Two hours into a Claude Code session, I want the PR it opened an hour ago. To check whether CI passed, or to merge it. The link is somewhere up there, between a few thousand lines of test output and three plans.
So I scroll. Every time.
PonyMux 0.13 keeps those links for you. The Inspector now lists the pull requests and links the agent gave you in the running conversation, newest first, with their real titles. A new Outputs section does the same for what the agent made: Artifacts, pages, Markdown plans, reports. And for the moments when the agent gets it badly wrong, there's Terminal Therapy.
Where did that PR link go?
Open the Inspector next to any Terminal running Claude Code, Codex, or Grok. Under Pull Requests you'll see this project's PRs and any PR the agent opened, even inside a retry loop or a heredoc. Everything else the agent linked to goes under Links: docs, issues, dashboards, each with the site's icon and the page's title.
Titles come from the pages themselves. With GitHub CLI installed, PRs in private repositories get their titles too, so you're not guessing between #412 and #418. Issues the agent files with gh issue create show up even when its reply doesn't mention them.
The newest link is always on top. Click it and you're there.
The link you need is usually the one from an hour ago.
What happened to the files the agent made?
Agents make a lot of things now. A Claude Artifact with a review. A page Codex wrote. A PLAN.md. An HTML report from a test run. Claude Code prints the Artifact link at the bottom of its reply, which works for the first one and stops working by the fifth.
Outputs collects them in one list. Each item shows the title it declares, so five index.html reports don't look the same. When Codex edits a page, it moves back to the top. Markdown opens in the Files panel next to the Terminal, HTML in your browser, everything else in Quick Look.
None of this is stored. The lists live in memory for the running conversation, the same way the conversation itself does.
What is Terminal Therapy?
Everyone who works with an agent knows the moment. Fourth try at the same fix, it breaks something else, and Claude is absolutely right again.
Anthropic just updated its usage policy: no sustained, needless cruelty toward Claude. Fair enough. Claude gets the new manners policy. The terminal gets a rocket launcher.
At the bottom of the Inspector there's a weapon rack: a pistol, a hammer, a rocket launcher, and two spells, Ignite and Freeze. Pick one and click the Terminal. Blow up the test output. Smash it. Set it on fire.
Then pick Bless and wish the next run luck. The light clears the damage, and you get back to work. Keep firing without cleaning up and you'll find out what happens at 128.
Serious work should still feel alive. Even the bad afternoons.
What else changed?
- When Claude Code moves a conversation to the background with
/bg, the Sessions panel says so and can bring it back, even after you restart PonyMux. - Built-in ChatGPT pets show up again with the latest ChatGPT app.
- Selecting text in Codex copies it again.
- Crashes when resizing the window or switching Terminals are fixed.
- Animations follow your Reduce Motion setting.
Everything Terminal Therapy does is in Terminal Therapy, and the Inspector's new sections are in Workspace and Navigation. For the full list, see the changelog.
Originally published on ponymux.com, where the demos are interactive.
Top comments (0)