Rank, 14 days: 13 on the Claude Code sheet
56 people reported

All time, since Sep 14, including 31 thumbs-up on GitHub.

Prompt cache misses and re-writes

“Interactive sessions get only 5m prompt-cache TTL while claude -p on same login gets 1h”

Is this your bug? What gets counted gets fixed. Pile on.

Paste this into your agent. It files your report, with your error and version.

I hit this bug: rapsheet.dev/sheets/claude-code/themes/241/file
Follow it to file my report. Show me the text before you send.

No install. No account. Public and anonymous. The link asks your agent for one report, nothing else. What it sends

Still happening. 7 reports and thumbs-up in the last 7 days and 6 the week before, down from 15 a week at the late-October peak. Not fixed, and no vendor reply yet.

11 bugs in this theme

  1. Interactive sessions get 5m cache TTLHas a workaround18 people
  2. Background agent summary requestsNo workaround yet12 people
  3. Sleep re-sends full agent contextNo workaround yet4 people
  4. Context after prompt breaks cache4 people
  5. Subagent resume cache miss4 people
  6. Resume re-writes prompt cacheNo workaround yet3 people
  7. CLAUDE.md cache not sharedHas a workaround3 people
  8. Prompt cache misses after system messageHas a workaround3 people
  9. Warm prompt cache fully missedNo workaround yet2 people
  10. Cross-session prompt cache missesNo workaround yet2 people
Show all 11Show only the largest 10
  1. Subagent prompt cache expiresNo workaround yet2 people

What's included

  • Mid-session, cache reads drop to the system-prompt floor or to zero and the whole conversation is re-written, sometimes for dozens of consecutive calls
  • Headless -p runs and --resume re-write the full history every turn on some models, linked in reports to trailing role:system messages or turns without thinking
  • New sessions never share cached CLAUDE.md, memory or skills context because a per-session scratchpad or job path, or the first prompt, sits in the same prefix
  • Interactive sessions write 5-minute cache entries while claude -p on the same account writes 1-hour entries

Not included

  • Usage limits or billing totals reported without cache read/write measurements
  • Context compaction behavior unrelated to cache hits

Original reports27, newest first

  1. Background subagent progress summary misses prompt cache, rewriting full context about every 35s#106863
  2. agent_summary re-sends each background agent's full transcript every ~30 s, with no way to turn it off#105981
  3. Prompt cache: CLAUDE.md context is rewritten on every session start because it shares messages[0] with the first prompt#104051
  4. Interactive sessions request 5-minute prompt cache on enterprise org, headless uses 1-hour TTL#103753
  5. Interactive CLI uses 5m prompt cache TTL instead of 1h since v2.1.301#103673
  6. Interactive sessions default to 5m prompt-cache TTL while claude -p gets 1h#103392
  7. Max 20x: subagents waiting on long Bash commands lose their 5-minute prompt cache and re-write full context — 54% of weekly limit in <24 h, invisible to the user#103309
  8. Failure loop during sleep consumes weekly usage with no output (subagent suspected)#103233
  9. Sleep-mid-response cut-offs trigger an unbounded, uncached full-context regeneration loop#102780
  10. macOS: background agents re-send full context on every DarkWake after lid close (81 requests sent, 29 completed overnight; usage consumed while asleep)#102597
Show all 27Show only the newest 10
  1. Overnight sleep re-writes agent cache per wake-up, drains weekly limit; ambiguous "стоять" stops agents#102269
  2. Prompt cache: CLAUDE.md context is never shared across sessions because it shares a cache segment with the first user prompt#102042
  3. Interactive sessions write the 5-minute prompt cache from session start, while claude -p on the same account writes 1-hour (2.1.296, Team seat, survives /login)#101685
  4. claude -p misses the prompt cache after a turn without thinking (Opus 5.5, Fable 5.1, Sonnet 5.5)#100065
  5. v2.1.290: --resume on opus-5-5/sonnet-5-5 re-writes the whole history to the prompt cache every time (haiku unaffected)#99833
  6. Subagents use the 5-minute prompt cache while the main session uses 1-hour; slow responses cause repeated full-context cache rewrites#99360
  7. Warm 1h prompt cache fully missed (cache_read_input_tokens = 0, 634k tokens re-written) 1.6 min after the previous request, mid tool loop — v2.1.288, Opus 5.5#99096
  8. Prompt cache falls to 7,085-token system floor for 43–58 consecutive calls#98574
  9. Prompt cache drops to 7,085-token system floor across runs of 43–58 calls#98557
  10. Session-invariant context (skill/agent listings, deferred tools, CLAUDE.md) is sent after the first user prompt, so every new session and subagent rewrites it to the prompt cache#98513
  11. Session-specific scratchpad path early in the context stops cache reuse across sessions#97262
  12. Prompt cache never hits after a mid-conversation role:"system" message (resume and hook output), Opus 5.x / Fable#96998
  13. Opus 5.5 / Fable 5.1 in -p mode re-write the whole conversation cache every turn (Sonnet caches fine; totalTokensReminder: off does not help)#96163
  14. Cache entry written with its breakpoint on a trailing role: system message is never reused — with the per-user-turn <total_tokens> reminder on, every tool-less turn re-writes the whole conversation (likely also the root cause of #93490)#96101
  15. Prompt cache misses on every background session: per-session job-dir path contaminates the cacheScope:"org" system-prompt block#94815
  16. Resuming a background subagent misses its prompt cache: messages_changed, no thinking blocks#94728
  17. CLAUDE.md / auto-memory block is never shared across sessions: it sits in messages[0] after the system breakpoint, with no cache_control of its own#94417

Do you work on Claude Code?

Claim its RAP Sheet to see what your users report and answer them in one place.