2026-08-08 11:54:05 +01:00
|
|
|
|
#!/bin/bash
|
|
|
|
|
|
# Claude Code Status Line
|
|
|
|
|
|
# Based on https://github.com/daniel3303/ClaudeCodeStatusLine
|
|
|
|
|
|
# Enhanced: git branch/ahead-behind, token bar, caching, improved visuals
|
|
|
|
|
|
|
|
|
|
|
|
set -f # disable globbing
|
|
|
|
|
|
|
|
|
|
|
|
# ===== Config =====
|
|
|
|
|
|
SHOW_GIT=true # git branch, dirty status, ahead/behind
|
|
|
|
|
|
SHOW_TOKENS=true # token usage bar
|
|
|
|
|
|
SHOW_THINKING=true # extended thinking indicator
|
|
|
|
|
|
SHOW_RATE_LIMITS=true # 5h / 7d rate limit bars
|
Show the per-model usage limit, and size the status line by measurement
/usage reports a weekly limit scoped to a single model (Fable) that the
status line did not surface. It is absent from both the status line's stdin
JSON and the legacy seven_day_opus/seven_day_sonnet fields, which are always
null now; it lives in the usage API's .limits[] under kind "weekly_scoped".
Render it labelled by scope.model.display_name so it follows whichever model
the limit applies to, inside the 7d segment since both are weekly limits
sharing one reset time.
Fitting it exposed a problem with sizing by width tier. Tiers only know the
terminal width, so content that varies with session state — branch name, cwd,
model display name — could push a line past the edge and be clipped by the
renderer. Measure the assembled line instead (vis_len) and emit the richest of
four rate-limit variants that fits, on one line or two: with reset times, with
extra-usage credits, bars only, or bars without the per-model segment. Every
branch is fit-checked, including with wrapping disabled. Two lines render
correctly in the status line.
Cap the cwd basename as well: nothing else shortened line one, so a long
project directory could overflow it on its own.
Correct the usable width. Claude Code exports COLUMNS but applies the `padding`
setting on top of it rather than deducting it first, so a line sized to COLUMNS
is clipped; deduct padding on both sides plus a column of margin.
Sanitise payload data before rendering. printf %b interprets backslash escapes,
which is how the colour variables work, so a cwd or model name containing a
literal \n — or a real newline, which jq decodes from the JSON — would split
the line and break both the width measurement and the two-line guarantee.
Parse the usage payload in one jq pass rather than ten. The status line runs on
every redraw and per-field parsing had become the dominant cost; this brings a
render back to roughly what it was before (~155ms vs ~115ms parent), the
remainder being the reset times the tiered version did not show at this width.
Fields are read one per line rather than via @tsv: tab is IFS whitespace, so
`read` collapses runs of it and an empty field — no scoped limit, which is the
common case — silently shifted every later field along by one.
Tests cover the API payload shapes including malformed and hostile input, the
width invariants (never exceed the usable width, never more than two lines,
never wrap unnecessarily), and a sweep over widths, branch lengths, model
names, cwd lengths and extra-usage. They restore the live usage cache on exit,
and remove the fixture outright when there was no cache to restore — otherwise
a test run would leave fabricated usage figures live for an hour.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-27 17:55:35 +01:00
|
|
|
|
SHOW_MODEL_LIMIT=true # per-model weekly limit bar (e.g. Fable), when the API reports one
|
|
|
|
|
|
WRAP_NARROW=true # wrap onto a second line rather than dropping segments or being clipped
|
|
|
|
|
|
WRAP_MIN_WIDTH=100 # below this, keep wide-tier line-one content (it can wrap)
|
2026-08-08 11:54:05 +01:00
|
|
|
|
BRANCH_MAX_LEN=28 # truncate branch names longer than this
|
Show the per-model usage limit, and size the status line by measurement
/usage reports a weekly limit scoped to a single model (Fable) that the
status line did not surface. It is absent from both the status line's stdin
JSON and the legacy seven_day_opus/seven_day_sonnet fields, which are always
null now; it lives in the usage API's .limits[] under kind "weekly_scoped".
