# Subagents get a 5-minute prompt cache. What I changed once I measured what that cost me — retained record

> Retained reddit record from reddit:ClaudeCode.

Canonical: https://highsignal.app/data/records/d62123b01335ab33

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:ClaudeCode)  published 2026-10-21 12:08 ingested 2026-10-22 02:42

[original source ↗](https://www.reddit.com/r/ClaudeCode/comments/1xble3m/subagents_get_a_5minute_prompt_cache_what_i/)

## retained text



```
On a subscription, the main session's prompt cache lasts 1 hour; subagents', teammates' and forks' last 5 minutes. When a subagent with a big context waits on something slow (a test suite, an e2e harness, a poll loop), its next turn writes the whole context to the cache again. Over four months, avoidable cache rebuilds were about a quarter of my usage at API prices, and subagents waiting like this were the biggest part.

What I changed:

1. **Agent teams off.** Teammates were 72% of my subagent usage: each kept its whole history across tasks and re-read it every turn.
2. **One task per agent**, then a fresh one with a written brief. I only message an agent again if it finished under 5 minutes ago and is still small.
3. **Nobody waits inside a big context.** Long commands run from the main session (which has the 1-hour cache) in the background, through a small wrapper that writes the full log to a file and prints only the exit code and the failure lines. A fresh agent gets the failures.
4. **Compaction on purpose:** a 400K auto-compact backstop (`autoCompactWindow`), a status line with context size and a cache countdown, and `/compact <focus>` at the end of each phase.
5. **Less noise:** Bash output compacted with RTK.

The week after, against the week before, on about the same number of requests: avoidable rebuilds 27% → 5% of usage, usage on contexts over 400K 68% → 3%, cost per request at API prices $0.11 → $0.06. Not from cheaper models: the mix got more expensive. One week is a small sample.

Generic versions of the agent files, the rules and the wrapper: [https://github.com/27medkamal/claude-cache-audit/tree/main/examples](https://github.com/27medkamal/claude-cache-audit/tree/main/examples)

To check your own transcripts: `uvx claude-cache-audit` (local only, no prompt text printed). Write-up: [https://27medkamal.com/blog/claude-code-usage/](https://27medkamal.com/blog/claude-code-usage/)
```



document fetched 2026-10-22 00:43 · canonical https://reddit.com/r/ClaudeCode/comments/1xble3m/subagents_get_a_5minute_prompt_cache_what_i
