Claude Code Platform Skill

This skill provides comprehensive expertise on using Claude Code effectively.

Skill Directory: ~/.claude/skills/claude-code-platform/

Core Tool Knowledge

File Operations

Read Tool: Read files with file_path (required), optional offset and limit for large files. Returns up to 2000 lines, supports images/PDFs/Jupyter notebooks.

Write Tool: Create files with file_path and content. MUST Read existing files first. Always prefer Edit over Write for existing files.

Edit Tool: Replace text with file_path, old_string, new_string, optional replace_all. String must be unique or use replace_all. MUST Read file first.

Glob Tool: Find files with pattern (e.g., **/*.ts) and optional path. Returns sorted by modification time.

Grep Tool: Search with pattern, optional path, output_mode (content/files_with_matches/count), -i for case-insensitive, -n for line numbers, multiline for cross-line patterns.

Execution

Bash Tool: Execute commands with command (required), description (recommended), timeout (default 120000ms, max 600000ms), run_in_background (for async execution). Always quote paths with spaces. Use && for dependent commands, parallel tool calls for independent commands.

Task Tool: Launch subagents with subagent_type and prompt. Available types: general-purpose, Explore (quick/medium/very thorough), pragmatic-lead, qa, plus custom agents.

Core Patterns

Parallel Execution: Call multiple independent tools in one message for dramatic speedup.

Tool Selection: Specialized tools (Read/Edit/Grep/Glob) > Bash commands for file operations.

Progressive Loading: Glob → Grep → Read (narrow before deep dive).

When You Need More Details

Complete Tool Parameters and Usage

For detailed parameter specifications, return formats, error handling, and advanced usage patterns, read:

cat ~/.claude/skills/claude-code-platform/tools-reference.md

MCP Integration Patterns

For Model Context Protocol server configuration, tool naming conventions (mcp__server__function), and integration with Linear/Calendar/Stripe etc, read:

cat ~/.claude/skills/claude-code-platform/mcp-integration.md

Advanced Features

For Skills system details, Checkpoints (save/restore state), custom Subagent creation, Hooks (PreToolUse/PostToolUse), and Haiku 4.5 model usage, read:

cat ~/.claude/skills/claude-code-platform/advanced-features.md

Performance Optimization

For parallel execution patterns, context efficiency, workflow patterns (investigation/debugging/research), git protocols, and error recovery, read:

cat ~/.claude/skills/claude-code-platform/best-practices.md

Troubleshooting

For debugging tool failures, permission issues, MCP connection problems, git errors, and common error patterns with solutions, read:

cat ~/.claude/skills/claude-code-platform/troubleshooting.md

Quick Reference Examples

Reading multiple files in parallel:

✓ Read file1.ts + Read file2.ts + Read file3.ts (one message)
✗ Read file1.ts → wait → Read file2.ts (sequential)

Finding and searching:

Step 1: Glob **/*.ts
Step 2: Grep "pattern" --type ts --output-mode files_with_matches
Step 3: Read specific matches

Bash best practices:

✓ git add . && git commit -m "msg" && git push (sequential with &&)
✓ cd "/path with spaces/" (quoted paths)
✗ cat file.txt (use Read tool instead)

MCP tool naming: mcp__server-name__function-name

Task tool for deep work: Use Explore agent (thoroughness: quick/medium/very thorough) for codebase exploration.

Haiku vs Sonnet: Haiku 4.5 = routine tasks (2x speed, 1/3 cost), Sonnet 4.5 = complex reasoning.

Key Principles

1. Specialized tools > Bash for file operations
2. Parallel when independent for speed
3. Progressive disclosure - Glob → Grep → Read
4. Read before Write/Edit always
5. Verify after critical operations
6. Clear tool descriptions for user transparency

This Skill Uses Progressive Disclosure

• Level 1 (This file): Core tool knowledge and quick reference
• Level 2 (Reference files): Detailed docs loaded via bash when needed
• Level 3 (On-demand): Specific sections of reference files as required

When you need deeper information, read the appropriate reference file from the skill directory using bash commands shown above.

Version: 2.0 - Updated Oct 16, 2025 for proper progressive disclosure