Claude Code subagents can't be paused or controlled, user says

A user who relies on Claude Code as a 'thinking partner' for analyzing large collections of documents, notes, and data — rather than for coding — describes a problem with parallel subagents that spawn with no clear way to stop or manage them. Each subagent independently reads files, calls tools, and burns tokens on its own. Once the 5-hour usage limit is hit, the Claude Code container freezes and the session ends.

When the container is reopened later, the subagents are already dead — including any that were mid-computation or blocked — and none of that work survives. This means tokens spent during the prior five hours are effectively wasted, since every subagent has to restart from scratch: rereading the same documents, rerunning the same analysis, and rebuilding the same understanding. A commonly suggested workaround is having each subagent periodically dump its progress every 5-10 seconds, but that forces subagents to spend time and tokens serializing their state instead of doing the actual reasoning work.

The user stresses they are paying for reasoning and exploration, not repeated summarization.