What did my agent actually do? Reading an agent's run log

Free options first

1. Ask the agent at the end of each task: "List what you did as one line per action, with the result." Paste the lines into a shared doc.
2. Keep that doc as the agent's log and point the next session at it.

This works for one agent and a few tasks. It stops working when the agent runs on a schedule, when you have more than one, or when the agent's own summary is the only record of what it did.

What a readable run log looks like

One line per finished run, written by the agent itself when the run ends: a name for the run, what it tried, the outcome (ok, failed, or still open), how long it took, and a link to anything it produced. Runs that started and never finished are shown as stale instead of vanishing.

Example (made-up runs):

Reading it takes a minute: two things ran fine, one needs a sign-in fixed, one never completed and should be re-run.

Set it up for any assistant that can add a custom connector

1. Sign in to Memoshelf with your email (free plan, no card).
2. Add Memoshelf to your assistant as a custom connector at https://memoshelf.com/mcp and sign in when asked; pick a name for that agent. That name is its lane.
3. Tell the agent, in its rules or at the start of a task: "When a run finishes, log it with log_run: name, what you tried, result, duration. Before starting, read what_happened_last for your lane."
4. Open the account page. The run log shows every agent's runs by lane, with stale runs flagged, and you can export it as JSON or markdown.

Try it

Free tier: three agents, a run log and notes, no card.