Fixed YAML parsing errors in 4 skill files that were missing the
closing '---' delimiter in their frontmatter, causing skill loading
failures.
Fixed files:
- skills/auto-superpowers/SKILL.md
- skills/cognitive-context/SKILL.md
- skills/cognitive-planner/SKILL.md
- skills/cognitive-safety/SKILL.md
Co-Authored-By: Claude <noreply@anthropic.com>
Add clipboard image helper scripts to enable easy image paste functionality
in Claude Code CLI using the Z.AI vision MCP server.
Features:
- clippaste-image: Simple clipboard image extractor (Wayland/X11)
- clipimage-paste: Advanced handler with JSON/base64/watch modes
- IMAGE-PASTE-README.md: Complete documentation
Integration:
- Works with Ctrl+V in Claude Code CLI
- Automatic Wayland (wl-paste) and X11 (xclip) detection
- Z.AI MCP server provides 8 vision analysis tools
Usage:
clippaste-image # Paste to temp file
clippaste-image ~/screenshot.png # Paste to specific file
clipimage-paste --json # Get JSON with metadata
clipimage-paste --watch # Watch for new images
Co-Authored-By: Claude <noreply@anthropic.com>
Added comprehensive Ralph Orchestrator integration documentation:
1. requirements.txt
- ralph-orchestrator (core dependency for /ralph command)
- pyyaml (for config parsing)
- Comments explaining each dependency
2. Updated supercharge.sh
- Enhanced Ralph installation with dual detection:
* Checks for 'ralph' command in PATH
* Checks for ralph_orchestrator Python package
- Installs from requirements.txt if available
- Falls back to direct pip install
- Provides helpful error messages with installation instructions
- Confirms successful installation with detailed status
3. Updated README.md with new section "Ralph Orchestrator"
- What is Ralph Orchestrator and how it works
- Installation methods (automatic, manual, system-wide)
- Detailed breakdown of our Ralph integration:
* bin/ralphloop wrapper script (6,290 bytes, 223 lines)
* skills/ralph/ Claude Code skill
* skills/brainstorming/ multi-AI integration
* hooks/ralph-auto-trigger.sh automatic triggering
* hooks/qwen-consult.sh Qwen AI consultation
- File structure created by Ralph
- Environment variables reference
- Troubleshooting guide
- Ralph Orchestrator source links
The /ralph command will now work properly on:
- This machine (ralph already installed)
- Fresh clones (supercharge.sh auto-installs ralph-orchestrator)
- Any machine with pip3 available
Co-Authored-By: Claude <noreply@anthropic.com>
Found and copied original files from ~/.claude installation:
- hooks/ - Original Qwen and Ralph hook scripts with full functionality
- commands/ - Original command definitions (brainstorm, write-plan, execute-plan)
- bin/ralphloop - Original 223-line Python wrapper (6,290 bytes)
- scripts/sync-agents.sh - Original sync script with GitHub/Gitea backup
- templates/ - Original config templates from working installation
- plugins/ - Original comprehensive plugin README
Files sourced from:
- ~/.claude/skills/skills/hooks/
- ~/.claude/skills/skills/commands/
- ~/.claude/skills/skills/templates/
- /home/uroma/obsidian-web-interface/bin/ralphloop
- ~/.claude/agents/sync-agents.sh
These are the production files from the working Claude Code
installation, replacing the placeholder files I created earlier.
Co-Authored-By: Claude <noreply@anthropic.com>
Added all directories and files expected by supercharge.sh:
- hooks/ (5 hook scripts for session management and AI consultation)
- commands/ (3 custom slash commands: brainstorm, write-plan, execute-plan)
- plugins/ (plugin references for glm-plan, rust-analyzer, marketplaces)
- bin/ralphloop (Ralph Orchestrator wrapper for autonomous iteration)
- scripts/sync-agents.sh (agent synchronization script)
- templates/ (config templates: settings, hooks, config.json)
This completes the repository structure so supercharge.sh can install
all components without warnings.
