Traction scoreGitHub stars can be faked, so popularity alone can be misleading. Traction Score looks for broader signs of recent attention, adoption, and active maintenance.
This collection provides a structured framework for context engineering, harness engineering, and the design of multi-agent architectures. It is designed for builders who need to optimize how agents handle information and maintain reliable execution patterns.
The repository offers tools for building, evaluating, and debugging agent systems, ensuring that context management remains effective as system complexity increases.
WHO IT'S FOR
AI agent / automation builders
designing multi-agent architectures and context management
LLM engineers
building book-to-SFT pipelines for voice replication
AI system debuggers
analyzing reasoning traces and context degradation
AI agent power users
implementing personal operating-system workflows for productivity
Compatible AgentsThe repository documents support for these agents. The skills may also work with other agents that can load SKILL.md files, but they may need some setup or small changes.
Installs: Direct manual reference to the standalone book-sft-pipeline skill documentation; this does not copy it into a skill directory. · Agents without Agent Skills directory-layout support
Before you start
An agent that does not support the Agent Skills directory layout
The evaluator, rubric, or outcome signal currently available
Any actions that must require human approval
Use the harness-engineering skill to design a minimal harness for [describe the autonomous workflow]. Classify its surfaces as locked, editable, append-only, or human-controlled, then define one fast, unambiguous feedback loop and the durable state needed to recover after interruption. Keep irreversible actions under explicit human approval and produce only the design—not an implementation or PR.
A description of the orchestrator-worker architecture
Runtime access details for worker KV tensors
The orchestrator and worker model families
Current token-cost and quality observations, if available
Use the latent-briefing skill to assess [describe the orchestrator-worker system]. Determine whether the runtime exposes and can rewrite worker KV state, whether the agents use a compatible model space, and whether repeated orchestrator-state replay is the main token cost. Recommend either a small latent-briefing experiment or a structured text handoff, and list the quality and token measurements the experiment should track.
Path to the book or extracted text, replacing [book or extracted-text path]
The author or target style
The intended fine-tuning platform or dataset format, if already chosen
Use the book-sft-pipeline skill to propose a small dataset-construction plan for [book or extracted-text path]. Explain how to segment the text into semantically coherent 150–400-word chunks at natural boundaries, separate style learning from plot memorization, and vary instructions and system prompts. Provide the plan and a few illustrative record outlines only; do not start training.
Claude
Codex
Cursor
Gemini CLI
Copilot