A task's status tells you where it is on the path from description to pull request.
| Status | Meaning |
|---|---|
PENDING | Accepted, waiting for a free agent slot |
IN_PROGRESS | Actively being worked on |
PENDING_REVIEW | Work is done and reviewed, awaiting integration or approval |
COMPLETED | Delivered |
NEEDS_HUMAN_REVIEW | Coroid cannot proceed alone and is asking you something |
PAUSED | Temporarily stopped, resumable |
FAILED | Execution could not complete |
CANCELLED | Stopped by you |
WONT_FIX | Closed deliberately without doing the work |
PENDING
Queued. The task is valid and waiting for capacity.
If it sits here longer than you expect: all your slots are busy, or
higher-priority tasks are ahead of it. Check priority (P0 to P3) and
capacity. Adding slots shortens the queue;
it does not speed up work already running.
IN_PROGRESS
An agent is working. You can watch the run live.
If it runs much longer than similar tasks: usually a slow or hanging test suite, or a specification broad enough that the agent keeps finding more to do. See When a task gets stuck or fails.
PENDING_REVIEW
The work is built and has been through an automated review pass. It is waiting on integration or on you.
What to do: review the pull request. See Review and merge the result.
COMPLETED
Delivered. For most tasks this means the pull request is open and ready; merging remains yours.
NEEDS_HUMAN_REVIEW
This is not a failure. The task hit a decision only a person can make — an ambiguous requirement, two equally valid approaches, a missing credential, a constraint that was not stated.
What to do: read the question, answer it, and the task continues from where it paused. Answering precisely is worth the extra sentence; a vague answer produces another guess.
If you see this status often, the usual cause is under-specified work. See Specifications.
PAUSED
Stopped on purpose and resumable. Its slot is released while paused.
FAILED
Execution could not complete. Distinct from work that ran and did not pass — that comes back as rework or findings.
Common causes:
- build or test commands that do not work in a clean workspace
- a missing dependency, credential or service the suite needs
- the repository connection losing access mid-run
- the task exceeding a run limit on your plan
What to do: read the failure output, which names the stage that failed. Most first-time failures are environment configuration — see Build and test configuration.
CANCELLED
You stopped it. The run halts and the slot frees immediately.
WONT_FIX
Closed deliberately without doing the work — already resolved, not applicable, or
decided against. Distinct from CANCELLED in intent: WONT_FIX is a judgement
about the work, CANCELLED is a judgement about the run.