| name | preflight |
| description | Before starting anything new, understand task, intent, and context. |
You're about to begin work on something: $ARGUMENTS
First: you need to understand what it entails. Your immediate task is to
correctly decide how much effort to spend on this preliminary research.
Too little is as bad as too much. You must estimate where the goldilocks zone is,
and arrive with maximum tokens intact.
You may be expected to:
- clarify a vague task
- methodically diagnose a fault
- write or critique a thorough and considered design doc which fits the
implementation surface
- execute a thoroughly prepared plan with efficiency
- preserve your own context by delegating to sub-agents
If you can't tell, ask now.
Learn just enough to ask the user the right clarifying questions. These
might include: how deeply should you research?
If you chase the questions which arise as you go, you'll disappear like a
helium balloon into the open sky.
Instead:
- Read the material in front of you
- Take stock of relevant
/retrieving-memories and /doctrine
- Decide up front, out loud:
- what, concretely, you need to know.
- when, concretely, you will stop even if you don't have all the answers.
- Remain curious and collect further questions, but don't run off after them.
- Stop when you reach your stopping conditions.
- Present any significant discoveries, open questions, unresolved tradeoffs to the user.
- Indicate your readiness to complete the assigned task; suggest the next step.
Repeat, if the user indicates to.
Do not begin implementation without acknowledgement from the user.