Render it labelled by scope.model.display_name so it follows whichever model
the limit applies to, inside the 7d segment since both are weekly limits
sharing one reset time.
Fitting it exposed a problem with sizing by width tier. Tiers only know the
terminal width, so content that varies with session state — branch name, cwd,
model display name — could push a line past the edge and be clipped by the
renderer. Measure the assembled line instead (vis_len) and emit the richest of
four rate-limit variants that fits, on one line or two: with reset times, with
extra-usage credits, bars only, or bars without the per-model segment. Every
branch is fit-checked, including with wrapping disabled. Two lines render
correctly in the status line.
Cap the cwd basename as well: nothing else shortened line one, so a long
project directory could overflow it on its own.
Correct the usable width. Claude Code exports COLUMNS but applies the `padding`
setting on top of it rather than deducting it first, so a line sized to COLUMNS
is clipped; deduct padding on both sides plus a column of margin.
Sanitise payload data before rendering. printf %b interprets backslash escapes,
which is how the colour variables work, so a cwd or model name containing a
literal \n — or a real newline, which jq decodes from the JSON — would split
the line and break both the width measurement and the two-line guarantee.
Parse the usage payload in one jq pass rather than ten. The status line runs on
every redraw and per-field parsing had become the dominant cost; this brings a
render back to roughly what it was before (~155ms vs ~115ms parent), the
remainder being the reset times the tiered version did not show at this width.
Fields are read one per line rather than via @tsv: tab is IFS whitespace, so
`read` collapses runs of it and an empty field — no scoped limit, which is the
common case — silently shifted every later field along by one.
Tests cover the API payload shapes including malformed and hostile input, the
width invariants (never exceed the usable width, never more than two lines,
never wrap unnecessarily), and a sweep over widths, branch lengths, model
names, cwd lengths and extra-usage. They restore the live usage cache on exit,
and remove the fixture outright when there was no cache to restore — otherwise
a test run would leave fabricated usage figures live for an hour.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-27 17:55:35 +01:00
|
|
|
|
CWD_MAX_LEN=20 # truncate the cwd basename longer than this
|
2026-08-08 11:54:05 +01:00
|
|
|
|
GIT_CACHE_SECS=10 # seconds to cache git status (git diff is slow on large repos)
|
2026-09-02 13:01:12 +01:00
|
|
|
|
USAGE_CACHE_SECS=300 # seconds to cache the usage API response (the 5h/7d bars)
|
Back off after a failed usage fetch instead of retrying every redraw
A failed fetch left the cache mtime untouched, so with no network every
redraw — and a redraw happens on every keystroke — tried curl again and
could wait --max-time 10 for it. Cutting the TTL from an hour to five
minutes made that start 55 minutes sooner, so it is worth fixing now.
The marker is a separate file rather than a touch of the cached
response: a cold /tmp has no response to touch, and that is exactly the
case with no stale data to fall back on, where the timeout is paid in
full with nothing to show for it.
It goes down before the request and is cleared when that lands, not
written afterwards. A fetch is in flight for as long as curl takes to
give up, and this cache dir is shared by every session, so redraws
starting inside that window are real rather than hypothetical: they now
serve stale data instead of each launching their own doomed request.
Every attempt writes it, so a repeated failure restarts the backoff, and
only an attempt does, so the redraws it suppresses cannot keep
re-stamping it and it always lapses.
Missing credentials stays outside the backoff — it costs no timeout, and
the redraw after a login should show the bars rather than wait out a
retry window it had no part in earning.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-09-02 13:12:45 +01:00
|
|
|
|
USAGE_RETRY_SECS=60 # seconds to wait before retrying a failed usage API fetch
|
2026-08-08 11:54:05 +01:00
|
|
|
|
TOKEN_BAR_WIDTH=8 # width of token progress bar
|
|
|
|
|
|
|
Wrap the status line rather than drop the per-model limit
The measured ladder tried every single-line variant before considering a
second line, so whenever line one plus the full group did not fit — from
around 92 usable columns upwards, depending on how long the branch, cwd and
model names are — it emitted rl_bare (5h and 7d only) and silently dropped
the per-model (Fable) bar that is the main reason the group is worth
rendering. On a 110-column laptop the limit was invisible.
