Claude/init project 7l gu i

.claude-plugin/INSTALL.md

@@ -0,0 +1,106 @@
+# claude-kit Installation
+
+## Full Mode (recommended)
+
+Full mode gives access to all features including `/forge` commands, agents, and practices pipeline.
+
+### Linux
+
+```bash
+git clone https://github.com/luiseiman/claude-kit.git ~/claude-kit
+cd ~/claude-kit && ./global/sync.sh
+```
+
+### macOS
+
+```bash
+git clone https://github.com/luiseiman/claude-kit.git ~/claude-kit
+cd ~/claude-kit && ./global/sync.sh
+```
+
+> Requires bash 3.2+ (pre-installed on macOS). If using Homebrew bash, both work.
+
+### Windows (WSL — recommended)
+
+```bash
+# From WSL terminal:
+git clone https://github.com/luiseiman/claude-kit.git ~/claude-kit
+cd ~/claude-kit && ./global/sync.sh
+```
+
+### Windows (Git Bash)
+
+```bash
+# From Git Bash:
+git clone https://github.com/luiseiman/claude-kit.git ~/claude-kit
+cd ~/claude-kit && ./global/sync.sh
+```
+
+> Symlinks require Developer Mode enabled in Windows Settings > Privacy & Security > For Developers. If symlinks are not available, `sync.sh` automatically falls back to file copies.
+
+### After installation
+
+In any project directory:
+```
+/forge bootstrap    # Initialize .claude/ with full config
+/forge audit        # Audit configuration and get a score (0-10)
+/forge sync         # Update config against current template
+```
+
+### Explicit target directory
+
+If `sync.sh` detects the wrong home directory, override with:
+```bash
+CLAUDE_HOME=/path/to/.claude ./global/sync.sh
+```
+
+## Plugin Mode (lightweight)
+
+Plugin mode installs a curated subset: hooks, rules, and commands.
+Skills, agents, and practices pipeline require the full repository.
+
+### What's included
+
+| Component | Included | Notes |
+|-----------|----------|-------|
+| block-destructive hook | yes | PreToolUse, blocks dangerous commands |
+| lint-on-save hook | yes | PostToolUse, auto-lint on file write |
+| warn-missing-test hook | yes | PostToolUse, strict profile only |
+| Common rules | yes | _common.md, memory.md, agents.md |
+| Commands | yes | audit, health, review, debug |
+| Skills (/forge) | no | Requires full repo |
+| Agents (6 subagents) | no | Requires full repo |
+| Practices pipeline | no | Requires full repo |
+| Stack rules | selectable | Pick stacks during install |
+
+### Installation
+
+```bash
+# When Claude Code plugin system is available:
+claude plugin install claude-kit
+
+# Until then, use git clone + sync.sh (see Full Mode above)
+```
+
+## Stack Plugins
+
+Individual stacks are also available as standalone plugins:
+
+```bash
+# When plugin system supports composition:
+claude plugin install claude-kit-stack-python-fastapi
+claude plugin install claude-kit-stack-react-vite-ts
+```
+
+Each stack plugin includes its `rules/*.md` and `settings.json.partial`.
+Multiple stack plugins compose via permission union merge.
+
+## Troubleshooting
+
+| Problem | Solution |
+|---------|----------|
+| `/forge` not found after install | Run `sync.sh` again — check that target path matches your `~/.claude/` |
+| Symlinks installed in wrong home | Set `CLAUDE_HOME` explicitly: `CLAUDE_HOME=~/.claude ./global/sync.sh` |
+| Windows: "symlink not supported" | Enable Developer Mode, or let sync.sh use file copies (automatic) |
+| macOS: "permission denied" on sync.sh | Run `chmod +x global/sync.sh` first |
+| WSL: changes not visible in Windows | Claude Code should be run from WSL, not Windows cmd |

