Runs, logs and cost

Every run is kept in Paperclip. You can read runs in the chat, through Claude, or in Paperclip’s dashboard.

In the chat

You ask	Claude runs	You get
“Which jobs do I have?”	list	Each job’s status, repo, schedule, next and last run, and link.
“Show the digest’s last runs.”	runs "<job>" --limit 5	One line per run: when, state, and what started it.
“What did the digest do this morning?”	log "<job>"	The latest run: task, summary, tokens, and the end of its log.
“Show the run before that.”	log "<job>" --run 2	The second latest run.
“Show the whole transcript.”	log "<job>" --full	The run’s full log instead of its last 30 lines.
“Why did the digest fail?”	log "<job>"	The task’s summary says why; Claude explains it.

runs prints one line per run, newest first:

1. 2026-11-03 22:45 America/Toronto  completed  manual

The state is one of Paperclip’s run states, and the last word is what started the run: schedule, manual (run now), api or webhook.

For log’s output, line by line, see Test a job.

In Paperclip

• A job’s page (Routines, then the job): its recent runs, each with its state, time and task.
• A task (Tasks, or a run’s task link): the result the Runner posted, and the task’s status.
• The Runner’s runs (Agents, Runner, Runs): every run, with its status, model, input, output and cached tokens, cost, duration and full transcript.

The transcript shows each step of the run: the command, the skill Claude used, its messages and tool calls, and the result.

Screenshot to comeA run's transcript in PaperclipAgents, Runner, Runs, a run, then its transcript: the git pull, the skill loaded from the repo, Claude's steps and the task marked done.

Tokens and cost

log lists each agent run’s tokens: input, cachedInput and output. Paperclip’s run page shows the same numbers.

Runs on a Claude subscription show no API cost. Look at the tokens: they count toward your plan’s usage. See What it costs for a measured example and how to change the model.

One agent run per job run

A run should take one agent run. When log shows a second one marked “follow-up of”, the first ended with its task still open, and Paperclip started another to close it. Ask Claude to make the skill end by closing its task.