social-search ยท diff
git:20260912.201ee8a to git:20260912.c0c8b18
6 added, 4 removed. Audit A to A.
---
name: social-search
description: Research bounded public source scopes, then return one independent ranked assessment.
---
- Reuse or establish [library context](../../references/library-context.md). Stay in the caller's context.
+ Stay in the caller's context. Reuse or establish [library context](../../references/library-context.md).
- Choose only source assignments likely to contribute useful evidence, preserving requested sources, dates and bounds. Closely related sources may share an assignment when that reduces agent overhead. Share the question and output requirements, divide caller bounds across assignments, and give each a separate evidence directory.
+ Choose assignments by their contribution to the question, preserving requested sources and dates. Group overlapping scopes when separate workers would repeat the same investigation. Give shared original-source checks one owner; other assignments collect distinct evidence, passing canonical leads to that owner and reusing its support.
+ Divide caller bounds across assignments. A total time limit includes evidence writing, gathering and final review: reserve time for those when setting read and handoff deadlines. Share the question, output requirements, ownership and deadlines; give each assignment a separate evidence directory.
+
Invoke [search-site](../search-site/SKILL.md) here once per assignment with deferred gathering. Launch independent assignments before gathering, within capacity and shared provider limits.
- Gather actual outcomes from every assignment. Await or stop unfinished workers at the caller's deadline; retain missing or failed assignments as gaps.
+ Gather every assignment's actual outcome by its handoff deadline; stop unfinished workers and retain partial artifacts and missing or failed assignments as gaps.
- Invoke [rank-evidence](../rank-evidence/SKILL.md) once with the question, all returned evidence and gaps, and the desired report location.
+ Invoke [rank-evidence](../rank-evidence/SKILL.md) once with the question, all returned evidence and gaps, remaining bounds and report location.
Return the assessment. For N assignments, the workflow uses N workers and one reviewer.