Open Decisions

These are deliberately unresolved. Do not let an AI agent quietly decide one while implementing an unrelated lesson.

ID	Decision	Why postponed	Decision point
OD-001	Does failed required verification hard-block human approval, or require explicit acknowledgement?	We need to use the real review loop first; both policies have tradeoffs.	Review milestone, after repeated use.
OD-002	Which real agent is the first production adapter?	Claude Code is the current assumption, but the fake adapter should make this cheap to reconsider.	Before M7.
OD-003	Exact Involvement Mode presets and which escalation reasons each subscribes to.	The mechanism needs to be proven across varied tasks before tuning product presets.	Decision/escalation milestone.
OD-004	Whether tiny human-authored tasks should avoid worktrees by default.	Worktree setup has a real per-task tax.	After M8 usage.
OD-005	Whether code integration defaults to merge, rebase, cherry-pick, or an explicit per-project policy.	We first need reliable review/revision semantics.	Integration hardening.
OD-006	Whether Project Knowledge becomes a structured knowledge subsystem or remains lightweight artifacts/notes.	Not central to first work loop.	Post-MVP.

How to close a decision

When evidence is sufficient:

1. record the chosen answer here;
2. update the canonical product/architecture/design doc;
3. add an ADR if the reasoning is important/non-obvious;
4. update course/plan if implementation order changes;
5. remove the row from the open table or move it into a “closed” section.