Reorder so a second line beats losing that bar: one line rich/mid/lean, then
wrap, and rl_bare only when WRAP_NARROW is false. Reset times and
extra-usage credits are still given up rather than wrapped for, which makes
the rendered content non-monotone in width; the comment on the ladder spells
that out.
Also stop the tests writing fixtures to the caches the live status line
reads. ~/.claude/statusline.sh is a symlink to the script, so a test run was
visibly rendering fabricated usage — a Fable bar at 10%, extra-usage credits
of $1234.56/$2000.00 — in whatever session happened to be open, and a run
killed before its trap fired would have left that in place for an hour. The
cache directory is now $STATUSLINE_CACHE_DIR (default /tmp/claude) and each
suite points it at a temporary directory, which also removes the
backup/restore dance and the deletion of the live git-status cache.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-31 17:28:55 +01:00
|
|
|
|
# Where the git and usage-API caches live. Overridable via the environment so a
|
|
|
|
|
|
# test run can render into its own directory: ~/.claude/statusline.sh is a
|
|
|
|
|
|
# symlink to this script, so a fixture written to the live cache is picked up by
|
2026-09-02 13:01:12 +01:00
|
|
|
|
# the next redraw and shown as real usage until USAGE_CACHE_SECS is up.
|
Wrap the status line rather than drop the per-model limit
The measured ladder tried every single-line variant before considering a
second line, so whenever line one plus the full group did not fit — from
around 92 usable columns upwards, depending on how long the branch, cwd and
model names are — it emitted rl_bare (5h and 7d only) and silently dropped
the per-model (Fable) bar that is the main reason the group is worth
rendering. On a 110-column laptop the limit was invisible.
Reorder so a second line beats losing that bar: one line rich/mid/lean, then
wrap, and rl_bare only when WRAP_NARROW is false. Reset times and
extra-usage credits are still given up rather than wrapped for, which makes
the rendered content non-monotone in width; the comment on the ladder spells
that out.
Also stop the tests writing fixtures to the caches the live status line
reads. ~/.claude/statusline.sh is a symlink to the script, so a test run was
visibly rendering fabricated usage — a Fable bar at 10%, extra-usage credits
of $1234.56/$2000.00 — in whatever session happened to be open, and a run
killed before its trap fired would have left that in place for an hour. The
cache directory is now $STATUSLINE_CACHE_DIR (default /tmp/claude) and each
suite points it at a temporary directory, which also removes the
backup/restore dance and the deletion of the live git-status cache.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-31 17:28:55 +01:00
|
|
|
|
CACHE_DIR="${STATUSLINE_CACHE_DIR:-/tmp/claude}"
|
|
|
|
|
|
|
2026-08-08 11:54:05 +01:00
|
|
|
|
# Terminal width detection.
|
Show the per-model usage limit, and size the status line by measurement
/usage reports a weekly limit scoped to a single model (Fable) that the
status line did not surface. It is absent from both the status line's stdin
JSON and the legacy seven_day_opus/seven_day_sonnet fields, which are always
null now; it lives in the usage API's .limits[] under kind "weekly_scoped".
Render it labelled by scope.model.display_name so it follows whichever model
the limit applies to, inside the 7d segment since both are weekly limits
sharing one reset time.
Fitting it exposed a problem with sizing by width tier. Tiers only know the
terminal width, so content that varies with session state — branch name, cwd,
model display name — could push a line past the edge and be clipped by the
renderer. Measure the assembled line instead (vis_len) and emit the richest of
four rate-limit variants that fits, on one line or two: with reset times, with
extra-usage credits, bars only, or bars without the per-model segment. Every
branch is fit-checked, including with wrapping disabled. Two lines render
correctly in the status line.