.claude-plugin/plugin.json

@@ -0,0 +1,71 @@
+{
+  "name": "claude-kit",
+  "version": "1.5.0",
+  "description": "Configuration factory for Claude Code — templates, stacks, hooks, rules, and audit tools",
+  "author": "luiseiman",
+  "license": "MIT",
+  "repository": "https://github.com/luiseiman/claude-kit",
+  "keywords": ["claude-code", "configuration", "hooks", "rules", "audit"],
+  "components": {
+    "hooks": [
+      {
+        "file": "template/hooks/block-destructive.sh",
+        "event": "PreToolUse",
+        "matcher": "Bash",
+        "description": "Block dangerous bash commands (rm -rf, DROP TABLE, force push)"
+      },
+      {
+        "file": "template/hooks/lint-on-save.sh",
+        "event": "PostToolUse",
+        "matcher": "Write",
+        "description": "Auto-lint on file save (Python, TypeScript, Swift)"
+      },
+      {
+        "file": "template/hooks/warn-missing-test.sh",
+        "event": "PostToolUse",
+        "matcher": "Write",
+        "description": "Warn when creating source files without tests (strict profile)"
+      },
+      {
+        "file": "template/hooks/session-report.sh",
+        "event": "Stop",
+        "matcher": null,
+        "description": "Generate session report on session end (requires FORGE_SESSION_REPORT=true)"
+      }
+    ],
+    "rules": [
+      "template/rules/_common.md",
+      "template/rules/memory.md",
+      "template/rules/agents.md"
+    ],
+    "commands": [
+      "template/commands/audit.md",
+      "template/commands/health.md",
+      "template/commands/review.md",
+      "template/commands/debug.md"
+    ]
+  },
+  "stacks": [
+    "python-fastapi",
+    "react-vite-ts",
+    "swift-swiftui",
+    "supabase",
+    "docker-deploy",
+    "data-analysis",
+    "gcp-cloud-run",
+    "redis",
+    "node-express",
+    "java-spring",
+    "aws-deploy",
+    "go-api",
+    "devcontainer"
+  ],
+  "install": {
+    "method": "git-clone",
+    "postInstall": "global/sync.sh",
+    "note": "Plugin mode installs hooks + rules + commands only. For full features (skills, agents, practices), use git clone + global/sync.sh"
+  },
+  "requires": {
+    "claudeCode": ">=1.0.0"
+  }
+}

.claude/agent-memory/architect/README.md

@@ -0,0 +1,3 @@
+# architect memory
+
+Learnings and discoveries from architect agent sessions.

.claude/agent-memory/code-reviewer/README.md

@@ -0,0 +1,3 @@
+# code-reviewer memory
+
+Learnings and discoveries from code-reviewer agent sessions.

.claude/agent-memory/implementer/README.md

@@ -0,0 +1,3 @@
+# implementer memory
+
+Learnings and discoveries from implementer agent sessions.

.claude/agent-memory/security-auditor/README.md

@@ -0,0 +1,3 @@
+# security-auditor memory
+
+Learnings and discoveries from security-auditor agent sessions.

.claude/commands/audit.md

@@ -14,7 +14,7 @@ Ejecutá una auditoría del proyecto actual contra la plantilla claude-kit.
    - `app.yaml`, `cloudbuild.yaml`, `gcloud` en scripts → gcp-cloud-run
    - `redis` en requirements/pyproject.toml → redis
 
-2. Leer el checklist de auditoría: `~/Documents/GitHub/claude-kit/audit/checklist.md`
+2. Leer el checklist de auditoría: `$CLAUDE_KIT_DIR/audit/checklist.md`
 
 3. Evaluar cada item del checklist contra el proyecto actual:
    - ¿Existe CLAUDE.md? ¿Tiene >20 líneas útiles?
@@ -24,6 +24,6 @@ Ejecutá una auditoría del proyecto actual contra la plantilla claude-kit.
    - ¿Los comandos build/test están documentados?
    - Items recomendados del checklist
 
