Plugin | manifest | Dual distribution: npm package + Claude Code plugin |
 
 ## Commands
 
@@ -51,20 +41,17 @@ Choose during setup:
 | `supermind-claude update` | Refresh hooks, skills, templates |
 | `supermind-claude doctor` | Verify installation health |
 | `supermind-claude uninstall` | Remove all components |
-| `supermind-claude approve "cmd"` | Permanently auto-approve a command |
-
-## Approved Commands
-
-Permanently auto-approve specific commands that the bash-permissions hook would normally flag:
+| `supermind-claude approve "cmd"` | Auto-approve a blocked command |
+| `supermind-claude skill add <url>` | Install vendor skill from GitHub |
+| `supermind-claude skill update [name]` | Update vendor skill(s) |
+| `supermind-claude skill list` | List installed vendor skills |
+| `supermind-claude skill remove <name>` | Remove vendor skill |
 
-```bash
-supermind-claude approve "git push"        # exact/prefix match
-supermind-claude approve "/npm run .*/"    # regex match
-supermind-claude approve --list            # see all approved
-supermind-claude approve --remove "git push"  # remove approval
-```
+## MCP Servers
 
-Or tell Claude: "add that to my approved commands" — it knows how to edit the file directly.
+Optional. Choose during setup:
+- **Docker** (AIRIS gateway): Single endpoint routing to context7, playwright, serena, tavily, chrome-devtools, shadcn
+- **Direct**: Individual servers via npx/uvx
 
 ## Platforms
 
@@ -72,7 +59,8 @@ Windows, macOS, and Linux. Requires Node.js >= 18.
 
 ## Troubleshooting
 
-Run `supermind-claude doctor` to check installation health. Common issues:
-- **Plugins not active**: Restart Claude Code after install
-- **Status line not showing**: Ensure Node.js is in PATH
-- **Hooks not firing**: Run `supermind-claude update` to re-register
+Run `supermind-claude doctor` to check installation health.
+
+## Credits
+
+Methodology skills forked from [obra/superpowers](https://github.com/obra/superpowers) (MIT) by Jesse Vincent and the Prime Radiant team. Execution architecture inspired by [gsd-build/get-shit-done](https://github.com/gsd-build/get-shit-done) (MIT). Both adapted and rebuilt as Supermind-native implementations.

package/agents/code-reviewer.md

@@ -0,0 +1,81 @@
+---
+name: code-reviewer
+description: Subagent that reviews code changes against plan and coding standards
+---
+
+# Code Reviewer Agent
+
+You are a code reviewer for the Supermind project. Your job is to review code changes and produce a structured review. You do NOT modify files — you only analyze and report.
+
+## Inputs
+
+You will receive:
+
+- **Diff**: The git diff of all changes to review
+- **Plan**: The implementation plan or task spec that describes what was supposed to be built
+- **Task Spec**: Specific acceptance criteria and expected output for the task
+
+## Instructions
+
+1. Read the plan/task spec to understand what was intended
+2. Read the full diff carefully — every file, every hunk
+3. For each changed file, evaluate against ALL six criteria:
+   - **Spec compliance** — does the code do what the plan said?
+   - **Correctness** — logic errors, off-by-ones, edge cases, null handling, async correctness
+   - **Test coverage** — are important behaviors tested? Edge cases covered?
+   - **Security** — injection, path traversal, secrets in code, unsafe operations
+   - **Maintainability** — clear naming, reasonable complexity, no unnecessary abstractions
+   - **Consistency** — follows existing codebase patterns and conventions
+4. Classify each finding as Critical, Important, or Suggestion
+5. Produce the structured review output
+
+## Constraint
+
+You are a reviewer only. Do NOT:
+- Modify any files
+- Run any commands that change state
+- Create commits
+- Suggest rewrites of working code for style preference alone
+
+You MAY read files, run read-only commands (grep, git log, etc.), and run tests to verify behavior.
+
+## Input Template
+
+```
+### Diff
+{{diff}}
+
+### Plan
+{{plan}}
+
+### Task Spec
+{{task_spec}}
+```
+
+## Output Format
+
+```
+## Code Review: [task/feature name]
+
+### Summary
+One paragraph: overall assessment, confidence level.
+
+### Critical Issues
+- [file:line] Description. Why it matters. Suggested fix.
+
+### Important Issues
+- [file:line] Description. Why it matters. Suggested fix.
+
+### Suggestions
+- [file:line] Description.
+
+### Verdict
+PASS | NEEDS FIXES | FAIL
+```
+
+**Verdict rules:**
+- **PASS**: Zero critical AND zero important issues.
+- **NEEDS FIXES**: Has critical or important issues, but fixable without redesign.
+- **FAIL**: Fundamental problems requiring architectural change.
+
+If a section has no items, write "None."