Follow Discord
Sweep 03 Nov 2026 · 22:27Z Build v2.1.318 536 read Stable v2.1.311 Latest v2.1.318 Next v2.1.318 Feeds RSS JSON llms.txt llms-full.txt Unofficial

Claude Code v2.1.318 ·

Incremental request reuse now keeps content per subagent and can split near the tools list

The off-by-default request-reuse mode now keeps a saved request for each subagent and can place a block boundary near the end of the tools list

Group of 3 You'll notice No documentation found Improvements
JSON All of v2.1.318
You'll noticeTier: how much it should matter to you
2Useful: my rating, 1 to 5
3Signal: worth watching, 1 to 5
Prompt CachingArea: what it touches
ImprovementsKind: in v2.1.318,
ImprovementsSection of the release

What

The request-reuse mode set by CLAUDE_CODE_INCREMENTAL_REQUEST_BODY reuses content from earlier requests to the model. It now does more.

  • Per-subagent reuse. A subagent is a helper agent Claude starts for part of a job. Before, only the main conversation's saved request was kept. Now a saved request is kept for each subagent, up to a count set remotely through tengu_vast_frost; a value of 0 clears them. If the server rejects a reused request because the bytes sent were wrong, that subagent's saved entry is cleared. Requests that reuse a subagent's saved content are tagged keptForSubagent. This applies only with CLAUDE_CODE_INCREMENTAL_REQUEST_BODY set to 2 and Anthropic's own API. Nothing has been read about how tengu_vast_frost is set.
  • Split near the tools list. When the request is broken into compressed blocks, a new cutNearToolsEnd option, controlled by a switch, can place a boundary near the end of the tools section. Before, boundaries were only placed at the messages and system sections. The code that finds saved content was also made more general; it used to match on the exact stored copy.
  • Request details sent to Anthropic now include whether the tools-end split was on and whether subagent content was reused, when set.

Why

Sessions that use many subagents could avoid rebuilding and resending the same large request again and again, which may also help prompt caching. The mode stays off unless you turn it on.

Read from

See this entry in the whole of v2.1.318 →

Feedback