Co-Authored-By: Claude <noreply@anthropic.com>
- Moved 3 new skill folders to skills/: python-patterns, react-best-practices, release-skills
- Removed 48 duplicate skill folders from root (already existed in skills/)
- Repository root now only contains: agents/, skills/, and config files
Co-Authored-By: Claude <noreply@anthropic.com>
Users are now warned that the PowerShell installer is experimental
and WSL is recommended for reliable installation.
Changes:
- README: Added warning banner, changed WSL to "Recommended"
- PowerShell script: Added interactive warning at startup
- WordPress article: Added yellow warning box before Windows section
PowerShell installer now shows:
- Yellow warning about experimental status
- WSL command for reliable installation
- Prompt to press Ctrl+C to exit or Enter to continue
Co-Authored-By: Claude <noreply@anthropic.com>
Changed from iterating Get-ChildItem to using Copy-Item with wildcard
This ensures proper recursive copying of all files and subdirectories
Before: Get-ChildItem | ForEach-Object { Copy-Item ... }
After: Copy-Item -Path "$Source\*" -Destination $Destination -Recurse
This prevents issues where nested directories aren't copied correctly.
Co-Authored-By: Claude <noreply@anthropic.com>
Gitea's raw file URL format requires /raw/branch/{branch}/ not /raw/{branch}/
This fixes the 403 redirect issue when downloading install-windows.ps1
Changed URLs:
- /raw/main/install-windows.ps1 → /raw/branch/main/install-windows.ps1
Also updated WordPress article with correct URLs.
Co-Authored-By: Claude <noreply@anthropic.com>
Windows has reserved device names that cannot be used as file names:
- CON, PRN, AUX, NUL
- COM1-9, LPT1-9
Changed backup function from Copy-Item to Robocopy which properly
handles these reserved names. This fixes the error when backing up
.claude directories that contain files with these names.
Robocopy exit codes:
- 0-7 = Success
- 8+ = Errors
Co-Authored-By: Claude <noreply@anthropic.com>
Features:
- Automatic download of repository from GitHub
- Full installation of all components (skills, agents, hooks, commands, plugins, scripts)
- Claude Code detection and optional installation via npm
- Dependency checking (Node.js, Git, Python, Ralph Orchestrator)
- Automatic backup of existing configuration
- Color-coded output for better user experience
- Error handling with clear messages
Usage:
irm https://.../install-windows.ps1 | pwsh -ExecutionPolicy Bypass -
OR right-click "Run with PowerShell"
Updated README with PowerShell installer as Option 1 (recommended)
Updated WordPress article with Windows PowerShell option
Co-Authored-By: Claude <noreply@anthropic.com>
- Remove broken one-line curl install commands
- Add proper git clone + run installer instructions
- Add clear explanation that repo must be cloned first
- Fix Windows WSL and Git Bash instructions
- Add warning about curl in PowerShell
The installer script needs access to files in the repository
(skills/, agents/, hooks/, etc.) so one-line piping won't work.
Co-Authored-By: Claude <noreply@anthropic.com>
Added 3 Windows installation methods:
1. WSL wrapper: wsl bash -c "curl ... | bash"
2. Git Bash instructions
3. Download & Run with Invoke-WebRequest
Added warning: Do NOT run curl directly in PowerShell.
Co-Authored-By: Claude <noreply@anthropic.com>
- Changed PowerShell references to WSL/Git Bash
- Added clear Windows user guidance
- Removed non-existent .ps1 script references
- Added note: "Windows Users: Run commands in WSL, Git Bash, or download scripts"
PowerShell versions will be added in future update.
