dev-tools

PDF parsing and structured data extraction specialist. Analyzes PDF content natively through Claude's multimodal capabilities, with Python3 fallback using pdfplumber and CLI tools for complex extractions.

Codebase-aware web research via Perplexity AI. Use ONLY when research needs codebase context (comparing current code to best practices). For simple lookups, call mcp__perplexity-ask__perplexity_ask directly instead.

Analyze Claude Code usage, cost, efficiency, and burn rate using ccusage and termgraph. Use when user says "usage", "cost", "spending", "tokens", "analyze usage", "how much did I spend", "usage report", "budget", "burn rate", "efficiency", "cache hits", "ccusage", "ccw", "ccp".

Turn ideas into designs through collaborative dialogue. Use when user wants to brainstorm, design features, explore approaches, or think through implementation before coding.

Dialectic thinking — spawn thesis and antithesis agents to stress-test ideas, then synthesize and verify against code. Use when user says "debate", "argue both sides", "devil's advocate", "stress test this idea", "pros and cons of approach", or wants rigorous evaluation of a design decision.

Audit Claude Code configuration against latest features and best practices. Use when user says "evolve", "self-improve", "audit config", "what's new in claude code", "upgrade configuration", "check for improvements", "are we up to date".

Explore public GitHub repositories via DeepWiki AI-generated documentation. Use for understanding architecture, patterns, design decisions, and code organization of popular open-source projects. Use when user asks "how does X repo work", "explain architecture of Y", "what patterns does Z use", "explore repo", "deepwiki", or needs codebase-level understanding beyond API docs.

Extract learnings and generate project-specific customizations (CLAUDE.md, commands, skills, hooks). Use when user says "learn", "extract learnings", "what did we learn", "save learnings", "adapt config", or wants to improve Claude Code based on conversation patterns.

Lint plugin agent/skill prompts against rules derived from Anthropic model cards (Opus 4.6, Sonnet 4.6). Use when authoring or reviewing skills and agents — "lint instructions", "audit prompts", "model card rules".

Library documentation via Context7. Use for API references, code examples, framework docs.

Query project history, past decisions, and known gotchas from claude-mem observations. Use when user asks "last session", "did we already", "what did we decide", "project history", "timeline", or "what happened with".

Web research via Perplexity AI. Use for technical comparisons (X vs Y), best practices, industry standards, documentation. Triggers on "research", "compare", "vs", "best practice", "which is better", "pros and cons".

Token-efficient code navigation via AST parsing. Use smart_outline for file structure, smart_search for cross-file discovery, smart_unfold for targeted function extraction. 10-20x fewer tokens than reading full files.

Consult Gemini for second opinions, brainstorming, or web search. Use when user says "ask gemini", "gemini search", "get gemini opinion", or wants a second AI perspective.

Creates isolated git worktrees for parallel development. Use when starting feature work needing isolation or working on multiple branches simultaneously. Not for simple branch switching or basic git operations.

Prefer modern CLI tools for better performance: rg (ripgrep) instead of grep for text searching, fd instead of find for file discovery, bat instead of cat for viewing files, sd instead of sed for text replacement, eza instead of ls for directory listing, dust instead of du, procs instead of ps. Use when writing bash scripts, optimizing command chains, working with file searches, or replacing grep/find/cat in workflows. These tools are faster, respect .gitignore, and have better output formatting.