{"version":"2.1.282","anchor":"new-request-replay-recording-in-transcripts-off-unless-teng","canonical_anchor":"new-request-replay-recording-in-transcripts-off-unless-teng","heading":"Claude Code can record the API requests it sends into session transcripts","tier":"internal","area":"Sessions","url":"https:\/\/changelogs.core-directive.com\/v\/2.1.282\/e\/new-request-replay-recording-in-transcripts-off-unless-teng","release_url":"https:\/\/changelogs.core-directive.com\/v\/2.1.282","markdown":"### Claude Code can record the API requests it sends into session transcripts\n\nNew, off-by-default recording writes each API request's system prompt, tools and message changes into the session transcript file\n\n**Unclear.** It is not known what reads these records back, or what value of `CLAUDE_CODE_ELEGANT_MEADOW` turns recording on.\n\n**What**\n\nA session transcript is the file on disk where Claude Code keeps the record of a conversation. Claude Code can now write a record of each request it sends to the model into that file. Requests from a subagent (a helper Claude runs to do part of a task) go into that subagent's own file. The records hold:\n\n- the system prompt and tool definitions, stored once and referred to by a fingerprint\n\n- how each request differs from the previous one in the same conversation\n\n- where caching markers sit, and a fingerprint of the whole request\n\nAnything added through `CLAUDE_CODE_EXTRA_BODY` is left out and listed as withheld. Nothing is recorded when the system prompt and tools are too big for one record, when the session is mirrored remotely, or when account memory text cannot be cleanly cut out. Only requests from the main conversation and subagents are recorded.\n\nRecording is off unless it is switched on remotely or through the `CLAUDE_CODE_ELEGANT_MEADOW` environment variable, which is checked first.\n\n**Why**\n\nWhen this recording is switched on, transcript files grow and hold more of what was sent to the model, including the system prompt, so they become larger and more sensitive to share.\n\n- Flag `tengu_elegant_meadow`: Not enough to say (read for one account on one subscription tier against v2.1.282; this account: no value returned, anonymous baseline: no value returned, compiled default: off) These values were read against a different version of Claude Code, so treat them as the nearest reading available instead of one taken on this release.\n- Flag `tengu_humble_hammock`: Off by default, switched on for this account (read for one account on one subscription tier against v2.1.282; this account: on, anonymous baseline: on, compiled default: off) These values were read against a different version of Claude Code, so treat them as the nearest reading available instead of one taken on this release.\n- Area: Sessions\n- Tier: Under the hood\n- Useful: 4\/5\n- Signal: 4\/5"}