Post-mortem when an agent breaks something.
An agent deleted the wrong folder, ran a migration, or left the build red. Postrun has the whole session, step by step, so the write-up starts from facts.
The problem
When an agent breaks something, the first question is “what exactly did it do?” The chat has scrolled away or been cleared. Shell history mixes the agent’s commands with yours. The final diff shows what changed in the repository, but not the commands it ran outside it, the output it acted on, or the order things happened in.
Without that sequence, a post-mortem is guesswork, and the fix for next time is a guess too.
How Postrun handles it
- STEP 01
The session is already on record
While capture is running, Postrun records every Claude Code and Cline session on your machine, whether or not you expected to need it. The record is in
~/.postrunbefore anyone asks what happened. - STEP 02
Find the session
Sessions are listed newest first, with the agent, the step count and how many steps failed. Open the one that ran when things went wrong.
- STEP 03
Walk the timeline to the first failure
Steps are in the order they happened, grouped into turns, so you can see which prompt led to which action. Failed commands and steps are flagged with their error type and exit code. The first red step is usually where the story starts.
- STEP 04
See exactly what changed
Expand a command to read its full output. Read each edit as a diff, including ones the agent made and then undid. The files list shows everything the agent created, edited or read, which is your checklist for what to inspect or restore.
- STEP 05
Share the record with the team
Export the session to one HTML file and attach it to the incident write-up. Secrets, credentials and home paths are masked first, and every masked value is listed for you to check. People reading it need no account and no install.
What Postrun doesn’t do
Postrun is a recorder, not a guard. It doesn’t block commands, stop an agent or roll back changes. It records sessions only while capture is running, and only for supported agents: Claude Code and Cline today, with Cursor and Codex next. What it gives you is an exact, ordered account of what happened, so you can decide what to undo and what to change.
Sessions stay on the machine that recorded them until you export one. See security for how that is enforced.
Have the record before you need it.
Postrun is in early access for Claude Code and Cline. Open the example report to see exactly what an export looks like, or join the waitlist for the install.