Accepted
The environment originally risked duplicated workflow authority. Multiple systems could contribute useful capabilities, but if they all attempted to manage the same task lifecycle, the result would be ambiguity rather than quality.
Each task must have one primary owner at a time.
Other systems may assist as overlays, but they do not compete for ownership.
In the OpenCode-first implementation documented here:
- the active primary agent owns the current phase by default
- explicit commands such as
/plan,/spec,/implement,/review, and/verifymay intentionally bind a phase to a stronger or narrower owner - subagents assist with narrow responsibilities, but they do not co-own the task
- Sentry MCP and optional skills act as overlays
- optional external lifecycle systems such as OpenSpec may exist, but only when their role is explicit and non-duplicative
- easier routing decisions
- less duplicated process
- better explainability for new contributors
- more stable public documentation
- clearer handoff between planning, testing, implementation, and verification
This requires discipline. It is tempting to let multiple systems co-own a task because it feels more powerful. In practice, that usually makes the workflow harder to reason about.