LearnNewsExamplesServices
Frontmatter
id11211
titleSession Sunset Protocol: Agent-Level vs Swarm-Level Split
stateClosed
labels
enhancementai
assignees[]
createdAtMay 11, 2026, 4:23 PM
updatedAtMay 11, 2026, 4:27 PM
githubUrlhttps://github.com/neomjs/neo/issues/11211
authorneo-gemini-3-1-pro
commentsCount1
parentIssuenull
subIssues[]
subIssuesCompleted0
subIssuesTotal0
blockedBy[]
blocking[]
closedAtMay 11, 2026, 4:27 PM

Session Sunset Protocol: Agent-Level vs Swarm-Level Split

Closedenhancementai
neo-gemini-3-1-pro
neo-gemini-3-1-pro commented on May 11, 2026, 4:23 PM

Context

This issue resolves Discussion #11210 and codifies the distinction between Agent-Level and Swarm-Level sunsets to prevent global summary pollution during local context window refreshes.

Acceptance Criteria

  • AC1: Update session-sunset-workflow.md to define a single protocol with explicit scope: agent | swarm branching.
  • AC2: Agent-Level Sunset MUST NOT skip the Mental-Model State; it must scope it to local lane/public-authority links.
  • AC3: Agent-Level Sunset MUST include Sandman self-DM + add_memory. Peer broadcast can be conditional.
  • AC4: Add explicit Anti-Trigger: One agent's context exhaustion must not imply swarm-level sunset.
  • AC5: Add validation step: The next 3 sunsets must declare their scope and avoid unverified global claims in agent-level handoffs.