xiaden

Methods for adaptive/token-budgeted visual generation — token merging, token pruning, early exiting, and self-budgeting tokenization in diffusion/image generation models.

Attention Residuals (AttnRes) implementation in the PDLI experiment. Covers AttnResDepth, AttnResBlock, and the L2RTP model that uses them. Essential for anyone modifying the transformer architecture or residual connections.

Comprehensive dependency maintenance — verifies import health, requirements.txt consistency, import hierarchy, and environment health.

Comprehensive paper library maintenance — cross-references papers.md against summaries, techniques, and PDFs. Detects orphans, missing registrations, and pulls missing papers.

Add a new research paper to the library — fetch the paper, write a structured summary to `docs/upstream/papers/`, implement its technique as a standalone Python module in `resources/techniques/`, and register it in `references/papers.md`.

Multi-scale/pyramidal approaches to denoising diffusion models. Covers the Ryu & Ye 2022 pyramidal DDPM paper and related coarse-to-fine diffusion architectures. Relevant to subdivision mode / spatial budget training and hierarchical latent backbones in the…

Comprehensive artifact maintenance — validates log naming, decision format, plan syntax compliance, staleness detection, and ensures the artifact corpus is parseable.

Dataset management maintenance — validates dataset integrity, version tracking, storage health, and cache consistency for training datasets.

Reference for the Nomarr Docker development environment. Use when working with Docker containers, running e2e tests, querying ArangoDB directly, debugging prod-like issues, testing DB migrations, or interacting with the containerized Nomarr API. Contains…

Generates a structured prompt for spawning Support-Librarian to gather relevant ADRs, logs, and design docs before a design, planning, or execution task. Use before starting any significant work — designing a feature, creating a plan, executing a plan, or…

Fetch authoritative, up-to-date external documentation using Context7 MCP tools. Use proactively — without being asked — whenever writing or reviewing code that involves a third-party library or framework API (method signatures, config keys, behavior); a…

Guide for building Python applications using the GitHub Copilot SDK. Use when integrating GitHub Copilot into Python apps: creating CopilotClient sessions, defining custom tools, handling streaming responses, managing session lifecycle, Pydantic parameter…

Use when executing implementation plans produced by the feature-planning skill. Orchestrates execution subagents (one plan phase at a time), dispatches review subagents for thorough quality enforcement after each plan, and manages fix cycles when review finds…

Use when decomposing a major feature design into dependency-ordered implementation plans. Handles the full pipeline from design document to validated, cross-referenced plan files with minimal drift. Trigger when user mentions planning a large feature,…

Use when checking code quality, finding violations, detecting complexity issues, or running QC checks. Provides scripts for linting, naming enforcement, dead code detection, and legacy code discovery.

Guide for creating effective skills. This skill should be used when users want to create a new skill (or update an existing skill) that extends Claude's capabilities with specialized knowledge, workflows, or tool integrations.

Create a subsystem orientation skill after deep research on a stable, non-churning part of the codebase when no skill exists for it. Use when you have just read many files, traced call chains, or explored architecture to understand how a subsystem works — and…

Use when executing implementation plans produced by the feature-planning skill. Orchestrates execution subagents (one plan phase at a time), dispatches review subagents for thorough quality enforcement after each plan, and manages fix cycles when review finds…

Use when decomposing a major feature design into dependency-ordered implementation plans. Handles the full pipeline from design document to validated, cross-referenced plan files with minimal drift. Trigger when **any** of the following are true, with user…

Use when creating or updating VS Code Copilot agent hook files in .github/hooks. Covers all 8 hook event types (PreToolUse, PostToolUse, SessionStart, SubagentStart, SubagentStop, Stop, UserPromptSubmit, PreCompact), the stdin/stdout JSON contract,…

Guide for creating effective skills. This skill should be used when users want to create a new skill (or update an existing skill) that extends Claude's capabilities with specialized knowledge, workflows, or tool integrations.

Guide for creating high-quality MCP (Model Context Protocol) servers that enable LLMs to interact with external services through well-designed tools. Use when building MCP servers to integrate external APIs or services, whether in Python (FastMCP) or…

Generates a structured prompt for spawning Support-Librarian to gather relevant ADRs, logs, and design docs before a design, planning, or execution task. Use before starting any significant work — designing a feature, creating a plan, executing a plan, or…

Use when exploring codebase structure, discovering module APIs, understanding imports, or checking what functions exist. Provides scripts that replace manual file reading with structured discovery.

Use when working with detail bands — the 1:N child row fan-out feature that produces vertical sub-row expansion under parent rows. Covers the DetailBandSpec contract, band column catalog propagation, per-band SQL generation, JS-side row stitching (separate…

Architectural orientation for the SQL query construction layer. Covers module responsibilities, data flow, invariants, and debugging guidance. Use before modifying any SQL generation code.

End-to-end report pipeline from data ingestion through SQL generation, report execution, result shaping, output layout, and export. Covers layer responsibilities, entrypoints, invariants, and navigation tips for the five-layer architecture.

Core state lifecycle: store, load, serialize, hydrate, apply, validate, and the reactive pub/sub pattern for app state in TableFlip Preact.

Use when working with the Preact UI layer — understanding the app shell composition, card/section/component hierarchy, store subscription pattern, ownership boundaries with core/query/report layers, AG Grid and XLSX vendor integration, export pipeline, or…

UMD GitHub Actions CI reality and Hatchet worker/scheduler deployment pins — run 33164294061 job results, the 14 postgres-job failures (python-multipart/ffmpeg/pg_dump/compose-interpolation, all environment defects with uncommitted working-tree fixes), the…

UMD environment configuration and deployment env-var naming rules — the pydantic-settings nested-name invariant (UMD_POSTGRES__DSN, UMD_OCFL__ROOT, UMD_PROJECTION__*), the alembic os.environ seam, all single/double-underscore references across…