[FEATURE] /context should default to graph only; /context all for full output
This has been filed at least four times. All previously closed without resolution:
- #18056 - /context --brief (closed)
- #19595 - /context -m flag (closed)
- #27592 - /context --compact (closed)
- #48798 - /context is too verbose (closed)
The problem has not been addressed and previous issues have been ignored or autoclosed.
Problem
The /context command exists to answer one question: how full is my context window?
The current output buries that answer under expanded listings of MCP tools, custom
agents, memory files, and skills - none of which the user asked for. Worse, that
output adds thousands of tokens to the session, so running /context to monitor
token usage actively makes the problem worse.
Proposal
Split into two modes:
/context - shows only the usage graph and category summary:
Context Usage
⛀ ⛁ ⛁ ⛁ ⛁ ⛁ ⛀ ⛁ ⛀ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ Opus 4.8 (1M context)
⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ claude-opus-4-8[1m]
⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ 35.8k/1m tokens (4%)
...
⛁ System prompt: 2.8k tokens (0.3%)
⛁ System tools: 20.2k tokens (2.0%)
⛁ Custom agents: 6.1k tokens (0.6%)
⛁ Memory files: 2.6k tokens (0.3%)
⛁ Skills: 4k tokens (0.4%)
⛁ Messages: 70 tokens (0.0%)
⛶ Free space: 964.2k (96.4%)
/context all - the current full output including expanded MCP tools,
agent listings, memory files, and skills. Those are also individually accessible
via /mcp, /agents, /memory, and /skills.
This is a low-risk change: the current behavior simply becomes /context all,
and the default gets tightened to what users actually need.
Why this keeps getting filed
Each prior issue was closed by automation - marked duplicate or stale - before
the feature shipped. The demand is real and consistent across all five filings.
Users do care about context usage - that is exactly why they run /context. But
what they need in the moment is a quick, high-level read: how full is the window
and where is the pressure coming from. The current command makes that harder, not
easier, by burying the answer in a wall of detail.
4 Comments
This is still desired, context as-is today is too noisy.
This is still desired, context as-is today is too noisy.
This is still desired, context as-is today is too noisy.
Two updates:
/context allalready exists, and in fullscreen mode the plain/contextview collapses the per-item listings (MCP tools, agents, skills) into one-line summaries. Also, since a recent release the graph is no longer sent to the model, so running/contextdoesn't add thousands of tokens to your session; Claude only receives a compact table. Outside fullscreen mode the default is still the full listing, so leaving this open for the compact-by-default part. Changelog: https://github.com/anthropics/claude-code/blob/main/CHANGELOG.md🤖 Generated with Claude Code