NLP

Orchestrate and run appropriate pensive review skills based on codebaseanalysis and context.Triggers: code review, unified review, full review, review orchestration,multi-domain review, intelligent review, auto-detect reviewUse when: general review needed without knowing which specific skill applies,full multi-domain review desired, integrated reporting neededDO NOT use when: specific review type known - use bug-review, test-review, etc.DO NOT use when: architecture-only focus - use architecture-review.Use this skill when orchestrating multiple review types.

Detect codebase bloat through progressive analysis: dead code, duplication, complexity, and documentation bloat.Triggers: bloat detection, dead code, code cleanup, duplication, redundancy, codebase health, technical debt, unused codeUse when: preparing for refactoring, context usage is high, quarterly maintenance, pre-release cleanupDO NOT use when: actively developing new features, time-sensitive bug fixes.DO NOT use when: codebase is < 1000 lines (insufficient scale for bloat).Progressive 3-tier detection: quick scan → targeted analysis → deep audit.

Gemini CLI delegation workflow implementing delegation-core for Google's Gemini models.Triggers: gemini cli, gemini delegation, google gemini, 1M context, large file analysis,gemini batch, gemini summarization, gemini extractionUse when: delegation-core selected Gemini, need Gemini's 1M+ token context window,batch processing or large document summarization requiredDO NOT use when: deciding which model to use - use delegation-core first.DO NOT use when: gemini CLI not installed or authenticated.Consult this skill when implementing Gemini-specific delegation workflows.

Construct temporary, session-specific memory palaces for complex projectsand conversations.Triggers: session context, project memory, conversation state, temporary storage,session palace, context preservation, complex project, extended conversationUse when: working on complex multi-step projects, preserving context acrossinterruptions, tracking session-specific stateDO NOT use when: permanent knowledge structures needed - use memory-palace-architect.DO NOT use when: searching existing knowledge - use knowledge-locator.Consult this skill for session-scoped temporary knowledge structures.

Reduce context usage with MECW principles (keep under 50% of total window).Triggers: context pressure, token usage, MECW, context window, optimization,decomposition, workflow splitting, context management, token optimizationUse when: context usage approaches 50% of window, tasks need decomposition,complex multi-step operations planned, context pressure is highDO NOT use when: simple single-step tasks with low context usage.DO NOT use when: already using mcp-code-execution for tool chains.Use this skill BEFORE starting complex tasks. Check context levels proactively.

Master Python asyncio, concurrent programming, and async/await patternsfor high-performance applications.Triggers: asyncio, async/await, coroutines, concurrent programming, async API,I/O-bound, websockets, background tasks, semaphores, async context managersUse when: building async APIs, concurrent systems, I/O-bound applications,implementing rate limiting, async context managersDO NOT use when: CPU-bound optimization - use python-performance instead.DO NOT use when: testing async code - use python-testing async module.Consult this skill for async Python patterns and concurrency.

Methodology for summarizing changes, extracting insights, and identifying follow-up actions.Triggers: catchup, what changed, summarize changes, context acquisition, handoff,progress review, recent changes, git log analysis, sprint summaryUse when: resuming work after absence, preparing handoff documentation, reviewingsprint progress, analyzing git history for contextDO NOT use when: doing detailed diff analysis - use diff-analysis instead.DO NOT use when: full code review needed - use review-core instead.Use this skill to quickly understand "what changed and what matters".

Transform tool-heavy workflows into MCP code execution patterns for token savings and optimized processing.Triggers: MCP, code execution, tool chain, data pipeline, tool transformation, batch processing, workflow optimizationUse when: >3 tools chained sequentially, large datasets (>10k rows), large files (>50KB), context usage >25%DO NOT use when: simple tool calls that don't chain.DO NOT use when: context pressure is low and tools are fast.Use this skill BEFORE building complex tool chains. Optimize proactively.

Foundational workflow for preparing and structuring detailed reviews(architecture, API, code quality).Triggers: review workflow, structured review, review scaffolding, evidence capture,review preparation, analysis framework, review templateUse when: starting any detailed review workflow, needing consistent structurefor capturing context and findings, ensuring comparable review outputsDO NOT use when: quick catchup without formal review - use catchup.DO NOT use when: diff-focused analysis - use diff-analysis.Use this skill at the BEGINNING of any detailed review for consistent structure.

Qwen CLI delegation workflow implementing delegation-core for Alibaba's Qwen models.Triggers: qwen cli, qwen delegation, alibaba qwen, qwen batch, multi-file analysis,qwen summarization, qwen extraction, 100K contextUse when: delegation-core selected Qwen, need Qwen's large context capabilities,batch processing or multi-file analysis requiredDO NOT use when: deciding which model to use - use delegation-core first.DO NOT use when: qwen CLI not installed or configured.Consult this skill when implementing Qwen-specific delegation workflows.

Minimize token usage through conservative prompting, work delegation,and quota tracking.Triggers: token usage, quota, token limits, prompt size, token conservation,usage tracking, delegation, context compression, token budgetUse when: session starts (mandatory), prompt sizes spike, tool calls increase,before long-running analyses or massive context loadsDO NOT use when: context-optimization already handles the scenario.DO NOT use when: simple queries with minimal context.Use this skill at the START of every session. This is MANDATORY for quota management.

Generate and edit images using the Gemini API (Nano Banana Pro). Use this skill when creating images from text prompts, editing existing images, applying style transfers, generating logos with text, creating stickers, product mockups, or any image generation/manipulation task. Supports text-to-image, image editing, multi-turn refinement, and composition from multiple reference images.

Comprehensive guide to PostgreSQL migrations - common errors, generated columns, full-text search, indexes, idempotent migrations, and best practices for database schema changes

Use when design is complete and you need detailed implementation tasks for engineers with zero codebase context - creates comprehensive implementation plans with exact file paths, complete code examples, and verification steps assuming engineer has minimal domain knowledge

Track user journeys with intent context and friction signals. Use when instrumenting onboarding, checkout, or any multi-step flow where you need to understand WHY users fail.

Creating and maintaining CLAUDE.md project memory files that provide non-obvious codebase context. Use when (1) creating a new CLAUDE.md for a project, (2) adding architectural patterns or design decisions to existing CLAUDE.md, (3) capturing project-specific conventions that aren't obvious from code inspection.

Educational GPT implementation in ~300 lines. Reproduces GPT-2 (124M) on OpenWebText. Clean, hackable code for learning transformers. By Andrej Karpathy. Perfect for understanding GPT architecture from scratch. Train on Shakespeare (CPU) or OpenWebText (multi-GPU).

Fast tokenizers optimized for research and production. Rust-based implementation tokenizes 1GB in <20 seconds. Supports BPE, WordPiece, and Unigram algorithms. Train custom vocabularies, track alignments, handle padding/truncation. Integrates seamlessly with transformers. Use when you need high-performance tokenization or custom tokenizer training.

Trains large language models (2B-462B parameters) using NVIDIA Megatron-Core with advanced parallelism strategies. Use when training models >1B parameters, need maximum GPU efficiency (47% MFU on H100), or require tensor/pipeline/sequence/context/expert parallelism. Production-ready framework used for Nemotron, LLaMA, DeepSeek.

Use when design is complete and you need detailed implementation tasks for engineers with zero codebase context - creates comprehensive implementation plans with exact file paths, complete code examples, and verification steps assuming engineer has minimal domain knowledge