Skip to content

Monitoring Jobs

Every time a workflow executes — manually, on a schedule, or triggered by a hook — Pencel creates a run. The Jobs surface (/jobs) gives you full visibility into what your agents are doing, what they've done, and what they need from you. This guide walks you through tracking and inspecting runs.

The Jobs page

Open Jobs from the sidebar. It lists your jobs — each with its trigger, latest run, and a trust bar summarizing recent runs as clean / repaired / failed. Open a job to see its full run history, most recent at the top.

Each run entry shows:

  • Workflow name — Which workflow was executed.
  • Status — The current state of the run (see status reference below).
  • Started at — When the run began.
  • Duration — How long the run has been running or how long it took to complete.
  • Agent — Which agent handled the run.

A job's trust rollup answers "can I rely on this unattended?" at a glance — the recent clean/repaired/failed mix plus the current clean streak.

Understanding run statuses

Runs move through a series of statuses as they progress. Here is what each one means:

StatusWhat it meansWhat you should do
QueuedThe run is waiting to start. Another run may be in progress, or the system is preparing.Wait. It will start shortly.
RunningThe workflow is actively executing. The agent is working through its instructions.Watch the progress if you want, or let it run.
Step RunningA specific step within the workflow is currently executing.No action needed — this is normal progress.
Awaiting ApprovalThe agent has completed a step and is waiting for your approval before proceeding.Review the step and approve or reject it.
Awaiting InputThe agent has a question and needs your answer to continue.Read the question and provide your response.
CompletedThe run finished successfully. All steps were executed.Review the output and artifacts.
FailedSomething went wrong during execution.Open the run to see the error details.
PausedThe run has been temporarily paused.Resume it when you are ready.
InterruptedThe run was stopped mid-execution, either by you or by the system.Decide whether to restart or discard the run.
CancelledThe run was cancelled before completion.No action needed. Review if you want to understand why.

TIP

The statuses you interact with most are Awaiting Approval and Awaiting Input — both need your attention before the run can continue. These also surface in the Active scope of Pulse, so you can clear them there without opening Jobs.

Inspecting a run

Click any run to open its detail view on the Jobs page. You see:

Step list

Each step the workflow defined is listed in order. For each step you see:

  • Step name — What the step is called.
  • Status — Whether it is pending, running, completed, failed, awaiting approval, or skipped.
  • Duration — How long the step took.
  • Auto-approve — Whether the step was configured to run without your approval.

Click an individual step to drill into its details.

Event log

Every step has an event log that records exactly what happened during execution:

  • Tool calls — Which tools the agent called, with what inputs, and what results came back.
  • Reasoning — The agent's thought process at each decision point.
  • Outputs — Any artifacts, messages, or data the step produced.
  • Errors — If something went wrong, the full error details appear here.

The event log is your primary tool for understanding agent behavior. When a run produces an unexpected result, the event log tells you why. Watch a run live to see each action as it happens, before the run completes.

INFO

Event logs persist permanently. You can inspect runs from weeks or months ago to understand historical decisions, audit agent behavior, or debug recurring issues.

Cost

Each run tracks its token usage and associated cost, shown on the run detail:

  • Total tokens — Consumed across all steps.
  • Input tokens — Context (instructions, memories, guidelines, tool results).
  • Output tokens — What the agent generated (responses, plans, tool calls).
  • Estimated cost — The approximate dollar cost based on the model's pricing.

For account-wide spend across all runs, see the Cost & Usage section in Settings.

Handling approval requests

When a run reaches Awaiting Approval, take action from the run detail (or the Active scope in Pulse):

  1. Open the run.
  2. Find the step waiting for approval.
  3. Review what the agent plans to do — you see the proposed actions and reasoning.
  4. Choose Approve (proceed), Reject (skip or stop, per configuration), or Modify (edit the plan before approving).

WARNING

Approval requests hold up the entire run until you respond. If you have scheduled workflows with approval enabled, check Pulse's Active scope or the Jobs page regularly to avoid a backlog. Pencel can also pause a job after repeated failures and flags silent (overdue) jobs that quietly stopped firing.

Responding to agent questions

When a run reaches Awaiting Input, the agent needs information from you:

  1. Open the run.
  2. Read the agent's question in the step detail.
  3. Type your response and submit.

The agent incorporates your answer and resumes. Common questions: clarifying ambiguous instructions, choosing between approaches, or missing information the agent couldn't find in the workspace.

Tips for effective monitoring

  • Clear Active first. Handle Awaiting Approval / Awaiting Input in Pulse's Active scope daily — they block runs.
  • Watch the trust bar. A job trending from clean to repaired/failed is a signal to revisit its workflow or connections.
  • Use per-run cost to optimize. If a workflow costs more than expected, look at the per-step token breakdown to find where the agent uses the most context.
  • Investigate failures promptly. A failed run often means a broken connection, an expired API key, or an instruction the agent can't follow. Fix the root cause to prevent repeats.
  • Compare runs over time. If a weekly workflow suddenly produces different results, compare recent runs to older ones — the event logs reveal what changed.

Next steps