Problem

When cron jobs or sub-agents spawn isolated sessions that fail, they leave behind "zombie" sessions that:

• Consume memory/context tokens
• Clutter the session list
• Cannot be cleaned up by the agent that created them

Current Gap

Existing tools:

• sessions_list — View sessions
• sessions_send — Message sessions
• sessions_spawn — Create sessions
• sessions_history — View history

Missing:

• sessions_kill — Terminate a specific session
• sessions_cleanup — Bulk kill orphans

Real-World Impact

During automation testing, 23+ isolated cron sessions were spawned. The agent could see them via sessions_list but had no mechanism to clean them up. Required a full gateway restart to clear ~250k tokens of wasted context.

Proposed Solution

sessions_kill:

• sessionKey (required): session to terminate
• force (optional): kill even if active

sessions_cleanup:

• filter (optional): "orphaned", "inactive", "cron:*", etc.
• olderThan (optional): minutes since last activity
• dryRun (optional): preview without killing

Security Considerations

• Agents should only kill their own sessions
• Cannot kill system/core sessions
• Bulk operations may need confirmation/elevation

Priority

Medium — Gateway restart works as workaround, but impacts resource management.