States and evidence
Every state an agent, a question and a result can be in, what each one means, and what you can do next.
Sculptured shows three separate things about an agent: whether its process is running (its lifecycle), what it is doing right now (its activity), and how much you can trust its result (its evidence). Each state has a shape and a word as well as a colour, so it reads the same without colour.
Agent lifecycle
An agent's lifecycle says whether a process exists for it. Your agent runner reports most changes as it sees them. A few are inferred by Sculptured, for example when the link to a computer is lost; those are labelled heuristic in the agent's history.
stateDiagram-v2
[*] --> planned: "team started"
planned --> launching: "runner got the launch"
planned --> attention: "no Claude Code / Codex on the computer"
launching --> running: "session ready"
launching --> attention: "exited before ready, or 60 s passed"
running --> stopping: "Stop sent"
stopping --> stopped: "process exited"
running --> attention: "exited unexpectedly"
running --> stale: "computer's link lost"
stale --> running: "computer back, process still alive"
stale --> attention: "computer back, process gone"
attention --> launching: "Restart, or Resume / Start fresh"
attention --> stopped: "Leave stopped"
stopped --> launching: "Restart, or start the team again"
| State | Shown as | What it means | What you can do |
|---|---|---|---|
| Planned | ○ planned | On the design, not launched yet in this run. | Wait for its turn, or start the team. |
| Launching | ◌ launching | The runner is starting Claude Code or Codex. It must be ready within 60 seconds. | Wait. |
| Running | ● running | A process is running. Its activity says whether it is working. | Send guidance, interrupt, soft stop or hard stop. |
| Stopping | ◑ stopping | A stop was sent and Sculptured is waiting for the process to exit. If the computer is offline, the stop is carried out as soon as it reconnects. | Wait. |
| Stopped | ■ stopped | No process. Its history is kept. | Restart the agent, or start the team again. |
| Stale | ◇ stale | The computer's link was lost while the agent was running. What you see is the last known state, not the current one. | Wait for the computer to reconnect, or check the computer. |
| Needs attention | ▲ needs attention | The agent stopped when it should not have, could not start, or was lost in a restart. The agent's page says why. | Read the reason, then Restart, Resume, Start fresh or Leave stopped. |
Activity
Activity is only current while the agent is running. While it is stale, the activity is the last known one. Otherwise it is not shown.
| Activity | Shown as | Meaning |
|---|---|---|
| Working | ⚙ working | A turn is in progress. |
| Waiting for input | ? waiting for input | The agent asked a question and is waiting for an answer. |
| Idle | · idle | The session is open, but no turn runs. An idle agent is not doing anything. |
| Unknown | ∅ unknown | Sculptured cannot tell. |
What the board shows
The board and the Run view combine lifecycle and activity into one word per agent. Blocked agents are sorted first.
| Board state | Shown when | What to do |
|---|---|---|
| ◉ needs you | The agent waits for your answer, or needs attention. | Open it and answer, or read its reason. |
| ▲ stuck | The agent is idle and its handoffs cannot go on by themselves. The pane says why. | Use Hand over again or Resend handoff, or send guidance. |
| ⧗ waiting for a slot | Its next message waits until fewer of the run's agents work at once (the run budget's Most agents working at once). The agent's page says "waiting for a slot to run". It is not stuck. | Nothing: it starts when a slot frees. Raise the limit in Run budgets to run more at once. |
| ⣟ working | A turn is running. | Watch, or send guidance. |
| ✓ done | A turn ended since you last looked. It does not mean the work is right. | Read the result and its evidence. |
| ○ idle | Running, with nothing to do. | Nothing, or send new work. |
| ∅ unknown | The computer is unreachable, or no state is known. | Check the computer. |
Launching, stopping, stopped and planned agents show their lifecycle word.
Questions
A question comes from an agent's request_input tool, from the CLI's own question prompt, or (for a restricted agent) from a permission prompt. See Questions and approvals.
| State | Meaning |
|---|---|
| Open | Waiting for your answer. It shows in Needs you. |
| Answered | You answered before the timeout. |
| Expired | The timeout (10 minutes by default) passed first. The agent decides for itself and says what it decided. A late answer is sent as an overriding instruction. |
| Withdrawn | The question can no longer reach the agent, for example because the agent stopped. The reason is shown. |
An open question stays open while a computer is briefly offline. You can answer it, and the answer reaches the agent when the computer reconnects. See When a computer drops off.
Evidence: reported, observed, verified
Every claim on the board carries where it came from. Sculptured never upgrades an agent's words to a fact.
| Label | Shown as | Where it comes from |
|---|---|---|
| Verified | ◆ verified | An independent check, such as a test command Sculptured ran itself, and its exit status. |
| Observed | ◆ observed | Something Sculptured saw: the runner's protocol, a file on disk, a captured tool call and result, or your own action. |
| Reported | ◇ reported | The agent's own words, such as its report or "done". |
| Heuristic | ≈ heuristic | Sculptured's own reading, for example how it classified a tool call. It never blocks an agent. |
| Unknown | ∅ unknown | No source is recorded. It stays unknown; nothing is invented. |
Outcome of a piece of work
A finished assignment shows one of these outcomes:
| Outcome | Meaning |
|---|---|
| Reported complete | The agent said it finished. Nothing checked it. |
| Captured tool result | A tool call's result was captured as evidence. |
| Verified | A check ran and passed. |
| Settled, unverified | The work ended, but no check ran. |
| Check failed | A check ran and failed. |
Work items in a workflow also show progress words: pending, in progress, waiting for you, blocked, reported complete, failed, canceled and handed off.