Statusline Configuration Skill

Statusline Configuration

Overview

Claude Code supports a custom status line via a shell script that receives JSON session data on stdin and outputs text to stdout. The script runs after each assistant message, after /compact, on permission mode changes, and on vim mode toggles (debounced at 300ms). Supports multiple lines, ANSI colors, and OSC 8 hyperlinks.

Setup

Configure in ~/.claude/settings.json (or project settings):

{
  "statusLine": {
    "type": "command",
    "command": "~/.claude/statusline.sh"
  }
}

Optional fields:

padding (int): extra horizontal spacing in characters (default: 0)
refreshInterval (int): re-run every N seconds in addition to event-driven updates (min: 1)
hideVimModeIndicator (bool): suppress built-in -- INSERT -- when script renders vim mode

Alternatively, use an inline command:

{
  "statusLine": {
    "type": "command",
    "command": "jq -r '\"[\\(.model.display_name)] \\(.context_window.used_percentage // 0)% context\"'"
  }
}

Or use /statusline <description> to have Claude generate one automatically.

Available JSON Fields

The script receives JSON on stdin with these fields:

| Field | Description | |-------|-------------| | model.id, model.display_name | Model identifier and display name | | cwd, workspace.current_dir | Current working directory | | workspace.project_dir | Directory where Claude Code was launched | | workspace.added_dirs | Additional directories added via /add-dir | | workspace.git_worktree | Git worktree name (if in linked worktree) | | workspace.repo.host/owner/name | Repository identity from origin remote | | cost.total_cost_usd | Estimated session cost in USD | | cost.total_duration_ms | Wall-clock time since session start | | cost.total_api_duration_ms | Time spent waiting for API responses | | cost.total_lines_added/removed | Lines of code changed | | context_window.used_percentage | Percentage of context window used | | context_window.remaining_percentage | Percentage remaining | | context_window.context_window_size | Max context window (200000 or 1000000) | | context_window.total_input_tokens | Input tokens in current context | | context_window.total_output_tokens | Output tokens from last response | | exceeds_200k_tokens | Whether tokens exceed 200k threshold | | effort.level | Reasoning effort (low/medium/high/xhigh/max) | | thinking.enabled | Whether extended thinking is enabled | | rate_limits.five_hour.used_percentage | 5-hour rate limit usage (Pro/Max only) | | rate_limits.seven_day.used_percentage | 7-day rate limit usage (Pro/Max only) | | session_id | Unique session identifier | | session_name | Custom name from --name or /rename | | version | Claude Code version | | vim.mode | Current vim mode (NORMAL/INSERT/VISUAL) | | agent.name | Agent name if using --agent | | pr.number, pr.url, pr.review_state | Open PR info for current branch | | worktree.name/path/branch | Worktree info during --worktree sessions |

Fields marked may be absent or null - use jq fallbacks like // 0 or // empty.

No terminal width in the JSON. There is no columns/width/lines field here. To lay out responsively (and wrap long lines), read the COLUMNS/LINES environment variables instead — see Responsive Width-Aware Wrapping.

Reference Script (Git + jj — Dual VCS)

Full-featured, width-aware script showing both jj and git status independently (both appear for colocated repos), plus context-window usage. It ends with a greedy segment packer (see Responsive Width-Aware Wrapping) so a long line wraps cleanly at segment boundaries on narrow terminals while rendering identically on wide ones:

#!/bin/bash

input=$(cat)

model_name=$(echo "$input" | jq -r '.model.display_name')
current_dir=$(echo "$input" | jq -r '.workspace.current_dir')
pct=$(echo "$input" | jq -r '.context_window.used_percentage // 0' | cut -d. -f1)

# Build VCS info — jj and git are checked independently (both show for colocated repos).
# Each discrete block is pushed as one breakable segment so the width-aware packer
# at the bottom can wrap cleanly at block boundaries on narrow terminals.
vcs_segments=()

