statusline-config

name	statusline-config
description	Configure the Claude Code statusline managed by the statusline-plugin. Use whenever the user wants to add, remove, reorder, recolor, or otherwise change segments shown in their statusline — including the first-time install of the runtime into ~/.claude/settings.json. Use whenever the user invokes /statusline-plugin:configure or asks "change my statusline to …".

statusline-config

You configure the Claude Code statusline rendered by the statusline-plugin runtime. Two artifacts:

1. ~/.claude/statusline-plugin/config.json — schema-driven layout (segments, colors, separator, secondLine). You edit this.
2. ~/.claude/settings.json — Claude Code's own settings; needs a one-time statusLine stanza pointing at the plugin runtime. Check it on first run.

The config file is auto-seeded with a sensible default the first time the runtime executes, so you can usually just read it, modify, and write it back.

Install (run on every /configure)

The plugin's renderer lives at a version-pinned path inside ~/.claude/plugins/cache/<marketplace>/statusline-plugin/<version>/bin/statusline.js — that path goes stale on every plugin upgrade. To survive upgrades, statusLine.command points at a stable wrapper at ~/.claude/statusline-plugin/run that resolves the latest cached version at exec time.

Do this every time the configure skill runs (it is fast, idempotent, and lets new wrapper logic propagate automatically when the plugin updates):

1. Install (or refresh) the wrapper:

   mkdir -p ~/.claude/statusline-plugin
   cp "${CLAUDE_PLUGIN_ROOT}/bin/run" ~/.claude/statusline-plugin/run
   chmod +x ~/.claude/statusline-plugin/run
   

   Always re-copy — the user might have a new plugin version with an updated resolver.

2. Point ~/.claude/settings.json at the wrapper. Read the file (merge — never overwrite). Resolve the user's absolute home dir (echo "$HOME"); settings.json does not expand ~. Then ensure:

   {
     "statusLine": {
       "type": "command",
       "command": "<HOME>/.claude/statusline-plugin/run"
     }
   }
   

   • If statusLine.command is missing: install the stanza above.
   • If statusLine.command matches the old version-pinned shape (*/.claude/plugins/cache/*/statusline-plugin/*/bin/statusline.js): rewrite it to the wrapper path. This is the migration path — once-per-machine, transparent to the user.
   • If statusLine.command already points at ~/.claude/statusline-plugin/run: leave it alone (the wrapper file refresh in step 1 is the only mutation needed).
   • If it points at something else (a hand-rolled command, a different plugin): leave it and tell the user — don't clobber their config.

3. Optional: "padding": 2 adds horizontal padding.

4. The change takes effect on the next prompt — no restart needed.

The wrapper itself has zero dependencies and exits cleanly (status 0, stderr hint) if no cached version is found, so a temporarily empty cache won't break the prompt.

Schema (authoritative)

{
  "$schema": "https://raw.githubusercontent.com/earchibald/statusline-plugin/main/schema/config.schema.json",
  "separator": " | ",
  "segments": [ /* Segment[] — first line */ ],
  "secondLine": {
    "separator": " ",
    "segments": [ /* SecondLineSegment[] — optional second line */ ]
  }
}

The secondLine key is optional. When present, the wrapper script (~/.claude/statusline-with-cache.sh) renders a second line below the main segments using the segment types listed below. The separator defaults to " " (space).

Every segment has a type and shares these optional common fields:

Field	Type	Default	Notes
color	enum	—	black red green yellow blue magenta cyan white gray brightRed brightGreen brightYellow brightBlue brightMagenta brightCyan brightWhite
bold	boolean	false
dim	boolean	false
italic	boolean	false
prefix	string	""	Rendered before the value (e.g. "⎇ ").
suffix	string	""	Rendered after the value.
hideWhenEmpty	boolean	true	Drop the segment entirely if the value resolves to empty.
joinPrev	boolean	false	When true, no separator is inserted before this segment — it concatenates directly to the previous visible segment. Use for tight clusters like Opus 4.7 [high].

Segment types

Type	Type-specific fields	Renders
text	value (string, required)	the literal value
model	format: name (default) | id	model display name (e.g. Opus 4.7)
cwd	format: tilde (default) | basename | full | brief; maxLen; briefDepth (default 1)	working directory. brief abbreviates intermediate components to one char (~/Projects/tts-me-baby/.claude/worktrees/tmb-28 → ~/P/t/.c/w/tmb-28); leading dots are preserved (.claude → .c). briefDepth keeps the last N components full (default 1 = leaf only); falls through to tilde-style output when ≥ total path components; ignored when format ≠ brief. maxLen truncates from the left with ….
git_branch	dirtySuffix (default *)	current branch + suffix when dirty; empty outside a repo
time	format: HH:mm (default) | HH:mm:ss | iso	local clock
tokens	which: total (default) | input | output	session token counter
context	format: percent (default) | remaining_percent | absolute | absolute_percent; scale: auto (default) | raw	context-window usage. auto scales numbers briefly (147001 → 147k, 1000000 → 1M); raw keeps full numbers. Only affects absolute / absolute_percent.
cost	unit: session (default)	session spend in USD, $0.00 shape
session	—	first 8 chars of session_id
output_style	—	active output style name
version	—	Claude Code version
agent	—	active subagent name (empty in main session)
effort	—	thinking-effort level (low / medium / high) when set
claude_ds	—	[DEEPSEEK] badge when CLAUDE_DS=1; hides otherwise

Unknown types render empty (forward-compat) — keep configs valid against the schema above.

