[FEATURE] Request: Persist per-turn token/time usage as text in the session transcript

Open 💬 0 comments Opened Jul 6, 2026 by 11pyo

Preflight Checklist

  • [x] I have searched existing requests and this feature hasn't been requested yet
  • [x] This is a single feature request (not multiple features)

Problem Statement

Problem

During a session, Claude Code shows a live indicator of elapsed time and token usage for the current turn (e.g. "5m 1s · 1.9k tokens"). This is currently ephemeral UI-only — once the turn completes, that data isn't preserved as readable text anywhere in the session transcript or history.

Feature Request

Persist this per-turn usage as a small text line appended to the transcript when each turn/task completes (e.g. "⏱ 5m 1s · 1.9k tokens"), so it remains visible in scrollback and exported session logs.

Motivation

  • Users currently have no lasting record of which prompts/tasks were token-heavy vs. efficient. Seeing this data accumulate over a session (and across sessions) builds concrete, personal intuition for what efficient prompting looks like — vague requests that trigger broad exploration vs. precise requests that resolve quickly.
  • At the scale of Claude Code's user base, better-informed prompting habits could meaningfully reduce aggregate token/compute consumption in Anthropic's infrastructure — which is a direct cost lever, not just a UX nicety.
  • This is qualitative/observational data users can act on themselves, without needing a separate analytics dashboard.

Possible implementation

  • Append a short usage line to the transcript on turn completion (toggleable in settings for users who find it noisy).
  • Optionally include it in exported/saved session logs by default.

Proposed Solution

Append a short usage summary line (e.g. "⏱ 5m 1s · 1.9k tokens · 8 layer thoughts · with 3agents ···") to the transcript automatically when each turn/task completes, so it persists in scrollback and exported session logs instead of disappearing once the turn ends. This could be toggleable in settings for users who prefer a cleaner transcript, and optionally included by default in exported/saved session logs.

Alternative Solutions

_No response_

Priority

High - Significant impact on productivity

Feature Category

Interactive mode (TUI)

Use Case Example

As a user working through a long multi-step session (debugging, research, or iterative code changes), I'd like to be able to look back at my own session and see roughly how much time/tokens each turn took.

Right now, while a turn is running, Claude Code shows a live indicator like "5m 1s · 1.9k tokens · 8 layer thoughts · with 3agents ···" — but once the turn finishes, that number disappears entirely. If I want to reflect on which of my prompts triggered a lot of work versus which resolved quickly and efficiently, I have no record to look back on within the session itself.

A simple fix: append that same number as a small persisted text line (e.g. "⏱ 5m 1s · 1.9k tokens") to the transcript once the turn completes, so it stays visible in scrollback and in exported session logs.

Additional Context

This is a minimal, low-risk request: it doesn't ask for any new telemetry or additional data access — it's simply persisting a number that Claude Code is already computing and displaying live during the turn, just not keeping afterward.

The main motivation is user-facing self-awareness: seeing this accumulate over a session naturally helps users build intuition for efficient prompting (e.g., noticing a vague request triggered broad exploration versus a precise one that resolved in a couple of steps) — similar to how cost/usage visibility in other developer tools helps people self-optimize.

Happy to see this behind a settings toggle for users who'd prefer a cleaner transcript.

View original on GitHub ↗