Output Style

Output style

slipstream ships an output style, output-styles/slipstream.md, tuned for terse, high-signal, token-lean responses. Day-to-day use under this style spends fewer tokens per turn because Claude stops restating the question, stops padding, and stops pasting back unchanged code.

What it changes

The style is a normal Claude Code output style with frontmatter (name, description) and a body of rules. The rules fall into four groups:

• Voice. Answer the question, do not restate it, no preamble, no closing pleasantries. Short declarative sentences. One code block or one short list over three paragraphs.
• Retrieval discipline. Orient with sp_map before reading files, pull one symbol with sp_symbol or a window with sp_lines, use sp_search to find things, check sp_budget when a turn has read a lot.
• Memory discipline. Persist durable decisions with sp_remember; recall a prior decision with sp_recall before acting on it.
• Output. When changing code, state the file and symbol and why in one line each; do not paste back unchanged surrounding code. When running a gate, report the command and the result, nothing more.

How to switch to it

/output-style slipstream

That sets the active style for the session. Switch back with /output-style default (or any other style). When the style is active, the statusline can show it in the skill segment.

When to use it, and when not

Use it for routine building, refactors, and long sessions where token spend matters. It pairs naturally with the MCP tools: the retrieval rules tell Claude to reach for sp_symbol instead of Read.

Do not use it when you want Claude to explain its reasoning at length, teach a concept, or write prose documentation. The style suppresses exactly the elaboration those tasks need. It is a working style, not a writing style.

Relationship to the other token features

%%{init: {'theme':'base','themeVariables':{'primaryColor':'#0d1117','primaryTextColor':'#f5f7fa','primaryBorderColor':'#38bdf8','lineColor':'#22d3ee','fontFamily':'monospace'}}}%%
flowchart LR
  Style[output style] -->|fewer tokens out| Spend[lower per-turn spend]
  MCP[MCP tools] -->|fewer tokens in| Spend
  Recall[smart recall] -->|smaller session start| Spend
  Spend --> Long[longer useful sessions]

Loading

The output style is the smallest of the token features and the easiest to reverse; it trims the response side, while the MCP tools and smart recall trim the input side.

See also

• Statusline for where the active style shows.
• MCP tools for the tools the style tells Claude to prefer.
• Token efficiency for the input-side savings.

---

SarmaLinux . sarmalinux.com . Repository