Skip to content

Runs

A run is one execution of a pipeline for one issue. Runs are started by triggers or by hand, queue up behind Max concurrent runs, and move through the statuses queued → preparing → running (↔ awaiting approval) → succeeded / failed / cancelled.

Runs in the sidebar, or a project’s Runs tab for just that project’s runs. The table shows every run, newest first:

Column Content
Run #number
Task Issue key and title (or the pipeline name), with the pipeline, branch and any links found below
Status Status pill
Progress One dot per step, coloured by step status
Source What started it: manual, linear, jira, github, schedule or webhook
Duration Elapsed time (live while running)
Cost Total reported cost, or —
Started How long ago

Filters across the top:

  • All, Active (queued, preparing, running), Needs approval, Succeeded, Failed (failed and cancelled), with counts;
  • All projects / No project / one project;
  • All pipelines / one pipeline;
  • a search box (Search issue, branch, #number) matching the issue key, title, pipeline name, run number and branch.

Click a row to open the run.

New run (top of the sidebar, Dashboard, Runs page, ⌘N, the ⌘K palette, the tray menu, or Run on a pipeline) opens Start a run:

Field Notes
Pipeline Lists every pipeline with its step count and project.
Task title Required. Becomes {{issue.title}}.
Issue key Optional, e.g. ENG-123. Used for branch names. If empty, TASK-<n> is used.
Issue URL Optional link back to the tracker.
Description Becomes {{issue.description}}.
vars.<name> One field per pipeline (and project) variable, with its default as the placeholder. Filled-in values override the defaults for this run only.

Start run (or ⌘↵) validates the pipeline and its workspace first. If something’s missing you get an error such as Can't start: Workspace: choose the local git repository, … and no run is created.

  • #number, the issue key and title (or pipeline name), the status pill, and the run’s error in red if it failed.
  • Issue ↗: opens the issue in the tracker (when the run has an issue URL).
  • One button per link found in step outputs, e.g. PR #57 or Preview. Jimothy recognises GitHub pull request URLs, GitLab merge requests, Bitbucket pull requests and *.vercel.app previews.
  • Open workspace in editor (</> icon): tries code, then cursor, and falls back to opening the folder in your file manager.
  • Reveal workspace folder (folder icon): opens the workspace in your file manager.
  • Cancel while the run is active; Retry and Delete run once it has finished.

The workspace buttons are hidden when you use Jimothy from another device through Remote access.

Pipeline (link to the editor), Trigger (what started it), Branch (with a copy button), Started, Duration, Cost and Tokens (12.4k in · 3.1k out; input includes cache reads). Graph collapses or expands the graph.

The pipeline graph shows every step as a node with its type, status icon, harness and model (or the shell command, or Human approval), and duration. Edges light up as work flows; feedback loops are drawn as amber arcs under the graph, and a node shows ↺ with the loop count once it has been re-run. Click a node to select that step.

Each step waiting for approval shows a banner above the step list, with the files the steps before it made. See Approvals.

The left column lists Setup & run log (workspace preparation, git operations and the run summary) followed by every step with its status, harness and model (or shell / approval), attempt count, loop count and duration.

When you open a run, Jimothy selects the step that most needs attention: the one awaiting approval, else the running one, else the failed one, else the last succeeded one.

The step header shows the status, name, type, harness, model and cost. When a step has more than one attempt, a dropdown switches between them (Attempt 2 (loop 1) · succeeded). Failed steps show their error in a red box.

Tabs:

Tab Content
Logs The attempt’s log, streaming live (see below). Files the step kept show as thumbnails under the Kept N artifacts line.
Data What the attempt got, next to what it produced. Input is what the step actually ran after template rendering (the prompt, command or approval message, plus the values it used); before the step runs, the template is shown instead. Output is its final output: Markdown for agent steps, plain text for shell steps, the comment for approvals. Long inputs and outputs show their first or last 20,000 characters, with a button to load the rest. Edit with this data opens the step in the editor with this run’s values.
Artifacts Agent and shell steps: the documents and images the attempt added or changed, with a preview and Download. See Artifacts.
Details Duration, tokens, cost, the harness session id (Claude Code, Codex, Pi), the approval decision (who, when, comment), the attempt’s artifacts, and the attempts table (#, loop, status, started, duration, exit code, artifacts). Click an attempt to open its logs.

