Skip to main content

Guide 5: Read the Results

After a pipeline run completes, this guide shows you where to find each piece of information and what it means.

The Run Detail Modal​

Open a run from the issue page (click View Details or the run history entry). The modal has three main areas:

Left panel — run metadata​

  • Status — final run status: Completed, Failed, Timed Out, Cancelled
  • Trigger — always "manual" for user-triggered runs
  • Started / Duration — when the run started and how long it took
  • Cost — total cost in USD across all stages (only shown if > $0)
  • Stage timeline — clickable list of all stages with status badges

Right panel — stage details​

Click any stage in the timeline to see its details:

Output tab:

  • Full transcript of the stage's stdout, formatted as:
    • Text messages from the AI
    • Tool calls (e.g., bash commands the AI ran)
    • Tool results (indented; red border = error)
    • Final result with cost and token counts
  • Toggle Raw JSON to see the raw event payloads (useful for debugging)
  • Toggle Verbose to show lifecycle events (launched, completed) and all tool results

Gates tab:

  • Shows pass/fail for each gate rule
  • Green checkmark = rule passed
  • Red X = rule failed, with the actual value that caused the failure
  • Overall verdict: Proceed, Hold, Rework, or Abort

The Activity Feed​

The activity feed on the issue page shows issue-level events — not stage output. This includes:

  • State changes (e.g., "Moved from Research to Implement")
  • Comments posted by skills (from the result document's comment field)
  • Blockers surfaced by skills

For stage output (the AI's actual work), always use the Run Detail Modal.

Understanding costs and tokens​

  • Cost is the sum of all stage costUsd values from the result documents
  • Tokens are broken into input and output, per stage
  • Cost only appears if the driver reported it in the result document metadata (meta.input_tokens, meta.output_tokens)

What to do when a run fails​

  1. Open the Run Detail Modal
  2. Check the Output tab — look for the last tool call and its error output
  3. Check the Gates tab — if a rule failed, the verdict and reason are shown there
  4. Check the issue's State and Status — Blocked status means a skill emitted blocked verdict or a needs_human signal and needs your attention
  5. Fix the underlying issue (update the skill prompt, fix the code, clear the blocker), then trigger a new run