https://docs.all-hands.dev/ logo
Join Slack
Powered by
<@U06RMLQGNGM> here is the hierarchy on the variou...
# proj-automation
q
@breezy-kilobyte-39430 here is the hierarchy on the various status informations associated with an automation run
👀 1
Automation run info has three layers: 1. Run lifecycle status Field: 
run.status
Meaning: Did the automation run finish at the system level? Values: 
PENDING
, 
RUNNING
, 
COMPLETED
, 
FAILED
, 
CANCELLED
, 
SKIPPED
Use for the primary run state badge. 2. System / execution error details Fields: 
run.error_detail
, 
run.status_detail
Meaning: Why the run infrastructure or script failed. Examples: sandbox timeout, callback failure, rate limit, transport/server error, environment unavailable. Use for technical failure messages, retry hints, and infrastructure blockers. 3. Task outcome Field: 
run.run_metadata.finish_tool_response
Meaning: What the agent says happened with the user’s actual task. Current shape:
Copy code
{
  "status": "success | partial_success | blocked | failed | unknown",
  "outcome_summary": "Concise human-readable summary"
}
Use for the user-facing task summary, blocker explanation, and next steps. Key distinction: a run can be system-level 
COMPLETED
but task-level 
blocked
. Example:
Copy code
{
  "status": "COMPLETED",
  "error_detail": null,
  "run_metadata": {
    "finish_tool_response": {
      "status": "blocked",
      "outcome_summary": "Could not post the report because Slack credentials were missing."
    }
  }
}
Meaning: • The automation ran successfully. • The agent completed its execution. • But the actual user task was blocked.
b
@quick-window-91705, I made some changes which you can find here. Image is attached. My main question, can we condense the error message into the one line (as shown in the preview). The Activity Log badge should answer “did my task work?”, not “did the process exit.” The three layers are right; I would recommend they don't share the same pill. Row (scan) One badge. Optional task summary under the timestamp (includes the error). No red 
error_detail
. • Pending / Running → lifecycle • Run crashed, cancelled, or skipped → lifecycle • `COMPLETED`+`finish_tool_response`→ task badge • `COMPLETED`with no task outcome → Successful Task status → Badge success → Successful blocked → Blocked failed → Failed partial_success → Partial unknown → Needs review The HubSpot case (
COMPLETED
+ blocked, missing key) should read Blocked plus the agent summary — not Successful with a red env error. Infra copy stays in details. Details (inspect) Label all three: Run (
COMPLETED
), Task (
Blocked
+ summary), System (
error_detail
/ 
status_detail
when present). Preview of every state in the real page chrome: 
/automations/preview-activity-log
on 
activity-log-row-preview
github.com/FraterCCCLXIII/openhands-main/tree/activity-log-row-preview Let me know what you think!
@quick-window-91705 thoughts on this?
q
this looks great! one line summary is hard to enforce cause the llm can decide to misbehave so some kind of option to expand would be good, otherwise looks great
b
Updated the branch to include that @quick-window-91705
🙌 1
q
i have an initial implementation here as well, there is a case where the user provides custom metadata for the agent to fill out could use some feedback there - ive attached screenshots github.com/OpenHands/OpenHands/pull/16937/changes
Open in Slack
PreviousNext
Powered by