Co-Authored-By: Claude <noreply@anthropic.com>
Added PowerShell equivalents for all installation commands:
- Quick one-liners section with Windows commands
- Detailed installation guide with PowerShell options
- Updated script references to include .ps1 files
- Cross-platform option documentation (--skip-deps / -SkipDeps)
Co-Authored-By: Claude <noreply@anthropic.com>
Root Cause: When script is executed via `curl | bash`, the prompt text
from `read -p "prompt"` is not displayed. The user sees nothing and
thinks the script is stuck/hanging, when it's actually waiting for input.
Solution: Use explicit `echo` statements for prompts instead of `read -p`:
Before (prompt hidden):
read -p "Launch coding-helper wizard now? [y/N] " -n 1 -r var < /dev/tty
After (prompt visible):
echo -e "${CYAN}Launch coding-helper wizard now?${NC} ${GREEN}[Y/n]${NC}"
>&2 echo -en "\033[0;36m> \033[0m"
read -n 1 -r var < /dev/tty 2>/dev/null || var="Y"
The key changes:
1. Echo the prompt text explicitly before calling read
2. Use >&2 to send to stderr (bypasses pipe)
3. Show visual indicator "> " using stderr
4. read command now just waits for input (no -p flag)
5. Default to "Y" if read fails (makes script usable non-interactively)
Fixed 5 wizard launch prompts:
- Line 652: All components installed case
- Line 668: Partial components case
- Line 697: Already installed case
- Line 724: Normal install success case
- Line 775: Sudo install success case
All now show:
Launch coding-helper wizard now? [Y/n]
>
And wait for user input with visible prompt.
Co-Authored-By: Claude <noreply@anthropic.com>
Root Cause: The script has `set -e` which causes immediate exit when
any command fails. When npm install fails, the script exits before
showing the error message. Additionally, `read < /dev/tty` can fail
in WSL or certain terminal environments, causing silent script exit.
Solution: Wrap all read commands and npm install with `set +e`/`set -e`
to prevent premature exit:
1. npm install command (line 708-712):
- Disables set -e before capturing output
- Re-enables after getting exit code
- Allows error handling to execute
2. All read commands with /dev/tty:
- Wrapped with set +e / set -e
- Added fallback defaults: || var="default"
- Added 2>/dev/null to suppress read errors
- Prevents exit if /dev/tty unavailable
3. Protected read locations:
- Line 698: Launch wizard prompt (already installed case)
- Line 722: Launch wizard prompt (success case)
- Line 761: Sudo retry prompt (defaults to "Y")
- Line 773: Launch wizard prompt (sudo success case)
This is critical for WSL users where /dev/tty may not work properly
with curl | bash execution.
Example fix:
```bash
# Before (exits on read failure):
read -p "Prompt? " -n 1 -r var < /dev/tty
# After (continues on read failure):
set +e
read -p "Prompt? " -n 1 -r var < /dev/tty 2>/dev/null || var="Y"
set -e
```
Co-Authored-By: Claude <noreply@anthropic.com>
Root Cause: The sudo and install prompts were not visually prominent,
causing users to miss them or not respond clearly. Additionally, the
prompts had [Y/n] format but didn't default to Yes on Enter.
Changes:
1. Main install prompt (line 674-676):
- Added prominent MAGENTA boxed prompt
- "Install Z.AI coding-helper now? [Y/n]"
- Defaults to Yes when user presses Enter
- Visual box makes it stand out from other output
2. Sudo retry prompt (line 747-748):
- Added prominent YELLOW boxed prompt
- "Try installing with sudo now? [Y/n] (recommended)"
- Defaults to Yes when user presses Enter
- Clear "(recommended)" label
3. Default handling:
- Changed from `[y/N]` to `[Y/n]` (Yes is default)
- Added check for empty input (Enter key) to treat as Yes
- `if [[ -z "$var" ]] || [[ $var =~ ^[Yy]$ ]]`
Example output:
═══════════════════════════════════════════════════════════
Install Z.AI coding-helper now? [Y/n]
═══════════════════════════════════════════════════════════
This makes it much harder to miss the prompts and provides a better
user experience by accepting Enter as "Yes" (the common case).
