[BUG] Scheduled /loop task fires far more frequently than its configured interval, burning premium-tier usage each firing

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report
• [x] I am using the latest version of Claude Code

What's Wrong?

A scheduled /loop task, on a background session, fired far more frequently than its configured interval. The configured interval was on the order of tens of minutes. Observed directly on-screen (Remote Control view of the background session): multiple /loop invocations firing back-to-back within minutes of each other, visible together on one screen, each one dispatching a fresh agent and running a real batch of tool calls (one instance ran 11 tools in a single firing, others ran fewer but still nontrivial batches).

This is not a cheap no-op poll. Each firing is a full premium-tier turn — a new agent dispatch plus real tool execution — so a firing-rate bug here directly and substantially inflates usage consumption, independent of and likely compounding a separate, already-reported issue where the background session itself was never actually stopped despite the user believing it was closed (see #67506, #73597, #56913).

Reproduction

1. Configure a /loop scheduled task on a background session with an interval on the order of tens of minutes.
2. Let the session run unattended in the background for an extended period.
3. Observe the session's transcript/Remote Control view: multiple /loop firings appear clustered together, well inside the configured interval, rather than spaced at the configured cadence.

Impact

Given how central /loop and scheduled tasks are to any long-running or autonomous Claude Code usage, a firing-rate bug in the scheduling mechanism would affect far more than one account — anyone relying on /loop for periodic background checks would have their actual usage consumption significantly higher than the configured interval implies, with no clear indication in the UI that firings are happening more often than scheduled.

Related

• #67506 (Fable usage/token consumption not matching description — this finding was surfaced while investigating that thread)
• #73597 (Opus subagents billed as Fable usage)
• #56913 (background session input-misrouting, same general session)