I built a beta tool for turning shell and AI agent sessions into reusable context

I’m shipping the first beta of Visr today and would love feedback from people building with AI agents. The basic idea: capture shell + agent sessions, then turn what happened into transcripts, runbooks, skills, and evals so useful context doesn’t disappear when the terminal session ends. The pain we’re trying to solve is that a lot of agent work is currently trapped in ephemeral sessions: commands, outputs, failures, fixes, project conventions, course-corrections, traces, and the small bits of context that make the next run better. It’s early and intentionally simple right now. Curious how other agent builders handle this today: \- Do you save useful context from agent runs anywhere? \- What context is worth preserving vs. noise? \- Would transcripts, runbooks, skills, or evals be the most useful output? \- What would make this fit your workflow instead of becoming another dashboard? Please check the comments for changelog/demo of the beta.