PrumoGrid

HomeLearnFeaturesHistory and writing

FEATURES / 18

History and writing

Where to look up what you sent to the agents and what PrumoGrid observed and wrote, how to clear the list without deleting anything, and what to gather when reporting a problem.

Updated for PrumoGrid 0.5.11

What it's for

After a day with several agents, you will want to know what you asked Codex, what time the analysis finished or what was written to CLAUDE.md. PrumoGrid keeps these answers in different places, each for one question:

QuestionWhere to look
What I sent to each terminalSent instructions
What happened, in order, with timesLOG: what happened, in the Stages tab
What was written to the agents' rule filesSee what was written, on a rules-applied line
What was typed to finish a stage, and the evidence observedSee the completion
Stages from terminals that were already closedView stages from closed sessions and the Results filter

When to use it and when not to

Use the history to reconstruct what happened: picking up the next day, understanding why a stage did not move forward, showing someone else what was asked and what was observed.

Don't use the history, or the terminals' scrollback, as a backup. It keeps facts and receipts within limits; the project's files live in Git and on disk. To share evidence, prefer Copy result on a stage, the text log or a reviewed screenshot.

Step by step

The example uses loja-api after the PRIMEIRA coupon fix, with three stages: analysis, fix and review.

Find a sent instruction

  1. In the Stages tab, look at Sent instructions · N and click View all. The same dialog opens from Tools → Sent instructions…, from the palette or with Ctrl+Shift+I.
  2. Filter by Project and by Terminal, for example the Codex terminal.
  3. In Search sent text, type ANALISE.md. The fix stage's instruction appears. Because the stage had a role, the text starts with "Your role for this stage is: implementer…", which is what PrumoGrid adds before your request.
  4. Click Go to position to see the instruction in the terminal, at the point where it was delivered, or Copy to take the text.
  5. If the list is long, Load more brings in older ones.

Only instructions PrumoGrid delivered, such as stage instructions, appear here. Commands you type by hand in PowerShell or CMD don't count. If the terminal was closed, the text stays in the list; the dialog tells you when its position is no longer available in the conversation.

Read what happened

  1. With the project selected, open the Stages tab and expand LOG: what happened. The title shows how many records there are.
  2. At the top, a summary counts the stages in the plan, how many have a criterion, how many were approved by a command and how many became done from the agent's notice.
  3. Below, read the timestamped lines, in order. For example: Codex CLI finished its turn: stage "Read ANALISE.md…" signaled, followed by check "stage completion": passed (exit 0). Waits also show up, such as a delivery waiting for you to answer a question on the agent's screen.
  4. Copy the log copies the list as text, to attach to a report.
  5. On a stage, use See the completion to see what was typed into the terminal, when, and the completion evidence observed.
  6. In the Flow, the stage's tracking view shows the same history from the stage's point of view.

Read the log before deciding what happened. "The agent said it finished" and "the verification approved" are different lines, and only the second one went through a command.

See what was written to the agents' files

In the log, lines such as rules written to 1 file carry the See what was written button. It opens the receipt of what was written to each agent's rule file: the paths, the files, the copies and the failures. It is not a recording of the terminal. It is also different from Writing in projects, which is the permission, folder by folder, for PrumoGrid to write the guidance block into CLAUDE.md or AGENTS.md (see Rules).

Clear previous records

Over time the list grows. To start the day with it short:

  1. Select a project. In All projects, the control asks you to pick a project first.
  2. Click Clear previous records. The explanation says the clearing applies to that project's previous records, in Flow and Stages, and that stages and pending items are kept.
  3. Wait for the notice Previous records hidden in this project. If saving fails, the notice says the list was kept, and it does not show as cleared.

Clearing hides. It does not delete stages, does not erase the evidence used to decide completions, does not end sessions and does not remove the original history files. The cutoff still applies when you reopen the app, and new events show up again. In this version there is no button to show what was hidden.

Stages from closed sessions

When a terminal is closed, its stages leave the default list and the active Flow. To read them, use View stages from closed sessions (N) and, in the Stages tab, the Results filter. Hide stages from closed sessions goes back to the normal list. A temporary disconnection from the terminal service does not count as closing.

What you see on screen

The Stages tab: the sent instructions, the timestamped log and the stages from closed sessions.

How much is kept

WhatHow much
Log eventsUp to 2,000 events. Token totals are kept in a separate collection and are not dropped with them.
Sent instructionsUp to 10,000 instructions. Those beyond that drop out, and the list tells you how many older records fell outside retention.
Each terminal's scrollbackScrollback lines per terminal, in Tools → Advanced options… › Terminals: from 1,000 to 50,000 lines, default 5,000. It applies to terminals opened after the change.
Agent conversationsThey stay in the CLI itself. PrumoGrid keeps the link to resume them, not a copy of the messages.

The terminal scrollback is there for you to scroll up and read what went by. It is not a backup, and what goes past the line limit does not come back.

The service log

The terminal service writes a log at %TEMP%\terminal-grid-daemon.log. Each file is capped at 50 MB, and at most three are kept: terminal-grid-daemon.log, terminal-grid-daemon.log.1 and terminal-grid-daemon.log.2. When the cap is passed, the oldest is discarded. To open the folder from PowerShell:

powershell
explorer $env:TEMP

How to report a problem

Gather what makes it reproducible:

  • The action you took and what you expected.
  • The project name, the shell and the agent, and whether it was Windows or WSL.
  • The exact message, copied from the screen, and the time.
  • If a stage is involved, Copy result on the stage and Copy the log in LOG: what happened, with the lines from that time.
  • If the terminal service is involved, the part of terminal-grid-daemon.log around that time.

Review screenshots, terminal excerpts and the log before sending them. They may contain paths, client names or keys that should not leave your machine. For other symptoms, see Troubleshooting.

Tips

  • Before repeating a stage that seems not to have started, look for the instruction in Sent instructions. If it was delivered, resume the existing stage instead of creating another.
  • Clear the records at the start of a new task in the project. That makes it easier to read only what happened from then on.
  • If you need to keep a long terminal output, save it to a file (npm test > saida.txt) instead of relying on the scrollback.

Known limitations

  • The log keeps up to 2,000 events; sent instructions, up to 10,000.
  • Clear previous records works per project, and this version has no way to show the hidden records again.
  • See what was written is a file receipt, not a screen recording.
  • History and scrollback do not replace a backup of the project's files.