# jj repository check
if jj root -R "$current_dir" --ignore-working-copy >/dev/null 2>&1; then
    # NB: every field needs a non-empty placeholder — IFS=tab read collapses
    # consecutive tabs (tab is IFS whitespace), so empty fields would shift
    # the columns.
    jj_data=$(jj --no-pager --ignore-working-copy -R "$current_dir" log --no-graph -r @ -T 'if(local_bookmarks, local_bookmarks.join(","), "-") ++ "\t" ++ change_id.short(8) ++ "\t" ++ if(conflict, "conflict", "-") ++ "\t" ++ commit_id' 2>/dev/null)

    if [ -n "$jj_data" ]; then
        IFS=$'\t' read -r bookmarks change_id conflict_status commit_hash <<< "$jj_data"
        if [ "$bookmarks" = "-" ]; then bookmarks=""; fi
        if [ "$conflict_status" = "-" ]; then conflict_status=""; fi

        modified_count=$(jj --no-pager --ignore-working-copy -R "$current_dir" diff --summary 2>/dev/null | wc -l | tr -d ' ')

        status_parts=()
        if [ -n "$conflict_status" ]; then
            status_parts+=("conflict")
        fi
        if [ "$modified_count" -gt 0 ]; then
            status_parts+=("${modified_count} modified")
        fi

        # Ahead/behind remote tracking (works in colocated repos via git)
        if [ -n "$bookmarks" ] && [ -n "$commit_hash" ]; then
            first_bookmark=$(echo "$bookmarks" | cut -d',' -f1)
            if git -C "$current_dir" rev-parse --verify "origin/$first_bookmark" >/dev/null 2>&1; then
                remote_ref=$(git -C "$current_dir" rev-parse "origin/$first_bookmark" 2>/dev/null)
                ahead=$(git -C "$current_dir" rev-list --count "$remote_ref..$commit_hash" 2>/dev/null)
                behind=$(git -C "$current_dir" rev-list --count "$commit_hash..$remote_ref" 2>/dev/null)
                if [ "$ahead" -gt 0 ]; then status_parts+=("${ahead} ahead"); fi
                if [ "$behind" -gt 0 ]; then status_parts+=("${behind} behind"); fi
            fi
        fi

        if [ -n "$bookmarks" ]; then
            display="${bookmarks} @ ${change_id}"
        else
            display="@ ${change_id}"
        fi

        if [ ${#status_parts[@]} -eq 0 ]; then
            vcs_segments+=("[jj ${display}]")
        else
            status=$(printf '%s, ' "${status_parts[@]}"); status=${status%, }
            vcs_segments+=("[jj ${display}: ${status}]")
        fi
    fi
fi

# git repository check (independent — shows alongside jj for colocated repos)
if git -C "$current_dir" rev-parse --is-inside-work-tree >/dev/null 2>&1; then
    branch=$(git -C "$current_dir" branch --show-current 2>/dev/null)
    if [ -z "$branch" ]; then
        # Detached HEAD (git checkout <hash>, bisect, jj-colocated repos):
        # --show-current prints nothing, so fall back to the short hash.
        detached=$(git -C "$current_dir" rev-parse --short HEAD 2>/dev/null)
        [ -n "$detached" ] && branch="detached @ ${detached}"
    fi
    if [ -n "$branch" ]; then
        status_parts=()

        staged_count=$(git -C "$current_dir" diff --cached --numstat 2>/dev/null | wc -l | tr -d ' ')
        if [ "$staged_count" -gt 0 ]; then status_parts+=("${staged_count} staged"); fi

        modified_count=$(git -C "$current_dir" diff --numstat 2>/dev/null | wc -l | tr -d ' ')
        if [ "$modified_count" -gt 0 ]; then status_parts+=("${modified_count} modified"); fi

        untracked_count=$(git -C "$current_dir" ls-files --others --exclude-standard 2>/dev/null | wc -l | tr -d ' ')
        if [ "$untracked_count" -gt 0 ]; then status_parts+=("${untracked_count} untracked"); fi

