Decide whether the report-only agents (qa-reviewer, principal-architect) should close the finding loop like the log-auditor now does

task-report-only-agents-finding-loop

task confidence inferred status backlog 2026-06-16 owner product-owner

source log-auditor — surfaced auditing DEC-0036; the candidate-follow-up the decision named, judged real and not already on the board

Decide the finding loop for the report-only agents

Upgrade the log-auditor from a backward-looking recorder into a recorder + standing auditor that closes the improvement loop closed the loop for the log-auditor: a surfaced improvement now becomes a durable, owned board task (or an explicit flag) instead of a sentence in a reply that disappears. The decision named a candidate follow-up — that the same pattern may belong on the other report-only agents — and left the scoping call to an owner. This task is that call.

The gap

Adversarial QA & Knowledge-Integrity Reviewer — its spec says, verbatim, "reports findings, does not silently fix" and "You report; you do not silently edit. Hand fixes to the owning agent, or propose them as decision records for the log-auditor when they're judgment calls." There is no durable home for those findings today: a QA pass that finds real issues has the same evaporating-prose problem DEC-0036 just fixed for the log-auditor.
Principal Platform Architect — drafts decisions for the log-auditor to record, but routes ripple-effects / downstream changes as prose. Possibly in scope; weaker case (it already hands consequential calls to the log-auditor, which now files).

Scope to decide (the why this is a decision, not a mechanical edit)

Does qa-reviewer adopt the file-a-task + structured-return pattern? (Strongest case — it is purely report-only with no durable outlet.)
Does principal-architect adopt it too, or is its existing "draft → log-auditor records" handoff sufficient now that the log-auditor files?
Whatever is adopted must carry the "don't duplicate an open task / don't clutter the board" bar so report-only agents do not flood the backlog with inferred items.

Notes

Owned by Product Owner (scope/what-and-why) — it changes more than one team member's mandate and is a deliberate operating-model call, not a one-line fix. Do not edit the agent prompts here; set the scope first, then the owning agents' prompts are updated as a follow-on. Provenance: surfaced by the log-auditor while auditing DEC-0036, confidence: inferred (agent-judged real, not human-curated).