Docs/Start

Watch it work

A worker runs unattended, but you can see everything it does: a live dashboard, the agent's transcript for each run, the scheduler log, and what is done per lump, read from git.

On this page
  1. Dashboard
  2. Transcripts
  3. Daemon status and log
  4. What is done
  5. Keep a record per item
You want to know Run
Is the worker up, what is it running right now? lumpcode dashboard or lumpcode daemon-status
What is the agent doing on this lump? lumpcode transcript --lumpName <lumpName>
Why did a pass skip or fail? lumpcode daemon-log
Which items are done, pushed, or left? lumpcode lump-status --lumpName <lumpName>

All of these run on the worker clone. lump-status also works on your laptop, because status comes from git.

Dashboard

terminal
lumpcode dashboard
# Dashboard: http://127.0.0.1:4747

A read-only page for this project's worker: each daemon, the lump and item it is working on, the next pass, and the supervisor that keeps daemons alive. It refreshes every few seconds and warns when its data is stale. Nothing on the page can start or stop anything.

It stays in the foreground until Ctrl+C. Change the port with --port. On a remote worker, keep the default 127.0.0.1 bind and tunnel to it:

terminal
ssh -L 4747:127.0.0.1:4747 my-worker

--bind 0.0.0.0 works but the page has no authentication, and the CLI says so.

Transcripts

Every worker run writes the agent's live output to a transcript file, one per lump per pass. Follow the latest one:

terminal
lumpcode transcript --lumpName backlog          # last 10 lines, then live
lumpcode transcript --lumpName backlog --lines 0 --noFollow   # whole file
lumpcode transcript --list                      # every transcript, newest first

On your laptop, lumpcode run prints the same feed in the terminal and writes no file.

Too noisy? Set outputDetail: "quiet" in lumpVariables (or per step in stepVariables) to keep the agent's messages and drop tool activity. Turn files off with "writeTranscripts": false on the lump, local.json, or project.json. Old transcripts are deleted after 14 days; change that with logRetention in local config.

Daemon status and log

terminal
lumpcode daemon-status            # every daemon for this project, plus the supervisor
lumpcode daemon-status --json     # inFlightRuns, nextTickAt, localConfigStale, …
lumpcode daemon-log               # follow the scheduler log
lumpcode daemon-log --noFollow --lines 100

The daemon log is the scheduler's own story: which lumps it found, which it skipped and why (disabled, too many open branches, a busy workspace), and errors. The agent's output is in the transcript, not here.

localConfigStale: true means you edited local.json or project.json after the worker started. Run lumpcode restart.

What is done

terminal
lumpcode lump-status --lumpName backlog

Each item is toDo, branchPushed (a branch with its marker is on the remote, not merged yet), or finished (merged). This is read from your git remote, so it is the same on every machine. See How a run works.

Keep a record per item

"keepHistory": true on a lump appends every step's prompt and output to .lumpcode/lumps/<lumpName>/history/<contextName>.yaml, on the laptop and the worker. Useful when you tune a prompt across several items. It is gitignored; agent output can contain secrets.

Edit this page