I love walking in San Francisco! To celebrate Walktober, I built an app to explore San Francisco’s many staircases. However, after working on it with my partner, I noticed a new feature: the app was assessing whether those staircases needed escalators.
Zut alors! I did not build that part.
So I used Entire to find the conversation behind it. Here’s the workflow, using agent sessions already captured by Entire.
1. Search for what looks unfamiliar
From my terminal, I ran:
entire search "escalator" --all-repos
Of course, for all Entire CLI commands, you could just tell your agent to do it. No need to remember commands anymore!
--all-repos searches across repositories. Entire searches the recorded conversations alongside the code, so I could find the prompt that led to the feature.
This differs from Codex, Claude Code, etc which search where the code is and if the code is there, but not the agent sessions that led to the change.
One result stood out: “Assess every stairway for escalator retrofit potential.”
There it was! I opened the result and found Patrick’s session, including his original prompt and an attached checkpoint. That gave me somewhere to start digging.
2. Get the explanation behind a checkpoint
Next, I copied a checkpoint ID and ran:
entire checkpoint explain <checkpoint-id>
The explanation walks through the session’s intent, what changed, the files involved, and what the agent learned on its journey.
In this session, the agent had been trying to pick up an elevation fix. The explanation showed that the fix was still in an unmerged PR, so the agent rebased onto that branch and updated the scoring to use cached step measurements.
That’s a lot more useful than trying to reconstruct the whole story from changed files. I could also read the transcript when I wanted the details.
3. Pick up the conversation!
From the checkpoint, I could resume the agent session with its existing context: the prompt, the plan, and the conversation that led to the change.
That’s handy when I’m returning to my own work weeks later, or if I want to pick up something a teammate (or another agent) started.
There’s also a web console if you’d rather browse the prompts, agent responses, checkpoints, and tool calls in a slightly-prettier (my opinion) way.
Mystery solved. Now, back to climbing actual stairs.
Happy Walktober🚶♀️! Let me know your favorite San Francisco staircase in the comments--you can check out the site here.
Top comments (0)