Appearance
Runs
A run is a single execution of a workflow by an agent — it tracks every step, decision, and output from start to finish.
What It Does
When you launch a workflow, Pencel creates a run. The run walks through each step in the workflow, records what happens, and gives you a complete log of everything the agent did. You can watch runs in real time or review them after they finish.
Runs are your primary way to see what agents are actually doing. Every tool call, every decision, every output is captured.
Run Statuses
A run moves through several statuses during its lifecycle.
| Status | Meaning |
|---|---|
| Queued | Waiting to start |
| Running | Agent is actively working |
| Step Running | A specific step is executing |
| Awaiting Approval | Paused — a step needs your review before continuing |
| Awaiting Input | Paused — the agent asked a question and is waiting for your answer |
| Completed | Finished successfully |
| Failed | Encountered an error |
| Paused | Manually paused by you |
| Interrupted | Stopped by you mid-execution |
| Cancelled | Cancelled before completion |
TIP
The most common pause reason is Awaiting Approval. By default, steps that change data or interact with external services require your sign-off. You can configure auto-approval on specific steps if you trust the agent to handle them independently.
Key Properties
| Property | Description |
|---|---|
| Workflow | Which workflow is being executed |
| Agent | Which agent is performing the work |
| Status | Current run status (see table above) |
| Step Cursor | Which step the run is currently on |
| Outputs | Artifacts and results produced by the run |
| Cost | Total token usage and dollar cost |
| Start Time | When the run began |
| End Time | When the run finished (or was stopped) |
| Triggered By | What started the run (manual, schedule, hook, or another run) |
Inspecting a Run
You inspect every detail of a run from the Jobs surface — open a job, then a run, for its full detail.
- Step list — see each step in order, with its status and duration
- Event log — drill into a step to see exactly what the agent did: which tools it called, what arguments it passed, and what results it received
- Outputs — view artifacts created during the run
- Costs — check token usage and dollar cost on the run
INFO
The event log captures everything except raw token data. You get a complete audit trail of the agent's reasoning, tool usage, and decisions without the noise.
How Runs Stay Reliable
Pencel applies the same execution guarantees to every run, whether it was started manually or by a schedule.
- Structured event log. Every tool call, decision, and output is recorded as a typed event. The audit trail is consistent run-to-run, which makes it easy to diff a healthy run against a failing one when something looks off.
- Automatic repair on failure. When a step fails for a recoverable reason — a transient network error, a malformed argument, a missing field — the agent proposes a fix and retries before surfacing the failure. You see the final outcome in the run timeline; the recovery attempts are logged underneath so you can inspect what was tried.
- Verified outputs. Steps that produce structured results — analyses, scoring, document sections — are validated against the expected output shape before the result is passed downstream. A malformed output fails the step instead of silently corrupting the next one.
TIP
To see repair activity, drill into a step in the event log and look for retry entries labeled "auto-repair." They show what was changed before the retry succeeded.
Approval Gates
Some steps pause for your approval before executing. This is Pencel's safety mechanism for actions that change data, send messages, or interact with external services.
When a run reaches an approval gate:
- The run status changes to Awaiting Approval
- You receive a notification (if configured)
- You review what the agent plans to do
- You approve to continue or reject to skip the step
WARNING
Approval gates are on by default for steps that modify data. Think carefully before turning them off — auto-approved steps execute without any human check.
Run Triggers
Runs can start in four ways:
| Trigger | How it works |
|---|---|
| Manual | You click "Run" on a workflow |
| Schedule | A time-based trigger fires (e.g., every Monday at 9 AM) |
| Hook | An event triggers the run (e.g., new Slack message, webhook) |
| Sub-run | Another run spawns this one as a sub-task |