The user has strong feelings about architecture, plan adherence, and especially about agents making impactful decisions or concessions without consultation.
DO NOT make them alone.
If you encounter an unanticipated obstacle, YOU MUST consider stopping to discuss it. If the obstacle:
- requires adaptations that could alter planned functionality
- presents multiple solutions with different tradeoffs
- requires substantial research, debugging, or diverts you from the task
- suggests introducing new imports or coupling vectors
- risks policy violations or doctrinal ambiguity (i.e. heresy — see
/canon) - tempts a shortcut or omission
YOU MUST STOP IMMEDIATELY.
Present the situation with context. Summarise:
- the relevant task code, plan,
design.md, or governing canon - what you are trying to do
- the nature of the obstacle
- relevant discoveries
- options and/or hypotheses
- any obvious tradeoffs
Then ask for guidance before proceeding.
If the obstacle is really a piece of latent work to defer rather than a
decision needing the user now — a follow-up, a newly-spotted risk, a chore — then
capture it with backlog new so it is not lost to the conversation, and say
so. Is it work, knowledge, or a decision? See the boundary in
lib:reference/using-doctrine.md.
Capture the consult's outcome via /knowledge: a resolved tradeoff → DEC; one
still open → QUE, with the blocked work gating itself (doctrine needs <work> QUE-n).