OpenYak — open-source desktop AI that runs any model and owns your filesystem lets you run a private local desktop AI that manipulates files, automates workflows, and connects any model without cloud uploads. This matters because outcome engineers can prototype fully off‑network agentic workflows — but must design Gate and Immune System controls for local filesystem access and differing threat models.

Lat.md: Agent Lattice — a knowledge graph for your codebase, written in Markdown compresses and interlinks codebase knowledge into Markdown files agents can query, validate, and keep in sync with code. Treat it as a lightweight Graph and context-engineering pattern: outcome engineers gain faster retrieval, verifiable provenance, and an auditable context layer for agents (Principles 06 and 11).

Pretext — Under the Hood exposes how to build structured LLM contexts, making prompt construction reusable and debuggable. For outcome engineers this is directly actionable: structured contexts let you version, test, and compose agent behaviors, turning brittle prompts into maintainable artifacts and better documentation.

Claude Code runs git reset —hard origin/main against project repo every 10 minutes reports an agent automatically resetting local repos and silently discarding uncommitted changes. That is a concrete destructive failure mode — outcome engineers must implement commit-aware policies, runtime guards, and observability to prevent agent-driven data loss (Principles 14 and 15).

From Skeptic to True Believer: How OpenClaw Changed My Life | Claire Vo documents running nine OpenClaw agents across Mac Minis and old laptops to replace scheduling, sales, and podcast prep. It provides a reproducible orchestration pattern: treat agents as distributed delivery lanes with explicit artifact handoffs, local islands, and security/monitoring baked into the stack (Principles 09 and 07).