Co-Authored-By: Claude <noreply@anthropic.com>
Root Cause: npm install -g @z_ai/coding-helper was failing with EACCES
(permission denied) when trying to create /usr/local/lib/node_modules/@z_ai,
but the error was being suppressed by 2>/dev/null in the original version.
The script would report "Installation completed" even though it failed.
Solution: Enhanced error handling with three paths:
1. Success Path (exit code = 0):
- Verify installation with 'npm list -g'
- Show "installed and verified!" message
- Offer to launch wizard
2. Permission Error Path (EACCES detected):
- Clear error message explaining the permission issue
- Show 3 options:
a) Use sudo (with interactive retry)
b) Fix npm permissions permanently (chown command)
c) Run without installation using npx
- Interactive prompt to try sudo immediately
3. Other Error Path:
- Show full error output
- Suggest manual installation with sudo
- Remind that npx works without installation
Changes:
- install_coding_helper(): Added EACCES detection, sudo retry option,
and clear error messages for all failure scenarios
- Error output is now captured and displayed for debugging
- Added verification step after sudo installation
The script now properly handles the common case where users don't have
write permissions to /usr/local/lib/node_modules and offers multiple
solutions including an interactive sudo retry.
Co-Authored-By: Claude <noreply@anthropic.com>
Fixed 3 bugs related to coding-helper detection and installation:
1. Detection Bug: Wizard and bilingual checks only looked for 'chelper'
command, but the npm package @z_ai/coding-helper doesn't create this
command - it's run via 'npx @z_ai/coding-helper'.
Fixed by adding 'npm list -g @z_ai/coding-helper' check to both:
- wizard detection (line 505)
- bilingual detection (line 542)
2. Misleading Success Message: When npm install failed, the script said
"Installation completed. You can run manually with:" which sounded
like it succeeded.
Fixed by:
- Capturing both stdout and stderr
- Checking actual exit code
- Showing "Installation failed!" with error output
- Providing clearer manual installation instructions
3. No Installation Verification: After npm install appeared to succeed,
there was no verification that the package was actually installed.
Fixed by:
- Adding npm list -g verification after install
- Showing "installed and verified!" on success
- Warning if installation appeared to succeed but verification failed
Changes:
- detect_coding_helper_components(): Added npm list check to wizard
and bilingual detection
- install_coding_helper(): Added error capture, exit code checking,
verification step, and proper error messages
- Removed misleading "chelper" command from success message (it's
npx @z_ai/coding-helper, not chelper)
Co-Authored-By: Claude <noreply@anthropic.com>
Added comprehensive component detection that shows which features are
already installed and which are missing.
New functions:
- detect_coding_helper_components(): Returns JSON with status of all
10 components (core, wizard, multi_tool, mcp_config, api_management,
bilingual, mcp_servers, marketplace, tool_detection, backup_sync)
- display_component_status(): Shows colored ✓ or ✗ with [Installed]
or [Not installed] status
Enhanced offer_coding_helper_addons():
- Detects and displays status for each component individually
- Shows summary: "X/10 components installed"
- Smart handling based on installation state:
* All installed: Offer to launch wizard
* Core installed, missing components: Offer to configure
* Not installed: Offer to install
Detection checks:
- Core: chelper command or npm package
- Wizard: chelper availability
- Multi-tool: ~/.chelper config directory
- MCP config: mcpServers in settings.json
- API management: Z.AI base URL in settings
- Bilingual: inherent to chelper
- MCP servers: plugins/*mcp* or *zai* count
- Marketplace: marketplace in config.json
- Tool detection: tool-discovery-agent skill
- Backup sync: sync-agents.sh script
Co-Authored-By: Claude <noreply@anthropic.com>
Fixed 8 echo commands missing -e flag that were displaying raw
escape sequences like \033[0;32m instead of colored output.