        upstream=$(git -C "$current_dir" rev-parse --abbrev-ref @{upstream} 2>/dev/null)
        if [ -n "$upstream" ]; then
            ahead=$(git -C "$current_dir" rev-list --count @{upstream}..HEAD 2>/dev/null)
            behind=$(git -C "$current_dir" rev-list --count HEAD..@{upstream} 2>/dev/null)
            if [ "$ahead" -gt 0 ]; then status_parts+=("${ahead} ahead"); fi
            if [ "$behind" -gt 0 ]; then status_parts+=("${behind} behind"); fi
        fi

        if [ ${#status_parts[@]} -eq 0 ]; then
            vcs_segments+=("[git ${branch}]")
        else
            status=$(printf '%s, ' "${status_parts[@]}"); status=${status%, }
            vcs_segments+=("[git ${branch}: ${status}]")
        fi
    fi
fi

# jj ⇄ git sync check (colocated repos): compare git's actual HEAD with jj's
# last-imported view of it (git_head()). They diverge when git moves without a
# jj command running afterward — and the jj calls above use
# --ignore-working-copy precisely so this script never triggers the re-import.
if [ -n "$jj_data" ]; then
    jj_git_head=$(jj --no-pager --ignore-working-copy -R "$current_dir" log --no-graph -r 'git_head()' -T 'commit_id' 2>/dev/null)
    git_head=$(git -C "$current_dir" rev-parse HEAD 2>/dev/null)
    if [ -n "$jj_git_head" ] && [ -n "$git_head" ] && [ "$jj_git_head" != "$git_head" ]; then
        git_ahead=$(git -C "$current_dir" rev-list --count "$jj_git_head..$git_head" 2>/dev/null || echo 0)
        jj_ahead=$(git -C "$current_dir" rev-list --count "$git_head..$jj_git_head" 2>/dev/null || echo 0)
        sync_parts=()
        [ "${git_ahead:-0}" -gt 0 ] && sync_parts+=("git +${git_ahead}")
        [ "${jj_ahead:-0}" -gt 0 ] && sync_parts+=("jj +${jj_ahead}")
        if [ ${#sync_parts[@]} -gt 0 ]; then
            sync=$(printf '%s, ' "${sync_parts[@]}"); sync=${sync%, }
            vcs_segments+=("[jj ⇄ git out of sync: ${sync}]")
        else
            vcs_segments+=("[jj ⇄ git out of sync]")
        fi
    fi
fi

# --- Width-aware greedy segment packer ---------------------------------------
# Each entry in segs[] is one breakable unit; seps[i] is the separator placed
# before segs[i] when it shares a line with the previous segment. A segment
# that wraps to a new line starts that line with no leading separator. On a
# wide terminal everything packs onto one line and reproduces the original
# "model | dir [jj ...] [git ...] | ctx N%" layout byte-for-byte; on a narrow
# terminal it breaks at segment boundaries instead of overflowing.
#
# Width source: Claude Code (v2.1.153+) exports COLUMNS before running this
# script; stdout is captured (not a TTY), so tput/stty cannot detect width.
cols=${COLUMNS:-80}            # 80 = fallback for older Claude Code / non-CC shells
budget=$((cols - 2))           # small margin; built-in padding width is undocumented
[ "$budget" -lt 10 ] && budget=10

segs=("$model_name" "$current_dir")
seps=("" " | ")
for seg in "${vcs_segments[@]}"; do
    segs+=("$seg"); seps+=(" ")
done
segs+=("ctx ${pct}%"); seps+=(" | ")

