Feature request: Reserve capacity for graceful wrap-up + low-power mode near limit

Problem

When the 5-hour usage limit is hit, Claude Code stops abruptly mid-task. There is no warning, no ability to save state, and no opportunity for an orderly session close. This results in:

• Lost context and in-progress work
• No ability to write wrap-up summaries or update memory/task files
• Abrupt end to multi-step operations (agents, file edits, etc.)

Proposed Solution

1. Reserve capacity (last ~5% of limit)

Hold back a small reserve (~5% of the 5-hour window) that is only accessible for wrap-up operations. When the main limit is exhausted:

• Claude can still run a lightweight wrap-up sequence (save memory, write summaries, close open files)
• Reserve is strictly gated — no new tasks, only finalization work
• Reserve expires after a fixed time window (e.g. 10 minutes) if unused

2. Low-power mode (last 10% of limit)

When usage reaches ~90% of the 5-hour window:

• Notify the user that the session is approaching its limit
• Enter a reduced mode: only short, bounded responses; no spawning new agents; no large file operations
• This gives the user and Claude time to wind down gracefully rather than hitting a wall

Why this matters

Power users running long agentic sessions (background agents, multi-step pipelines, automated maintenance) have no way to anticipate or handle the hard cutoff. A graceful degradation model — similar to how operating systems handle low memory — would significantly improve reliability for agentic workloads.

Workaround currently needed

Users must manually monitor usage via /usage or claude.ai/settings/usage and preemptively run wrap-up commands. This is error-prone and interrupts flow.

---
Filed on behalf of a user via Claude Code CLI after /feedback command failed to submit.