/context in Verboo Code: how to see what is filling your context window

The agent starts forgetting what you agreed on, answers get slower, and you can't tell whether the cause is the long conversation, a heavy MCP server or an AGENTS.md that grew too big. The /context command in Verboo Code answers that in one step: it shows how much of the context window is in use and why, category by category. This guide shows how to read the output and what to do with each line.

What does /context show in Verboo Code?

A colored grid with current context window usage, the total in tokens and as a percentage, and a list of what is taking up that space. The command description in the source is literal: "Visualize current context usage as a colored grid".

/context

The first line shows the model and the total, in the format model · used/window tokens (N%). Right below comes the Estimated usage by category block, then the detail for each source: MCP, agents, instruction files and skills.

The number is what the model actually receives, not the raw history on screen. Before counting, the command applies the same transforms the CLI applies before calling the API: it drops everything before the last compaction and shrinks old tool results. That is why the total can be smaller than the conversation looks.

What does each category mean?

Each line is a source of tokens. Some you control directly, others you don't. The table shows the names exactly as they appear on screen and where to act:

Category	What it is	Where to act
System prompt	The CLI's base instructions	Not adjustable day to day
System tools	Definitions of the built-in tools (read, edit, run command)	Not adjustable day to day
MCP tools	Tools from connected MCP servers	/mcp
Custom agents	Available custom agents	/agents
Memory files	Loaded AGENTS.md, CLAUDE.md and rules	/memory
Skills	Headers of installed skills	/skills
Messages	The conversation: your messages, replies and tool results	/compact or /clear
Autocompact buffer	Headroom reserved for automatic compaction	/config (Auto-compact)
Free space	What still fits	Nothing to do

Everything up to Skills is fixed cost: it goes into every call, even before you type the first message. Messages is what grows during the session.

Why do some lines show N/A?

Those are the MCP tools (deferred) and System tools (deferred) categories. When tool search is on, these tools stay out of the context and only load when the agent needs them. /context shows their size for visibility, but they don't count toward the total, so the percentage shows as N/A. In the MCP detail, the heading becomes MCP tools · /mcp (loaded on-demand) and the list splits into Loaded and Available.

What is the Autocompact buffer?

It is reserved space, not used space. With Auto-compact on, Verboo Code compacts the conversation when usage reaches that headroom, and the grid shows the reserve so you can see how far you are from it. With Auto-compact off, the line becomes Compact buffer, a smaller reserve that keeps room for a manual /compact.

How do I find what is taking up space?

Look at the largest fixed category first. It is sent on every call, so trimming there pays off more than compacting the conversation. Below the category block, /context lists each item with its cost:

Memory files · /memory
└ AGENTS.md: 6.2k tokens
└ docs/architecture.md: 4.8k tokens

The example above is illustrative, but the format comes from the source: each section shows the command that manages that source next to its heading. If the largest number is a file imported with @ in AGENTS.md, remove the import and keep only what the agent would get wrong without reading. If it is an MCP server you don't use in this project, turn it off in /mcp.

What do the warnings at the end of /context mean?

The command ends with suggestions generated from the session numbers. They only show up when a threshold defined in the source is crossed:

Warning	When it shows	What to do
Context is N% full	Usage at 80% or more	Run /compact now, saying what to keep
Bash results using ... tokens	Command output above 15% of the window and 10k tokens	Ask it to filter with head, tail or grep
Read results using ... tokens or File reads using ... tokens	File reads above 15% of the window (or read results above 5%), always over 10k tokens	Read only the part you need, without re-reading the whole file
Memory files using ... tokens	Instruction files above 5% of the window and 5k tokens	Trim them with /memory
Autocompact is disabled	Auto-compact off and usage between 50% and 80%	Turn it on in /config or compact by hand

The 80% warning depends on your settings. With Auto-compact on, it warns that automatic compaction will fire soon and discard older messages, and suggests running /compact first so you decide what stays. With it off, it suggests /compact or turning it back on in /config.

How do I free context after reading /context?

In the order of the diagram: fixed cost first, then the conversation.

1. Trim the fixed cost. /memory to slim down instructions, /mcp to turn off unused servers, /skills to review skills.
2. Summarize the conversation when Messages dominates. Say what must not be lost:

   /compact keep the changed files and the decision to stay on API v1

3. Start over with /clear if the next task doesn't depend on the previous one. Instruction files stay on disk and are read again.

Run /context again at the end to check how much you freed. The step by step for compaction, including the Auto-compact environment variables, is in the /compact in Verboo Code guide.

Can I check context usage without opening a session?

Yes. In non-interactive mode /context returns the same breakdown as text, in Markdown tables, instead of the colored grid:

verboo -p "/context"

To measure the most recent conversation in the current folder, combine it with -c, which continues the last session:

verboo -c -p "/context"

The output starts with ## Context Usage, has the **Model:** and **Tokens:** lines and then tables by category, MCP, agents, instruction files and skills. It is plain text, so you can save it to a file and compare the fixed cost before and after editing AGENTS.md.

How does this fit with the other commands?

• /memory edits the files listed under Memory files.
• The MCP guide shows how to add and remove the servers listed under MCP tools.
• /plan gets you to the code with fewer back-and-forths, which keeps Messages from ballooning.