# NB: no ANSI/OSC 8 escapes are emitted here, so ${#seg} == display width.
# If color/hyperlinks are added later, measure an escape-stripped shadow copy
# (sed -E 's/\x1b\[[0-9;]*m//g') so segments don't wrap early.
line=""; linelen=0; out=()
for i in "${!segs[@]}"; do
    seg="${segs[$i]}"
    [ -z "$seg" ] && continue
    # A single segment wider than the line (usually a long current_dir) can't
    # break at a boundary — truncate it, keeping the tail, with a leading "…".
    if [ "${#seg}" -gt "$budget" ]; then
        seg="…${seg: -$((budget - 1))}"
    fi
    slen=${#seg}
    sep="${seps[$i]}"; seplen=${#sep}
    if [ -z "$line" ]; then
        line="$seg"; linelen=$slen
    elif [ $((linelen + seplen + slen)) -le "$budget" ]; then
        line+="${sep}${seg}"; linelen=$((linelen + seplen + slen))
    else
        out+=("$line"); line="$seg"; linelen=$slen   # wrap to next line
    fi
done
[ -n "$line" ] && out+=("$line")
printf '%s\n' "${out[@]}"

Output Examples

On a terminal wide enough to fit the line, everything packs onto one row (identical to a non-wrapping script):

# Colocated jj+git repo (both shown)
Opus 4.6 | /path/to/project [jj feature @ znnuytsz: 1 modified, 2 ahead] [git feature: 1 untracked] | ctx 12%
# jj-only (non-colocated)
Opus 4.6 | /path/to/project [jj @ uoylmlmx] | ctx 12%
# jj with conflict and remote tracking
Opus 4.6 | /path/to/project [jj main @ kntqzsqt: conflict, 3 modified, 1 behind] | ctx 12%
# git-only repo
Opus 4.6 | /path/to/project [git master: 2 staged, 1 modified] | ctx 12%
# Detached HEAD (checkout by hash, bisect, jj-colocated)
Opus 4.6 | /path/to/project [git detached @ d7ead68: 1 modified] | ctx 12%
# Colocated repo where git moved without jj noticing (e.g. direct git commit)
Opus 4.6 | /path/to/project [jj @ snzksmsp] [git detached @ 76137d1] [jj ⇄ git out of sync: git +1] | ctx 12%
# No VCS
Opus 4.6 | /tmp | ctx 12%

On a narrow terminal the same line wraps at segment boundaries instead of overflowing (the git-only repo above at COLUMNS=40):

Opus 4.6 | /path/to/project
[git master: 2 staged, 1 modified]
ctx 12%

Design Notes

Dual VCS: Both jj and git blocks run independently; colocated repos show both
jj ahead/behind: Uses commit_id (renders as full git commit hash) then git rev-list --count against the first bookmark's remote tracking ref
Detached HEAD fallback: git branch --show-current prints nothing when detached, so the git segment falls back to detached @ <short-hash>. Without this, jj-colocated repos would never show the git segment — jj keeps git permanently detached
jj ⇄ git sync check: compares git rev-parse HEAD with jj's last-imported git_head(); the segment appears only when they diverge, with directional counts (git +N / jj +N). Divergence happens when git commands run without a jj command following — and stays visible because this script's --ignore-working-copy calls never trigger jj's auto-import
IFS tab gotcha: every jj template field needs a non-empty placeholder ("-") — tab is IFS whitespace, so read collapses consecutive tabs instead of preserving empty fields. An empty field shifts the columns (this bug once made every repo show conflict and silently disabled jj ahead/behind)
--ignore-working-copy: Prevents expensive snapshot operations in the statusline
--no-pager: Ensures jj doesn't page when running non-interactively
Width-aware wrapping: the final block packs segments (model, dir, each VCS block, ctx) onto lines using a parallel seps[] array, so a wide terminal renders the original single line unchanged (model→dir and →ctx join with " | "; VCS blocks hug the dir with a space) while a narrow one wraps at segment boundaries. Width comes from COLUMNS (${COLUMNS:-80}); an over-wide current_dir is tail-truncated with a leading …. See Responsive Width-Aware Wrapping
Performance: 2-3 subprocess calls for jj, 3-5 for git, plus 2-3 for the sync check in colocated repos; acceptable for statusline frequency

Multi-Line Example

For a deliberately multi-line layout (independent of the width packer), add a context bar as a second line:

