Academy
3. Clarifications and To Decide
Know why @oupi asks questions before starting, how to answer them efficiently, and how to use the To decide list to approve, reject, snooze or cancel pending missions.
When you give @oupi a mission, it doesn't just dive in and guess. If something essential is missing or ambiguous — the audience, the format, a key constraint — it pauses and asks you before starting. This might be a single question or a short series. Your answers are saved with the mission, so @oupi (and you) can always refer back to them. This is the clarification step, and it exists to make sure the work that follows actually matches what you need. A mission built on assumptions costs more to fix than one built on a quick answer upfront.
Keep your answers brief and direct. If you'd rather @oupi proceed with reasonable assumptions instead of waiting for your reply, just say so — it will move forward and note what it assumed. This is especially useful for low-stakes or exploratory missions where speed matters more than precision.
Every clarification and every result awaiting your approval lands in the "To decide" list inside the Assistant. Think of it as your decision inbox: nothing is sent, published, or finalized on your behalf while a mission sits here. From this list you can:
• Answer a clarification so the mission can proceed. • Approve a finished result to finalize delivery. • Reject a result so @oupi knows it missed the mark. • Snooze a mission to come back to it later. • Cancel a mission to stop work entirely (what was produced so far is kept).
Checking "To decide" regularly keeps your missions flowing and avoids bottlenecks.
Write your original mission clearly — state the outcome, who it's for, and the format you want — and you'll get fewer clarification questions. @oupi plans better from a clear destination than from a long list of steps.
Once a mission finishes, @oupi runs a verification pass that compares the result to your original goal. The verdict is one of three:
• Verified — the result matches the goal. • Failed — something went wrong. • Could not be verified — the checker couldn't read the output (e.g., a binary file or an external action). This is not a failure; it just means you should review it yourself.
If the result needs changes, reply directly from the mission page. @oupi keeps the full context of what it did and concluded, so your follow-up starts a child mission instead of starting from scratch.
Open the Assistant and go to "To decide." If any missions are waiting, try each action: answer a clarification, approve or reject a result, snooze one for later, or cancel one you no longer need. Notice how nothing is finalized until you act.
@oupi asks clarifications when key information is missing — answer briefly or tell it to assume. All pending decisions live in the "To decide" list: answer, approve, reject, snooze, or cancel. Nothing is delivered until you act. After delivery, check the verification verdict, and reply from the mission page to refine results without losing context. Clear goals up front mean fewer questions and lower costs.