- Z.AI coding-helper feature list (5 lines)
- Manual command hint
- Discount link display (2 lines)
Co-Authored-By: Claude <noreply@anthropic.com>
All `read -p` commands now redirect stdin from /dev/tty instead of
piped stdin. This fixes the bug where menu prompts immediately failed
with "Invalid choice" when script was executed via curl | bash.
Fixed 10 read commands throughout the script:
- Provider selection menu
- API key input prompts
- Installation confirmation prompts
- Coding-helper installation prompts
Co-Authored-By: Claude <noreply@anthropic.com>
Major Features Added:
- Provider Switching: Detect current API provider (Anthropic/Z.AI) and offer switching
- API Key Update: Easy update of existing API keys without reinstalling
- Z.AI Coding-Helper Integration: Optional installation of @z_ai/coding-helper
Provider Detection:
- Detects Z.AI by checking for dashscope-intl.aliyuncs.com URL
- Detects Anthropic by checking for anthropic.com or ANTHROPIC_API_KEY
- Shows current provider and API key (masked) when script runs
Provider Switching Options:
1. Switch to Z.AI (GLM models: glm-4.5-air, glm-4.7)
2. Switch to Anthropic (Official Claude models)
3. Update API key only (keep current provider)
4. Skip configuration
Z.AI Coding-Helper Addons:
- Interactive wizard for easy setup
- Multi-tool management (Claude Code, OpenCode, Crush, Factory Droid)
- MCP service configuration
- API key management for Global and China plans
- Bilingual interface (English/Chinese)
- Automatic tool detection and installation
- Configuration backup and sync
Configuration Management:
- Automatic backup of settings.json before changes
- Smart merging of existing configurations
- Proper model switching (GLM ↔ Claude models)
- Base URL configuration for Z.AI
Enhanced Banner:
- Shows all three main features (Z.AI, Provider Switching, Coding-Helper)
README Updates:
- Documented new features in install-claude-code.sh section
- Added provider switching workflow explanation
- Listed coding-helper addon features
Added comprehensive one-line installation options for all scripts:
Quick One-Liners Section (top of README):
- SuperCharge installer one-liner
- Claude Code + Z.AI installer one-liner
- Manual instructions one-liner
Detailed Installation Guide:
- SuperCharge Installer with all options
- Claude Code + Z.AI Installer with all options
- All scripts support argument passing via bash -s --
Available Scripts Reference:
- supercharge.sh: Main installer with --skip-deps, --dev-mode
- install-claude-code.sh: Claude Code + Z.AI with --auto, --manual, --skip-install
Total: 16 one-liner command examples
Features:
- curl | bash for direct execution
- bash -s -- for argument passing
- All script options documented
- Clear tables showing options and descriptions
Added discount offer link in multiple locations:
- README.md: In "What is Z.AI API?" section
- install-claude-code.sh: In hero banner, API key prompt, and manual instructions
Discount link: https://z.ai/subscribe?ic=R0K78RJKNW
Features:
- New install-claude-code.sh script for automated Claude Code installation
- Z.AI API integration (GLM models: glm-4.5-air, glm-4.7)
- Automated and manual installation modes
- Supercharge.sh now offers to install Claude Code if not present
- Updated README with comprehensive Claude Code installation guide
Z.AI Models:
- glm-4.5-air: Fast, efficient model (Haiku class)
- glm-4.7: Powerful model (Sonnet/Opus class)
Documentation: https://docs.z.ai/devpack/tool/claude
Fixed Windows line endings (CRLF) in all shell scripts to Unix format (LF):
- supercharge.sh (main installation script)
- agents/*.sh (all agent management scripts)
- hooks/*.sh (all hook scripts)
- skills/ui-ux-pro-max/scripts/wordpress_safe_update.sh
All scripts now pass bash syntax validation.
·