-4. Calcular score según `~/Documents/GitHub/claude-kit/audit/scoring.md`
+4. Calcular score según `$CLAUDE_KIT_DIR/audit/scoring.md`
 
 5. Generar reporte con formato estándar de auditoría.

.claude/hooks/block-destructive.sh

@@ -1,26 +1,57 @@
 #!/bin/bash
 # PreToolUse hook: block dangerous bash commands
 # Matcher: Bash
+# Supports FORGE_HOOK_PROFILE: minimal | standard (default) | strict
 COMMAND=$(echo "$TOOL_INPUT" | jq -r '.command // empty' 2>/dev/null)
 
-DANGEROUS_PATTERNS=(
+PROFILE="${FORGE_HOOK_PROFILE:-standard}"
+
+# Minimal: only the most catastrophic patterns
+MINIMAL_PATTERNS=(
   'rm -rf /'
   'rm -rf \*'
   'rm -rf ~'
+  'git push.*--force.*main'
+  'git push.*--force.*master'
+)
+
+# Standard: current behavior (minimal + broader destructive ops)
+STANDARD_PATTERNS=(
   'DROP TABLE'
   'DROP DATABASE'
   'TRUNCATE TABLE'
   'DELETE FROM .* WHERE 1'
-  'git push.*--force.*main'
-  'git push.*--force.*master'
   'git reset --hard'
   'docker system prune -a'
   'chmod -R 777'
 )
 
+# Strict: standard + risky execution patterns
+STRICT_PATTERNS=(
+  'curl.*\|.*sh'
+  'wget.*\|.*sh'
+  'eval '
+  'chmod 777'
+  'chmod -R 777'
+  '> /etc/'
+  'tee /etc/'
+  'dd if=.* of=/dev/'
+)
+
+# Build active pattern list based on profile
+DANGEROUS_PATTERNS=("${MINIMAL_PATTERNS[@]}")
+
+if [[ "$PROFILE" == "standard" || "$PROFILE" == "strict" ]]; then
+  DANGEROUS_PATTERNS+=("${STANDARD_PATTERNS[@]}")
+fi
+
+if [[ "$PROFILE" == "strict" ]]; then
+  DANGEROUS_PATTERNS+=("${STRICT_PATTERNS[@]}")
+fi
+
 for pattern in "${DANGEROUS_PATTERNS[@]}"; do
   if echo "$COMMAND" | grep -qiE "$pattern"; then
-    echo "BLOCKED: destructive command detected" >&2
+    echo "BLOCKED: destructive command detected [$PROFILE profile]" >&2
     echo "Pattern: $pattern" >&2
     echo "Command: $COMMAND" >&2
     exit 2

.claude/rules/agents.md

@@ -37,6 +37,8 @@ Spawn an Agent Team ONLY when ALL of these hold:
 Team structure pattern:
 - **Lead**: coordinates, synthesizes, DOES NOT implement
 - **Teammates**: max 3-4 (diminishing returns beyond that)
+- Each teammate MUST use `isolation: "worktree"` to work on an isolated copy of the repo
+- Lead agent coordinates merges from worktree branches into the main branch
 - Require plan approval before any teammate writes code
 
 ## Context Management

.claude/rules/memory.md

@@ -9,7 +9,8 @@ description: "Memory management policy — always loaded"
 - Before modifying code, read CLAUDE_ERRORS.md for known issues in the affected area
 - After fixing a bug, record it: date, area, root cause, fix applied, derived rule
 - If same error appears 3+ times across sessions, promote the derived rule to _common.md or a stack-specific rule
-- Format: markdown table with columns Date | Area | Error | Cause | Fix | Rule
+- Format: markdown table with columns Date | Area | Type | Error | Cause | Fix | Rule
+- Type must be one of: `syntax`, `logic`, `integration`, `config`, `security`
 
 ## Agent Memory (.claude/agent-memory/)
 - Agents with `memory: project` persist learnings in .claude/agent-memory/<agent-name>/