[FEATURE] /context summary — print only the usage grid, skip MCP tools/Memory/Skills listings

Preflight Checklist

• [x] I have searched existing requests — no open request exists; four prior closed/locked requests for this are referenced in Additional Context below
• [x] This is a single feature request (not multiple features)

Problem Statement

/context prints four sections: the Context Usage grid, then full listings of every MCP tool, every memory file, and every skill. With an ordinary setup (claude.ai connectors + Claude in Chrome + a few skill packs ≈ 50 MCP tools, 30+ skills), that is ~100+ lines of scrollback for what is usually a two-second "how full is my context?" check. The grid — the reason you run the command — immediately scrolls off, and you scroll back up past the listings every time.

There is also a token cost: the command's full output is inserted into the conversation transcript, so each invocation adds several thousand tokens to the Messages category, with the tool/skill listings dominating. Repeatedly checking context usage during a long session therefore consumes meaningful context — ironic for the monitoring command (previously raised in #18056 and #61907).

Verified on v2.1.215: the only recognized argument is all (which expands detail further). Anything else — /context --summary, /context --brief — is silently ignored and the full output prints anyway.

Proposed Solution

A summary argument that prints only the grid plus the "Estimated usage by category" block:

/context summary        (or --summary / --brief)

Argument parsing already exists for all, so this slots in beside it:

• /context — unchanged (full output)
• /context all — unchanged (expanded per-item breakdown)
• /context summary — grid + category totals only; the MCP tools / Memory files / Skills sections are omitted (each already has a dedicated command: /mcp, /memory, /skills)

Alternative Solutions

• Inverting the default would also solve it: bare /context shows the summary, and the listings appear only under /context all — the docs already describe all as "expands the per-item breakdown".
• Current workarounds: read the top block and scroll past the rest; or run /context inside a forked subtask (/subtask /context) so the dump stays out of the main conversation — that works but costs a context fork, and the full output still prints to the terminal.

Priority

Medium - Would be very helpful

Feature Category

Interactive mode (TUI)

Use Case Example

1. Deep in a long session, I want to know how close I am to auto-compaction.
2. I run /context: the grid appears, followed by ~100 lines listing 50 MCP tools, memory files, and 30+ skills.
3. I scroll back up to find the grid, and the transcript just grew by a few thousand tokens.
4. With /context summary, the answer is the last dozen lines on screen, and repeated checks stay cheap enough to run casually.

Additional Context

This has been requested at least four times, and every instance is now closed and locked without the feature landing:

• #15042 (--summary flag) — closed
• #27592 (--compact flag) — auto-closed as duplicate of #18056
• #27596 (collapsible sections) — auto-closed as duplicate of #27592
• #18056 (--brief flag, the canonical issue) — went stale and was auto-closed as not planned

Filing fresh per the auto-lock message on those issues ("please file a new issue and reference this one if it's relevant"). The recurrence suggests steady demand for the same small feature.

Environment: Claude Code v2.1.215, Windows 11.

🤖 Generated with Claude Code