Working with program activities
Activities are where a program's work happens. This guide covers configuring an activity, reading its state, and every action the activity inspector offers. For the concepts behind them see Programs.
Configure an activity
You configure activities in two places:
- In the program editor (New program, or ⋯ → Edit program) — each activity is a card under its phase. Changes save as you type.
- On the program page — open the activity and expand Details in the inspector. Make your changes and select Save changes. If you collapse Details with unsaved edits, the inspector shows Unsaved changes. The name and description can only be changed in the editor.
| Field | Where | What it does |
|---|---|---|
| Activity title | Editor | Required. A leading ticket key (for example FYAI-514: …) becomes the activity's key on the board. |
| Description | Editor | What the activity is. Used as the task text when Prompt is blank. |
| Owner type | Both | Workflow, Agent, Team or Person. A type the program's access sets to None can't be chosen. |
| Workflow / Agent / Team / Person | Both | The executor. Person lists the members of your project. |
| Workflow attributes | Both (Workflow) | A form for the chosen workflow's attributes, pre-filled with their defaults. These values are passed to the workflow every time the activity runs. |
| Prompt | Both (Agent, Team) | The task text sent to the agent or team. Leave it blank to use the description. |
| Risk level | Both | None, Low, Medium, High or Critical. Any level other than None adds a Risk note field. |
| Knowledge bases | Both | Knowledge bases given to the agent or team when this activity runs, on top of the program's own. |
| Attachments | Both | Reference links: enter a Title and URL and select Add attachment. |
| Done when — acceptance criteria / Acceptance criteria | Both | What a successful run must achieve. Used to verify each run. |
| Deliverable — expected artifact types / Expected artifact types | Both | Artifact types a run must report, such as pull_request. Press Enter to add each one. |
| Depends on | Both | Activities, in any phase, that must succeed before this one can start. |
| When a phase run may run this | Both | The run policy: Once, Always or Until success. See Run policy. |
Preview what an agent or team receives
Under Prompt, select Preview what the assignee receives. The preview shows the exact message the run will be sent — your prompt (or the description) followed by the program context — and updates as you type. Select Hide preview to close it.
When an executor is deleted
If the agent, team or workflow an activity points at has been deleted, Details shows "This agent no longer exists. Pick a replacement below, then save." Choose a replacement and select Save changes. An activity with no executor, or one outside the program's access, is refused when you try to run it.
The activity inspector
Select an activity on the program page to open the inspector. From top to bottom it shows:
- Header — the activity key and phase name, with Open run, Copy link and Close.
- Title and state.
- State section — details for the current state (below).
- Deliverables — what the latest run produced. Tickets the run mentioned are counted as Tickets mentioned rather than listed, because they are references picked up from the transcript rather than outputs.
- Run history — every run, newest first.
- Details — the editor described above.
- Actions — buttons for the current state, at the bottom.
Open run takes you to the latest run: a workflow run opens its mission in Missions; an agent or team run opens its transcript in the chat pane. Copy link copies a link that reopens this activity.
Run history
Each run shows its status (Queued, Running, Completed, Failed or Lost contact), when it ran, its live progress while running, the summary it wrote, any error, and the artifacts it produced. For agent and team runs, Watch it work (while running) or Open transcript (afterwards) opens that run's transcript.
Artifacts appear as chips — Pull request, Ticket, Wiki page, Document, Link, or a platform entity such as Dashboard, Agent, Team, Workflow, Function, Data source, Knowledge base, Mission or Element. Platform entities link to the matching page in the app.
States and actions
Ready
The activity can run now. Select Run now (R). For an activity assigned to a person, select Mark complete instead once the work is done — it records who completed it and when, and clears any failed verification against it.
Waiting
The inspector lists what the activity is waiting on ("Waiting on A2 … (Running)"). The table shows Queued behind with the same keys. Select Edit dependencies to open Details at the dependency list.
Running or Queued
The inspector shows the run's live progress line, elapsed time, time since the last heartbeat, and tokens used so far when the run reports them. Select Open run to watch it.
The Stop button shown for running activities does not currently take effect from the inspector. To cancel a run, call the activity's stop endpoint through the Platform API. A queued run is cancelled straight away; a running one is marked as cancelling until the agent service confirms it has stopped.
Done
The inspector shows the run's summary and its verification:
- Verified with the score, if a verifier passed it.
- Not verified: no acceptance criteria, with Add criteria to open Details at the acceptance criteria.
- Not verified if criteria exist but no verdict was recorded.
Actions: Re-run (ignores the run policy) and Re-verify (V).
Failed
The inspector explains the failure in plain language, with the raw error behind Raw error (with a copy button), and lists under Blocks any activities that can't start until this one succeeds. For a context-window overflow it also shows a gauge of how far over the model's limit the run went.
The recovery actions depend on what went wrong:
| Shown as | Actions offered |
|---|---|
| Ran out of context | Retry on a larger-context model, Split into smaller activities, Retry with guidance |
| Model call timed out | Retry, Retry on a different model, Retry with guidance |
| Rate-limited by the provider | Retry, Retry on a different model |
| Model credentials rejected | Open model settings |
| Couldn't start | Reassign, Retry |
| Model provider error | Retry, Retry with guidance |
| Finished without a result | Retry with guidance |
| Failed (anything else) | Retry, Retry with guidance |
The model, guidance and split actions apply to agent activities only, so they are left out for teams and workflows.
- Retry (
R) runs the activity again unchanged. - Retry on a larger-context model lists your models that have a larger context window than the one that overflowed, largest first. Retry on a different model lists every model. Pick one and select Retry. The override applies to that run only — the agent's own model doesn't change.
- Retry with guidance (
G) asks for extra instructions (up to 4,000 characters) that are sent with the next run only. - Split into smaller activities (
S) opens the program editor with a ready-made request for the Program Assistant to split the activity into smaller ones while keeping its dependencies. Select the suggested prompt in the chat to send it. The request is also copied to your clipboard. - Open model settings opens Models, where you can fix the model's credentials.
- Reassign opens Details so you can choose a different executor.
Verification failed
The run finished, but the verifier judged that it didn't meet the acceptance criteria. The inspector shows the verdict ("Verification failed · score … · judged by …"), the verifier's notes, the Done when criteria, and the Run summary.
- Re-verify (
V) runs the judge again on the latest completed run — useful after you've corrected the criteria. - Accept anyway (
A) marks the run as verified. You must enter a note explaining why the work is good enough; your username is recorded as the verifier. - Re-run with guidance (
G) runs the activity again with extra instructions (agent activities only).
Needs answer
An agent stopped and asked a person a question rather than guessing. The inspector shows who raised it — An agent is asking for help, The system blocked this or A person blocked this — how long it has been waiting, the reason and the question.
- Type Your answer.
- Select Answer and resume to run the activity again straight away with your answer passed to the agent, or Clear without resuming to clear the block and leave the activity for the next phase run.
Answering doesn't release activities that depend on this one — they wait until it actually succeeds.
Needs approval
The phase's checkpoint is waiting for a decision. Select Approve or Reject and then confirm (a note is required to reject). See Running a program.
Lost contact
The run stopped sending heartbeats, so the platform can no longer tell whether it is working. The inspector shows when it last heard from the run.
- Retry (
R) starts a fresh run. - Mark failed ends the run the platform still thinks is open, after you confirm Mark this activity failed?.
Paused, Skipped and Cancelled
Paused activities are held and won't run until resumed. Skipped and cancelled activities are listed under Skipped / Cancelled, which starts collapsed. Pausing and resuming individual activities is available through the Platform API; in the app you pause a whole phase with Pause this phase.