Agent Skills: Export System Prompt

Export System Prompt

Export Claude Code's complete system prompt (all instructions, tool definitions, CLAUDE.md content, memory imports, and session context) to the console or a markdown file.

Arguments

• No arguments: Output formatted system prompt directly to console (default)
• --output: Write to auto-generated path in .claude/temp/
• --output <path>: Write to specified path (.md extension added if missing)

Execution Workflow

Step 1: Parse Arguments

1. Check if --output flag is present in $ARGUMENTS
2. If --output present, check if a path follows it
3. Determine output mode:
   • console (no flag)
   • file_auto (flag without path)
   • file_custom (flag with path)

Step 2: Get Current Timestamp

Run: date -u +"%Y-%m-%dT%H:%M:%SZ" to get UTC timestamp for metadata.

Step 3: Capture System Prompt

The "system prompt" includes everything Claude sees as instructions at the start of each turn:

• Base Claude Code system instructions
• Tool definitions and descriptions
• CLAUDE.md file contents (project instructions)
• Memory file imports (@-file references)
• Hook-injected content (<system-reminder> tags)
• Session-specific context (gitStatus, environment info)

Important: Output the COMPLETE system prompt without truncation. This is the entire context Claude operates under.

Step 4: Format Output

Format as markdown with metadata header:

# System Prompt Export

**Exported:** {UTC timestamp}
**Working Directory:** {cwd from environment}
**Model:** {model name from system context}

---

## Complete System Prompt

{Everything from the system context - tools, instructions, CLAUDE.md, memory, hooks, etc.}

---

*Exported by `/claude-ecosystem:export-system-prompt`*

Step 5: Output Based on Mode

Console Mode (default):

Output the formatted markdown directly. The user will see the complete system prompt in the conversation.

File Mode (--output):

1. Determine output path:

   • If path provided: Use it (append .md if missing)
   • If no path: Generate .claude/temp/YYYY-MM-DD_HHmmss-system-prompt-export.md

2. Write formatted output to file using Write tool

3. Report success: Saved system prompt to: {path}

Output Template

# System Prompt Export

**Exported:** {timestamp}
**Working Directory:** {cwd}
**Model:** {model}

---

## Complete System Prompt

### Tool Definitions

{List all available tools with their descriptions}

### System Instructions

{Base Claude Code instructions}

### Project Instructions (CLAUDE.md)

{Contents of CLAUDE.md and imported memory files}

### Session Context

{gitStatus, environment info, hook-injected reminders}

---

*Exported by `/claude-ecosystem:export-system-prompt`*

Verification Protocol

Before reporting completion, verify:

• [ ] Arguments parsed correctly (detected output mode)
• [ ] Timestamp captured
• [ ] Complete system prompt included (not truncated)
• [ ] Output formatted as markdown with metadata
• [ ] If file mode: File written successfully and path reported
• [ ] If console mode: Full output displayed

Notes

• Large output warning: System prompts can be 10K+ tokens. For easier handling, consider using --output to save to a file.
• Security awareness: The exported prompt contains your repository's CLAUDE.md instructions and configuration. Be mindful when sharing.
• Debugging use case: This command is useful for understanding exactly what instructions Claude is operating under, especially when troubleshooting unexpected behavior.