Skip to main content
Every writeback report includes an outcome — a structured label that describes what happened. There are five possible values.

The five outcomes

Everything worked exactly as expected. The agent completed the task without issues, confusion, or guesswork.Use this when the tool performed correctly and the instructions were clear. Success reports are just as valuable as failures — they tell you what’s working.Example: “Generated a 5-tab xlsx with pivot tables. All formulas working correctly. Instructions were very clear.”
The tool returned an error, an unexpected result, or didn’t work at all.Use this when something went technically wrong — a 4xx/5xx response, an empty result when data was expected, or output that clearly didn’t match the request.Example: “search_index returned 0 results for ‘machine learning papers’ but the index has 2,000+ entries. Possible encoding issue with the query.”
The tool worked (or might have worked), but the instructions were unclear, ambiguous, or contradictory — leading to guesswork.Use this when the agent had to guess at parameters, tried multiple approaches, or wasn’t sure which tool to use. The tool doesn’t need to have failed — confusion is its own category.Example: “The skill says to ‘run the generator’ but there are 3 tools with ‘generate’ in the name. Unclear which to use for PDF output.”
The agent could not complete the task and stopped trying.Use this when the agent exhausted its options and gave up — whether due to repeated failures, total confusion, or hitting a dead end with no path forward.Example: “Tried 4 different parameter combinations for the export tool. None produced valid output. Gave up and told the user I couldn’t complete the task.”
The agent wanted a feature or capability that doesn’t exist yet.Use this when the tool works fine but is missing something that would make it significantly more useful. This is product feedback, not a bug report.Example: “The export tool only supports PDF. A CSV option would make this much more useful for downstream processing.”

Choosing the right outcome

If multiple outcomes seem to apply, use the one that best describes the primary thing that happened:
SituationOutcome
Tool returned a 500 errorfailure
Tool returned success but result was wrongfailure
Instructions were unclear but tool workedconfusing
Had to guess at a parameterconfusing
Tried everything, nothing workedgave_up
Tool works but is missing a featurerequest
Task completed without issuessuccess

The details field

Always include a details string with your report. This is where the real value is — a natural language description of what you tried, what happened, and why you chose this outcome. Be specific:
  • What input did you use?
  • What did you expect to happen?
  • What actually happened?
  • Which tool or endpoint were you using?