Towards a Science of AI Agent Reliability. Authors define 12 reliability dimensions, benchmark 14 models, and launch an interactive dashboard to measure AI agent reliability. Outcome engineers get a concrete measurement framework and a dashboard to audit agents across reproducible reliability signals — essential for Validation (Principle 16) and for building an immune system around agent behavior.

Hugging Face Agent Skills. Hugging Face publishes a standardized, interoperable “Agent Skills” repository that lets agents perform dataset, training, and evaluation workflows across platforms. This reduces integration friction and makes reusable agent capabilities practical for production pipelines — a key advance for legible tool landscapes and the Graph of reusable artifacts (Principles 06 and 11).

Vouched launches Agent Checkpoint. Vouched ships Agent Checkpoint to add governance, human checkpoints, and auditability to agent workflows. Outcome engineers can embed auditable stop-points and human approvals to meet compliance and operational safety requirements — directly supporting Gate and Law concerns (Principles 15 and 10).

Emdash — Open-source agentic development environment. Emdash runs multiple coding agents in isolated Git worktrees, enabling parallel agent-driven feature development and remote SSH workflows. Teams get a reproducible, developer-first environment to iterate agent-generated code and deliver artifacts safely, putting Teamwork and Artifacts (Principles 03 and 08) into practice.

Ian Webster & Joel de la Garza: Promptfoo on Agent Security. Promptfoo reframes agents as LLMs that act and makes security testing the essential pre-production gate for enterprise agent deployments. Use it to codify tests, fuzz prompts, and enforce safety gates so agents don’t become a supply-chain risk — tooling that operationalizes the Immune System and Gate principles (14 and 15).