Feature request: standalone /contextSize command, and record the context variant in session transcripts

Status Open
Maintainer reply None cached
Activity 0 comments · opened Aug 31, 2026

Feature request: a standalone /contextSize command (or similar), independent of /model, completing the trio with /effort.

Today the context window rides on the model string (opus[1m], fable[1m]): one token in the /model picker selects two orthogonal things at once — which model, and how much context. /effort already demonstrates the right shape for an orthogonal dial.

Why it matters beyond ergonomics — observability: the chosen context variant is not recorded anywhere machine-readable. Session transcripts record message.model as the bare id (claude-opus-5) with no [1m] marker or context field, so external tooling that wants to answer "which of my sessions run 1M context?" has to scrape /model command echoes out of transcript text and fall back to guessing from settings defaults. A first-class context-size setting — with the choice reflected in the transcript records (or in message.usage) — would make it scriptable and auditable the way model and effort already are.

Ask:

  1. /contextSize (session-scoped, like /model and /effort), with the equivalent settings key.
  2. Record the effective context size per assistant turn in the session .jsonl (field on the record, or in usage), so the choice is observable after the fact.

Context: building local tooling that aggregates per-session model usage across a machine; the context variant is the one dimension of the model choice that leaves no trace.

View original on GitHub ↗