{"slug": "show-hn-agents-ship-diffs-visr-keeps-trajectories", "title": "Show HN: Agents ship diffs. Visr keeps trajectories", "summary": "Visr, a new open-source tool, transcribes the entire trajectory of terminal sessions—including every command, output, retry, and fix—rather than just the final diff. The tool, installable via Homebrew, generates a transcript link that can be shared with future agents or teammates, preserving the full problem-solving process and dead ends that a standard diff discards. This trajectory data serves as the raw material for downstream artifacts like CLAUDE.md files, MCP servers, and evals, enabling agents to learn from past work without repeating mistakes.", "body_md": "[Changelog](/changelog)\n\n# Your agents ship diffs. Visr keeps trajectories.\n\nDiffs tell you what changed. Trajectories tell the next agent how.\n\n$ brew install sourishkrout/visr/visr\n\n$ visr shell\n\n## Keep the Trajectory, Not Just the Outcome\n\n`visr shell`\n\nstarts a transcribed shell in your own terminal. Run Codex, Claude Code, Cursor, or plain commands inside it. Visr transcribes the whole trajectory: every command, output, retry, and fix. It hands you a transcript link when you're done. A goldmine for future agents: every dead end already ruled out, so they don't walk back in.\n\nThat's baseline memory for everything downstream: the raw material behind a `CLAUDE.md`\n\n, an `AGENTS.md`\n\n, an MCP server, a skill, evals.\n\n## From Trajectories to Knowledge Graph\n\nA transcript is more than a log. The retries, the dead ends, the fix that finally worked: that's the tribal know-how a diff throws away. And because one `visr shell`\n\ncan span many agent sessions, the transcripts connect: one session to the next, one fix to the next run, trajectories linking into a graph.\n\nIn v0.2.0, the transcript link shows up in your terminal status bar as soon as the session starts transcribing. When the shell exits, Visr prints a short summary and drops the transcript link on its own line, easy to copy, open, and share.\n\nYou run your agents, inspect output, fix bugs, retry a command, and exit with a transcript of the whole trajectory, not just the final outcome. From there it can seed a follow-up review, a skill, a runbook, an eval.\n\n## Your Next Agent's Trajectory\n\nTry it on one real task. Run `visr shell`\n\n, work the way you normally would, and exit when you're done. Visr hands you the transcript link, and that transcript is the artifact worth carrying forward.\n\nShare it with the next agent, the next teammate, the next workflow, so they don't have to guess what happened from the final diff alone. That's where this goes next, the same path the v0.1.0 changelog laid out: [Agents run on terminal memories. Bottle yours.](/changelog/bottle-terminal-memories)\n\n$ brew install sourishkrout/visr/visr\n\n$ visr shell\n\nGive it a try, and let us know what breaks. Thanks!\n\n### Previously\n\n[v0.1.0](/changelog/bottle-terminal-memories)bottled terminal memories.", "url": "https://wpnews.pro/news/show-hn-agents-ship-diffs-visr-keeps-trajectories", "canonical_source": "https://visr.dev/changelog/keep-the-trajectory", "published_at": "2026-06-03 19:36:56+00:00", "updated_at": "2026-06-03 19:50:06.290028+00:00", "lang": "en", "topics": ["ai-agents", "ai-tools", "ai-infrastructure", "artificial-intelligence", "mlops"], "entities": ["Visr", "Codex", "Claude Code", "Cursor", "CLAUDE.md", "AGENTS.md", "MCP"], "alternates": {"html": "https://wpnews.pro/news/show-hn-agents-ship-diffs-visr-keeps-trajectories", "markdown": "https://wpnews.pro/news/show-hn-agents-ship-diffs-visr-keeps-trajectories.md", "text": "https://wpnews.pro/news/show-hn-agents-ship-diffs-visr-keeps-trajectories.txt", "jsonld": "https://wpnews.pro/news/show-hn-agents-ship-diffs-visr-keeps-trajectories.jsonld"}}