[witbrock]

Blocked coding tasks currently wait without automatically giving a supervisor a repair task. This change uses the task's report_to_concept_id, falling back to the assignee's ontology supervisor, and creates one linked repair task per unresolved episode through the existing controller. Creator, organisation, assignee and native-writer checks still govern execution; reporting responsibility does not grant execution authority.

The handoff preserves source instructions, comments, evidence, execution preferences and prior attempts. A completed repair alone cannot restart its source: the supervisor must provide fresh matching observations, which the controller records and the shared retry gate consumes once. Lost creation acknowledgements, repeated polls and same-blocker manual retries reuse the repair. Cycles, changed source instructions, reassignment and inaccessible routes remain visible without launching another coding run. The feature is opt-in through operator configuration; the private Mac adapter uses the same shared methods.

Validation: 221 focused backend/controller/task/release tests passed, including canonical task/message replay, lost acknowledgements, two-hop escalation to a human, other-task progress and single continuation. The private Mac adapter canonical replay also passed without launching a coding model. Four handoff tests were rerun successfully after adding source comments/notes to the repair context. Existing unrelated Ruff findings remain unchanged; changed/new module checks and diff checks passed.

Merge decision: ready on the bounded Tier 2 evidence. Activation uses existing idle controller locks and retained rollback; no public web deployment. Legacy reporting defaults require an exact reviewed migration, preserving explicit choices. Unclassified legacy blockers require reviewed continuation rather than invented recovery observations.

Von task: #V#task_agent_c5b7f4c66b71aa506cbe3f9e2fbac25c