# RossLabs toolkit

> Tools, guides, and research drawn from building AI systems.

## Entries

- [Prompt Decision Aid](https://rosslabs.ai/toolkit/prompt-decision-aid): Compares 14 research-backed or clearly labeled practical techniques with fit, risks, prompt patterns, and representative papers.
- [TruePace](https://rosslabs.ai/toolkit/flodoro): Runs Micro, Pomodoro, Extended, Flow, or Daily focus blocks with pauses, breaks, and scheduled check-ins where supported.
- [Agent Architectures](https://rosslabs.ai/toolkit/research/agentic-ai-architectures): A research reference for choosing an agent architecture, harness, memory system, and production discipline, supported by 78 cited sources.
- [Model Selection Guide](https://rosslabs.ai/toolkit/research/model-selection): Provides dated pricing and a task-based routing starting point that must be validated on your own evaluation set.
- [Agent Memory Patterns](https://rosslabs.ai/toolkit/agent-memory-picker): Maps a memory need to a starting pattern, teaches the correctness condition and failure mode behind it, then links the production and research implementations that match.
- [Design of Experiments](https://rosslabs.ai/toolkit/patterns/design-of-experiments): Explains factors, two-level matrix designs, replication, uncertainty, and confirmation runs for testing several settings together.
- [Build Loop](https://rosslabs.ai/toolkit/build-loop): Walks build-loop as verified on June 24, 2026: the four core phases, the gates between them, and the memory file that carries decisions into the next run.
- [Factory perspectives](https://rosslabs.ai/factory/options/): Turns one sentence into an answer through playful factory views, guided labs, and inspectable retrieval steps.
- [Graph Engineering](https://rosslabs.ai/toolkit/research/graph-engineering): Separates knowledge, workflow, reasoning, and state-memory graphs, then maps each to starter schemas, invariants, queries, failure modes, and storage choices.
- [Knowledge Graph Retrieval](https://rosslabs.ai/toolkit/patterns/knowledge-graphs): Separates index-time graph construction from query-time retrieval and shows how four RossLabs projects implement each stage.
- [Agentic Stack Builder](https://rosslabs.ai/toolkit/agentic-stack-builder): Combines a framework, coordination pattern, memory approach, and harness controls into a starting architecture with compatibility notes and failure risks.
- [Agent Anatomy](https://rosslabs.ai/toolkit/agent-anatomy): Shows the core design decisions in an agent and how several frameworks express each decision in code, configuration, or Markdown.
- [Multi-Agent Coordination](https://rosslabs.ai/toolkit/patterns/multi-agent-coordination): Walks the nine commands two agents run to share one checkout without overwriting each other, and shows where the automatic check stops reaching.
- [Tyrone's AI operating stack](https://rosslabs.ai/toolkit/personal-stack): Walks the layered inventory of coding agents, tools, orchestration, memory, and infrastructure behind this site — the connections between them, and the trigger that sends work to each.
- [Vector Search Stack](https://rosslabs.ai/toolkit/vector-search-stack): Flip the levers at each stage of a real vector search stack and see recall, precision, cost, and latency trade off.
- [Rally Point Lab](https://rosslabs.ai/toolkit/patterns/multi-agent-coordination/play): Guided Story: nine short exchanges show how presence, claims, refusals, handoffs, receipts, liveness, and durable history change what happens when two agents share one repo. Scenario Runner: pick a question, set 2 to 20 agents, files, layout, handoffs, injection, and rallyd, and watch one deterministic engine append the ledger and project the room.
- [Verification-Gated Agent Development](https://rosslabs.ai/toolkit/patterns/verification-gated-agents): Explains a synthesized five-step pattern that separates execution, deterministic checks, independent review, and lessons for the next run.
- [LLM Wiki Builder](https://rosslabs.ai/toolkit/llm-wiki-builder): Turns five choices about outcome, audience, capture, sources, and complexity into a blueprint, manifest, and build prompt you download locally; it uploads nothing and connects no tools.

HTML version: https://rosslabs.ai/toolkit/