AI Engineering

LeanCTX compresses LLM context by up to 99% using 79 MCP tools, 10 read modes, 95+ shell compression patterns, and tree-sitter AST parsing for 26 languages. Use it when reading files, running shell commands, searching code, or exploring directories.

APM Issue Autopilot drives any open microsoft/apm issue from raw intake to a mergeable PR with triage as the central gate. It batches issues, runs the triage panel per issue, presents one consolidated review, escalates by default, and only auto-implements clear, bounded accepts approved by the maintainer.

APM Starter is a starter package for Agent Package Manager that demonstrates how to create reusable agent primitives like instructions, prompts, and agents. Use it to bootstrap new APM packages or learn the structure.

APM Triage Panel orchestrates multiple specialist personas to triage a single GitHub issue in the microsoft/apm repository, producing a synthesized comment with a triage decision, label set, milestone, and suggested next action.

Docs Impact Classifier classifies the documentation impact of a pull request diff into no change, in-place edit, or structural change. It uses deterministic path gates, symbol extraction, and a bounded LLM call to keep docs-sync costs low when no docs work is needed.

Shepherd Driver drives one existing pull request to a mergeable state: classifies review, runs the review panel, folds recommendations, pushes, watches CI, and resolves conflicts under caps. Use it as the composed per-PR stage in a batch bug shepherd or issue autopilot orchestrator, not standalone.

PerfUp autonomously researches, benchmarks, and implements performance optimizations for vllm-mlx, then ships production-ready pull requests when improvements are confirmed.

Add Domain extends an existing knowledge system by deriving a new domain configuration through conversation, then generating folders, templates, and vocabulary while connecting to the current architecture. Use it when adding a new subject area to a multi-domain note system.

Ars Contexta Setup scaffolds a complete knowledge system by detecting the platform, asking a few questions, deriving a configuration, and generating files. Use it to create a structured vault with processing pipeline, automation hooks, and navigation maps for an AI agent.

Ask queries the bundled research knowledge graph to answer methodology questions, routing through WHY, HOW, and WHAT IT LOOKS LIKE tiers plus reference documents. Use it when you need research-backed guidance on knowledge system design, workflows, or domain examples.

Contextual Help provides contextual guidance and command discovery for the Ars Contexta knowledge system. It shows available commands, active skills, and intelligent suggestions in narrative, contextual, or compact modes based on vault state.

Knowledge System Architect analyzes health reports, friction patterns, and derivation history to propose specific changes with research justification. It never auto-implements; recommendations require explicit approval.