#!/bin/bash
input=$(cat)
model_name=$(echo "$input" | jq -r '.model.display_name')
current_dir=$(echo "$input" | jq -r '.workspace.current_dir')
pct=$(echo "$input" | jq -r '.context_window.used_percentage // 0' | cut -d. -f1)
cost=$(echo "$input" | jq -r '.cost.total_cost_usd // 0')

# Line 1: model + dir + VCS (add VCS logic from above)
echo "${model_name} | ${current_dir}"

# Line 2: context bar + cost
bar_width=20
filled=$((pct * bar_width / 100))
printf -v bar '%*s' "$filled" ''
printf -v empty '%*s' "$((bar_width - filled))" ''
echo "[${bar// /▓}${empty// /░}] ${pct}%  \$${cost}"

Responsive Width-Aware Wrapping

The reference script above ends with a greedy segment packer so a long statusline wraps cleanly instead of overflowing — Claude Code's docs warn that overflow "may get truncated or wrap awkwardly." This section explains how that packer works so you can adapt it.

Width source — COLUMNS/LINES env vars only. Claude Code sets COLUMNS and LINES to the current terminal dimensions before running the script (requires Claude Code v2.1.153+). Fall back to a default for older versions:

cols=${COLUMNS:-80}

What does NOT work, and why: Claude Code captures the script's stdout (it is not connected to the TTY) and pipes the JSON in on stdin, so there is no terminal for the script to query. tput cols, stty size, language-level width detection, and /dev/tty are all non-functional/undocumented — COLUMNS is the only supported mechanism.

Greedy segment packer. The status is built as discrete logical segments — model, dir, each [jj …]/[git …]/[jj ⇄ git …] block, and ctx — in a segs[] array, with a parallel seps[] array giving the separator that precedes each segment when it shares a line. Per-segment separators (rather than one uniform separator) are what let a wide terminal reproduce the original model | dir [jj …] [git …] | ctx N% layout byte-for-byte: model→dir and →ctx join with " | ", while VCS blocks hug the dir with a bare space. The packer walks the segments, appending to the current line while the next segment fits the width budget and breaking to a new line otherwise:

cols=${COLUMNS:-80}          # set by Claude Code v2.1.153+; 80 = fallback
budget=$((cols - 2)); [ "$budget" -lt 10 ] && budget=10

# segs[] / seps[] built as in the reference script (model, dir, VCS blocks, ctx)
line=""; linelen=0; out=()
for i in "${!segs[@]}"; do
    seg="${segs[$i]}"; [ -z "$seg" ] && continue
    sep="${seps[$i]}"; slen=${#seg}; seplen=${#sep}
    if [ -z "$line" ]; then
        line="$seg"; linelen=$slen
    elif [ $((linelen + seplen + slen)) -le "$budget" ]; then
        line+="${sep}${seg}"; linelen=$((linelen + seplen + slen))
    else
        out+=("$line"); line="$seg"; linelen=$slen   # wrap to next line
    fi
done
[ -n "$line" ] && out+=("$line")
printf '%s\n' "${out[@]}"      # each line = a separate status row

Wide terminal → one line (identical to the non-wrapping output); narrow → wraps cleanly at segment boundaries.

Caveats:

ANSI/OSC 8 escapes inflate ${#seg} — it counts the escape bytes, so colored/hyperlinked segments wrap too early. The reference script emits no escapes, so ${#seg} equals the display width; if you add color, measure a plain-text shadow copy of each segment (strip escapes with sed -E 's/\x1b\[[0-9;]*m//g') and pack on that length while emitting the colored version.
A single segment wider than budget (usually a long current_dir) can't break at a boundary. The reference script truncates such a segment to a leading … plus the tail (seg="…${seg: -$((budget - 1))}") rather than letting it overflow.
${#var} counts code points, not display columns — fine for ASCII, the ▓░ bar glyphs, and the ⇄ sync arrow (1 column each), but wide CJK/emoji may miscount. Acceptable as a statusline heuristic.

Prerequisites

jq (for parsing JSON input)
git (for git repo detection)
jj >= 0.36 (optional, for jj repo support)

Agent Skills: Statusline Configuration

Install this agent skill to your local

Skill Files