mybigday

Analyse what real BRICKS devices and their users actually did, from `bricks device metrics` (uptime heartbeat, memory, disk) and `bricks activity-log` (every runtime event, plus screenshot history when enabled). Use for fault diagnosis (freeze, restart, wrong…

Use when the user wants to extract a reusable template (prompt) from an existing BRICKS application. Analyzes the project's design, data flow, and key features, then produces a concise 5-10 item requirement list that can be used as a prompt to recreate a…

Use before adding any AI feature to a BRICKS app — chat, Q&A, voice interface, transcription, semantic search / RAG, or AI-generated media. Picks the right generator for the feature and target platform from the built-in catalog - local LLMs (llm/GGML, mlxLlm,…

Use when choosing or changing the actual model for an AI feature in a BRICKS app — which GGUF, which quantization, which size — and when checking whether target devices can run it. Encodes hardware sizing rules - RAM budgets (the llm/GGML generator requires…

Use when implementing a complete AI feature in a BRICKS app end-to-end, after generator and model choice. Proven wiring patterns with events, actions, and Property Bank state - voice kiosk (VAD → STT → LLM → TTS, or the assistant generator with MCP tools and…

Use when the user brings a Word document (.docx) into their BRICKS project and wants its content in the app — menus, price lists, policies, product sheets, event programs, manuals, service guides. Extracts headings, paragraphs, lists, tables, and embedded…

Use when the user brings a PDF into their BRICKS project and wants its content in the app — menus, price lists, schedules, catalogs, brochures, reports, or scanned documents. Extracts text on the design machine with a bundled script, routes the results into…

Use when the user brings a presentation (.pptx) into their BRICKS project — company decks, menus designed as slides, event programs — and wants it on screen. Extracts per-slide text, titles, speaker notes, and embedded images on the design machine with a…

Set up training and inference on AMD Ryzen AI Max+ 395 / Strix Halo (gfx1151, RDNA 3.5) with TheRock nightly ROCm wheels. Triggered when the host has gfx1151, when `rocminfo` shows Strix Halo, or when the user mentions Strix Halo / Ryzen AI Max / gfx1151 /…

Recreate the ml-intern-kit Python environment on a new machine (laptop, rented GPU box, Docker, fresh checkout). Triggered when the user is on a new host, sees ImportError on a core dep (torch/transformers/trl/peft/accelerate), or wants to install flash-attn…

Evaluate a trained or downloaded language model with `lm-eval-harness` standard tasks (arc, hellaswag, gsm8k, mmlu, truthfulqa, ifeval, ...). Triggered when the user wants to benchmark, eval, or compare a model — pre- or post-training.

Audit a Hugging Face dataset before training to confirm splits, columns, format, sample rows, distributions, and duplicates. Triggered before any training/fine-tuning script runs, when a user mentions a new dataset, or when you hit a KeyError / format…

Submit a training/inference script to Hugging Face Jobs (`hf jobs run`). Triggered when the user wants to run training in the cloud, scale beyond local hardware, or kick off a multi-hour fine-tune. Enforces pre-flight: hub_model_id required, ≥2h timeout for…

Run a literature-first crawl before writing ANY ML training/fine-tuning/inference code. Spawns an Explore sub-agent that mines papers, citation graphs, methodology sections, and matched HF datasets to produce a ranked list of training recipes attributed to…

Direct Preference Optimization (DPO) fine-tune with TRL `DPOTrainer`. Triggered when the user wants to align a model on preferences / pairwise comparisons / chosen-vs-rejected data, or improve an existing SFT checkpoint with a preference dataset.

LoRA fine-tune a causal LM natively on Apple Silicon (M1/M2/M3/M4) via `mlx-lm`. Triggered when the user is on macOS arm64, mentions MLX, or wants better throughput than torch-MPS. Use this instead of `train-sft` when the host is a Mac with no NVIDIA GPU.

Guidelines for developing the codec.cpp project, ensuring parity with llama.cpp architecture and conventions.

Guide for adding a new codec model (GGUF conversion + ggml graph integration) in codec.cpp.

Guide for adding or modifying ggml ops in codec.cpp with backend-safe patterns.

Build and run the llama.rn example app on real hardware to validate native (cpp/) changes end-to-end - iOS-on-Mac ("Designed for iPad" on Apple Silicon, real Metal GPU) and Android on a connected Snapdragon device with ggml-hexagon (HTP/NPU) enabled. Use this…

Upgrade React Native version in example/ directory using rn-diff-purge methodology. Use when the user wants to upgrade react-native, migrate to a new RN version, or update React Native dependencies in the example app. Handles diff retrieval, systematic file…

Run Maestro UI tests on the example app in iOS/Android simulators. Use when testing UI flows, verifying test suite results, automating the example app, running maestro commands, creating test flows, or validating changes via the on-device test runner.…