The log viewer streams each line with a timestamp and an icon for its kind:

Icon Kind Examples
· system $ claude -p …, Retry 1/1, Feedback loop iteration 1
◆ assistant the agent’s messages
∴ thinking reasoning blocks
⚙ tool Bash: npm test, Edit: src/api.ts, $ pytest
↳ tool result command output, file contents (clipped)
✓ result success · 12 turns · $0.4213 · 81234 in / 2210 out tokens
› stdout plain-text harness output, shell output
! stderr stderr lines
✗ error failures, tool errors, verdict failures

Toolbar: Filter logs (text search), a Thinking checkbox to hide reasoning, a live badge while the attempt runs, the line count, Copy logs, and Follow output (auto-scroll; scrolling up pauses it, scrolling to the bottom resumes).

What you get depends on the harness’s output format: Claude Code, Codex and Pi show structured tool calls and thinking; plain-text harnesses show raw stdout. See Output formats.

Cancel on an active run:

  • a queued run is cancelled immediately;
  • running steps are stopped by killing the whole process tree (SIGTERM, then SIGKILL after 5 seconds), so agents and anything they spawned stop;
  • approval gates stop waiting;
  • remaining steps are marked cancelled and the run ends cancelled.

The workspace and branch are left as they were. A cancelled triggered run counts as a failure for write-back.

Once a run has finished (succeeded, failed or cancelled):

Button Re-runs
Retry (header) Every step that didn’t succeed, plus everything downstream of it. Succeeded steps keep their output.
Retry from here (step header) The selected step and everything downstream of it, even if they succeeded.

Retries reuse the run’s workspace and branch if the folder still exists, so work done so far is kept. They use the pipeline’s current definition, so you can fix a prompt or command in the editor and retry the same run. Changes you made to a step on the run page are kept on top of it (see Fix a run in place). Steps you added since get a fresh entry; history is kept for existing steps. Feedback-loop counters start again from zero.

You can change or skip a step from the run page without opening the pipeline editor. Both buttons sit in the step header, next to Retry from here.

Edit opens Edit step for the selected step. What you can change depends on the step type:

Step type Fields
Agent Name, Harness, Model, Prompt, System prompt, Timeout (minutes), Retries
Shell Name, Command, Run in, Timeout (minutes), Retries
Approval Name, Message shown to the approver
Pipeline Name, Timeout (minutes), Retries

You can’t change a step’s id, type, dependencies or feedback loops here, because that would change the order the run works through. Condition steps have no Edit button. Make those changes in the pipeline editor, then Retry.

The change applies to this run only:

  • While the run is active, Save applies the change from the step’s next attempt, loop or retry. An attempt that is already running carries on as it started.
  • Once the run has finished, Save keeps the change for the next retry, and Save and retry from here saves it and re-runs this step and everything after it.

Tick Also save to the pipeline, so new runs get it to copy the same change into the pipeline too. The checkbox only appears while the run’s pipeline still exists.

Jimothy checks the edited step before saving and rejects invalid changes, such as a harness that isn’t configured. The run log records each change, e.g. Step "Run tests" changed for this run: command.

Run-only changes survive retries: a retry reloads the pipeline’s current definition, then applies your run-page changes on top of it.

Button When it shows What it does
Skip The run is active and the step hasn’t finished Skips the step when its turn comes.
Don’t skip The step is marked to be skipped Undoes Skip; the step runs normally.
Skip and continue The run has finished and this step failed or was cancelled Skips the step and resumes the run.

A skipped step shows as skipped with Skipped by hand. Its output is empty, and the steps after it run as if it had passed. The step stays skipped on later retries until you click Don’t skip.

Approval and condition steps can’t be skipped: approve or reject an approval, and let a condition decide which steps run.

Delete run removes the run and its logs. You can only delete a finished run (Cancel the run before deleting it). The workspace folder is kept; remove it yourself or enable Delete a run’s workspace after it succeeds in Settings.

Each run has its own folder, runs/<run-id>/, in the data folder: run.json, one JSONL log per step attempt, and a folder per attempt with its full output and artifacts. See Data & security.