Second-line segment types

These are rendered by ~/.claude/statusline-with-cache.sh, not the plugin runtime. They read the session transcript to display prompt-cache cliff information.

Type	Fields	Renders
cache_cliff_1h	count (integer, default 3)	1h[Xk@HH:MM, Yk@HH:MM, …, N more] — next count expiry groups of the 1h cache, grouped by minute, soonest first. Appends N more when groups exist beyond the shown count. Color: green > 30m, yellow > 10m, red ≤ 10m on the soonest cliff.
cache_cliff_largest	prefix (string, default "largest ")	<prefix>Xk@HH:MM — the single group with the most tokens across all live 1h blocks. Independent of cache_cliff_1h. Color follows the same thresholds.

Methodology: A "group" is all 1h cache blocks whose ts + 3600 falls in the same wall-clock minute. Within a group, the maximum ephemeral_1h_input_tokens value is taken (the field reports total cache size, not incremental additions — summing would double-count overlapping cache blocks). Only blocks with ts > now − 3600 are included (live window). cache_cliff_largest picks the globally largest group regardless of its position in the expiry order.

Examples:

"1h cache cliff with 3 examples and largest" →

"secondLine": {
  "segments": [
    { "type": "cache_cliff_1h", "count": 3 },
    { "type": "cache_cliff_largest" }
  ]
}

Renders: 1h[15k@23:42, 2k@23:44, 4k@23:45, 27 more] largest 191k@00:17

"largest: prefix, then 4 examples" →

"secondLine": {
  "segments": [
    { "type": "cache_cliff_largest", "prefix": "largest: " },
    { "type": "cache_cliff_1h", "count": 4 }
  ]
}

Renders: largest: 191k@00:17 1h[15k@23:42, 2k@23:44, 4k@23:45, 3k@23:46, 26 more]

Edit recipes

Be additive when the user's request is ambiguous: ask which slot the new segment goes in (left, right, replace) only if there's real ambiguity. Otherwise infer from intent — e.g. "add the time" → append to the segments array.

"Show me model and current directory only."

{
  "separator": " | ",
  "segments": [
    { "type": "model", "format": "name", "color": "cyan", "bold": true },
    { "type": "cwd",   "format": "tilde", "color": "blue" }
  ]
}

"Add cost on the right." Append a cost segment to the end:

{ "type": "cost", "unit": "session", "color": "gray" }

"Add a clock."

{ "type": "time", "format": "HH:mm", "color": "gray" }

"Make the model name green." Find the existing model segment, set "color": "green". Don't add a duplicate.

"Show context usage as a percentage."

{ "type": "context", "format": "percent", "color": "yellow", "prefix": "ctx " }

"Show effort right after the model with no separator" — e.g. Opus 4.7 [high]. Add an effort segment with joinPrev: true, plus a leading-space prefix and bracket suffix:

{ "type": "effort", "joinPrev": true, "prefix": " [", "suffix": "]" }

"Show a brief cwd" — abbreviates parents, keeps the last folder full.

{ "type": "cwd", "format": "brief", "color": "blue" }

Renders ~/P/t/.c/w/tmb-28 for ~/Projects/tts-me-baby/.claude/worktrees/tmb-28.

"Brief cwd, but keep the last two folders readable." Set briefDepth to 2:

{ "type": "cwd", "format": "brief", "briefDepth": 2, "color": "blue" }

Renders ~/P/t/.c/worktrees/tmb-28 for the same path. briefDepth ≥ total path components falls through to tilde-style output (no abbreviation).

"Show context as absolute(percent)" — brief by default, e.g. 147k/1M (15%).

{ "type": "context", "format": "absolute_percent", "color": "yellow" }

"Show full numbers, not the k/M scale."

{ "type": "context", "format": "absolute_percent", "scale": "raw", "color": "yellow" }

"I want a tag in front." Use a text segment first:

{ "type": "text", "value": "[claude]", "color": "magenta", "bold": true }

Workflow

1. Read ~/.claude/statusline-plugin/config.json. If missing, write the default below first (the runtime will create it lazily, but creating it now lets the user see what you're modifying).
2. Apply the user's change minimally — preserve unrelated segments, ordering, separator, and field values.
3. Validate: every segment has a known type, all color values are from the enum, text.value is present.
4. Write the file with 2-space indentation and a trailing newline.
5. Show the user the diff in plain English: "Added a cost segment after the git branch, gray." Don't dump JSON unless they ask.
6. Mention that the change takes effect on the next prompt.

Default config (use when seeding)

{
  "$schema": "https://raw.githubusercontent.com/earchibald/statusline-plugin/main/schema/config.schema.json",
  "separator": " | ",
  "segments": [
    { "type": "model", "format": "name", "color": "cyan", "bold": true },
    { "type": "cwd", "format": "tilde", "color": "blue" },
    { "type": "git_branch", "color": "magenta", "prefix": "⎇ ", "dirtySuffix": "*" },
    { "type": "context", "format": "percent", "color": "yellow", "prefix": "ctx " },
    { "type": "cost", "unit": "session", "color": "gray" }
  ]
}

Don'ts

• Don't introduce segment types not in the schema — they render empty.
• Don't put prefix/suffix strings inside the value of a text segment when the user wanted a real segment (e.g. don't fake a git branch via text — use git_branch).
• Don't change separator unless the user asked for it; it touches every segment.
• Don't write the config to anywhere other than ~/.claude/statusline-plugin/config.json.
• Don't restart Claude Code or instruct the user to — the runtime is invoked fresh on every render.