AlexiosBluffMara

Delegate coding tasks to Blackbox AI CLI agent. Multi-model agent with built-in judge that runs tasks through multiple LLMs and picks the best result. Requires the blackbox CLI and a Blackbox AI API key.

Configure and use Honcho memory with Mercury -- cross-session user modeling, multi-profile peer isolation, observation config, dialectic reasoning, session summaries, and context budget enforcement. Use when setting up Honcho, troubleshooting memory, managing profiles with Honcho peers, or tuning observation, recall, and dialectic settings.

Query Base (Ethereum L2) blockchain data with USD pricing — wallet balances, token info, transaction details, gas analysis, contract inspection, whale detection, and live network stats. Uses Base RPC + CoinGecko. No API key required.

Query Solana blockchain data with USD pricing — wallet balances, token portfolios with values, transaction details, NFTs, whale detection, and live network stats. Uses Solana RPC + CoinGecko. No API key required.

Structured decision-making framework for technical proposals and trade-off analysis. When the user faces a choice between multiple approaches (architecture decisions, tool selection, refactoring strategies, migration paths), this skill produces a 1-3-1 format: one clear problem statement, three distinct options with pros/cons, and one concrete recommendation with definition of done and implementation plan. Use when the user asks for a "1-3-1", says "give me options", or needs help choosing between competing approaches.

Control Blender directly from Mercury via socket connection to the blender-mcp addon. Create 3D objects, materials, animations, and run arbitrary Blender Python (bpy) code. Use when user wants to create or modify anything in Blender.

Generate flat, minimal light/dark-aware SVG diagrams as standalone HTML files, using a unified educational visual language with 9 semantic color ramps, sentence-case typography, and automatic dark mode. Best suited for educational and non-software visuals — physics setups, chemistry mechanisms, math curves, physical objects (aircraft, turbines, smartphones, mechanical watches), anatomy, floor plans, cross-sections, narrative journeys (lifecycle of X, process of Y), hub-spoke system integrations (smart city, IoT), and exploded layer views. If a more specialized skill exists for the subject (dedicated software/cloud architecture, hand-drawn sketches, animated explainers, etc.), prefer that — otherwise this skill can also serve as a general-purpose SVG diagram fallback with a clean educational look. Ships with 15 example diagrams.

Generate real meme images by picking a template and overlaying text with Pillow. Produces actual .png meme files.

""Optimize and deploy ML inference on edge devices. Use when: running LiteRT on Snapdragon XR1, configuring GPU/NNAPI delegates, optimizing memory for 6GB RAM constraint, reducing inference latency below 50ms, deploying Gemma 4 on phone, running MLX on Apple Silicon, profiling thermal throttling, battery-aware inference scheduling.""

""Quantize and optimize ML models for edge deployment across all four tiers. Use when: converting models to LiteRT, ONNX export, INT8/FP16 quantization, GGUF conversion, MLX format, BitNet 1-bit quantization, model size reduction, inference optimization, Adreno GPU delegate, NNAPI configuration, profiling model performance on target hardware.""

Create or update a skill and mirror it to both .github/skills/ and .claude/skills/

""Fine-tune and train ML models using Unsloth Dynamic QLoRA for the Duchess platform. Use when: fine-tuning Gemma 4, training Qwen2.5-VL, creating domain adapters (Safety, Spanish, Engineering, Electrical), configuring QLoRA hyperparameters, running training on RTX 5090, dataset preparation for construction safety, model evaluation against benchmarks.""

""Build Android apps for Vuzix M400 glasses and companion smartphones. Use when: developing Kotlin Android code, AOSP without Google Play Services, Vuzix SDK integration, companion phone app, BLE communication, camera pipeline, sideloading APKs, foreground services, Jetpack Compose UI, Android Gradle builds, min SDK 33 for glasses, min SDK 30 for phone.""

""Design AR interfaces for Vuzix M400 glasses and companion phone app. Use when: designing heads-up display layouts, 640x360 OLED optimization, voice UI commands, safety alert presentation, glove-compatible interactions, bilingual UI layout, notification design, wearable ergonomics, OLED power optimization, accessibility for construction workers.""

""English/Spanish localization for construction safety communication. Use when: translating safety alerts, creating bilingual UI strings, construction-register Spanish terminology, code-switching training data, regional Spanish variants, OSHA bilingual compliance, LoRA adapter vocabulary for Spanish construction jargon, toolbox talk translations.""

StreamSession, video frames, photo capture, resolution/frame rate configuration

Plan or revise Savitar architecture, ADRs, contracts, system boundaries, phase sequencing, or multi-surface integration. Use when scoping features, defining interfaces, or breaking work into clean-room milestones.

Add or change a Savitar model route, MCP server, transport, browser automation feature, or memory workflow. Use when implementing new integrations, extending contracts, or wiring new capabilities into the runtime.

Design or refine the Savitar persona. Use when defining speech patterns, public bio, reply style, identity consistency, or human-conversational behavior across CLI, messaging, and web surfaces.

Create or update a Savitar provenance log entry for clean-room work. Use when a feature depends on external documentation, parity targets, third-party products, or public APIs and the repo needs an auditable record of what was consulted.

Review Savitar for security risks. Use when working on public web UI, Google auth, messaging transports, remote execution, MCP/tool safety, secret handling, or prompt-injection-resistant workflows.

Test or validate Savitar runtime changes. Use when adding unit tests, doctor checks, CI coverage, regression validation, or release-readiness checks.

Design or extend Savitar messaging transports and the shared gateway. Use when working on Discord, WhatsApp, iMessage, message normalization, delivery flows, or channel identity policy.

Plan or build the Savitar public web UI. Use when working on Google login, operator dashboard, browser-facing controls, user flows, or public-internet session handling.

Run the complete Jemma offline brain-response pipeline on any media file. Uses Gemma 4 E4B for visual scene description and seven-tier audience narration (toddler to neuroscience researcher), and TRIBE v2 for cortical BOLD prediction. Generates a peak-cortex PNG, multi-atlas analysis, and narrations for all seven expertise levels. Use when the user wants to analyse a video/audio clip for brain response and get explanations for multiple audiences.

Run TRIBE v2 brain-response prediction on a local media file (video, audio, or text). Returns predicted BOLD activity across 20,484 fsaverage5 cortical vertices at 2 Hz, plus multi-atlas ROI analysis (Schaefer-400, Harvard-Oxford, Jülich), Yeo-7 network laterality, temporal dynamics, and a peak-cortex PNG. Use when the user wants to predict brain responses to a video clip, audio recording, or text stimulus.

End-to-end model deployment for Jemma: Ollama local, HuggingFace Hub, Google Cloud Run, GGUF export. Use when: deploying, publishing, exporting, packaging, or releasing a trained model.

Run, monitor, debug, or resume the Jemma overnight training pipeline. Use when: training, fine-tuning, data ingestion, RAG indexing, QLoRA, checkpoint management, or pipeline troubleshooting.