lauren

A walkthrough of automating release and QA with two team bots: contributor notifications, a fuzz swarm of 10+ agents testing the build, and an engineer bot that triages, fixes and cherry-picks patches.

Argues that self-verification is the foundation of an agentic loop: agents exercise the real interface and produce videos, screenshots or formal models as reviewable proof, with honest caveats about approximation.

Part 2 of a pstack guide on agent-driven research, planning and architecture. Covers indirect prompts, history-aware context recall, README-driven design, parallel prototyping, and a cross-model architecture arena, all anchored on empirical verification rather than abstract plans.

pstack v0.15.27 adds a beta meta-skill that builds formal verification skills for agents, covering TLA+ modeling and Kani checks for Rust. The author explains the approach and reports bugs found in their own code. Thread collection was partial.

The pstack author explains how 24 concise principle skills work: a mode skill loads an index, agents must name each principle applied and the decision it changed, and users steer with principle names. He also argues much of this will compress as models improve.