Cap the cwd basename as well: nothing else shortened line one, so a long
project directory could overflow it on its own.
Correct the usable width. Claude Code exports COLUMNS but applies the `padding`
setting on top of it rather than deducting it first, so a line sized to COLUMNS
is clipped; deduct padding on both sides plus a column of margin.
Sanitise payload data before rendering. printf %b interprets backslash escapes,
which is how the colour variables work, so a cwd or model name containing a
literal \n — or a real newline, which jq decodes from the JSON — would split
the line and break both the width measurement and the two-line guarantee.
Parse the usage payload in one jq pass rather than ten. The status line runs on
every redraw and per-field parsing had become the dominant cost; this brings a
render back to roughly what it was before (~155ms vs ~115ms parent), the
remainder being the reset times the tiered version did not show at this width.
Fields are read one per line rather than via @tsv: tab is IFS whitespace, so
`read` collapses runs of it and an empty field — no scoped limit, which is the
common case — silently shifted every later field along by one.
Tests cover the API payload shapes including malformed and hostile input, the
width invariants (never exceed the usable width, never more than two lines,
never wrap unnecessarily), and a sweep over widths, branch lengths, model
names, cwd lengths and extra-usage. They restore the live usage cache on exit,
and remove the fixture outright when there was no cache to restore — otherwise
a test run would leave fabricated usage figures live for an hour.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-27 17:55:35 +01:00
|
|
|
|
# Claude Code exports COLUMNS for this subprocess. There is no controlling
|
|
|
|
|
|
# terminal, so the stty fallback fails; when both fail we default to 80
|
|
|
|
|
|
# (safe/compact) rather than wide.
|
2026-08-08 11:54:05 +01:00
|
|
|
|
#
|
Show the per-model usage limit, and size the status line by measurement
/usage reports a weekly limit scoped to a single model (Fable) that the
status line did not surface. It is absent from both the status line's stdin
JSON and the legacy seven_day_opus/seven_day_sonnet fields, which are always
null now; it lives in the usage API's .limits[] under kind "weekly_scoped".
Render it labelled by scope.model.display_name so it follows whichever model
the limit applies to, inside the 7d segment since both are weekly limits
sharing one reset time.
Fitting it exposed a problem with sizing by width tier. Tiers only know the
terminal width, so content that varies with session state — branch name, cwd,
model display name — could push a line past the edge and be clipped by the
renderer. Measure the assembled line instead (vis_len) and emit the richest of
four rate-limit variants that fits, on one line or two: with reset times, with
extra-usage credits, bars only, or bars without the per-model segment. Every
branch is fit-checked, including with wrapping disabled. Two lines render
correctly in the status line.
Cap the cwd basename as well: nothing else shortened line one, so a long
project directory could overflow it on its own.
Correct the usable width. Claude Code exports COLUMNS but applies the `padding`
setting on top of it rather than deducting it first, so a line sized to COLUMNS
is clipped; deduct padding on both sides plus a column of margin.
Sanitise payload data before rendering. printf %b interprets backslash escapes,
which is how the colour variables work, so a cwd or model name containing a
literal \n — or a real newline, which jq decodes from the JSON — would split
the line and break both the width measurement and the two-line guarantee.
Parse the usage payload in one jq pass rather than ten. The status line runs on
every redraw and per-field parsing had become the dominant cost; this brings a
render back to roughly what it was before (~155ms vs ~115ms parent), the
remainder being the reset times the tiered version did not show at this width.
Fields are read one per line rather than via @tsv: tab is IFS whitespace, so
`read` collapses runs of it and an empty field — no scoped limit, which is the
common case — silently shifted every later field along by one.
Tests cover the API payload shapes including malformed and hostile input, the
width invariants (never exceed the usable width, never more than two lines,
never wrap unnecessarily), and a sweep over widths, branch lengths, model
names, cwd lengths and extra-usage. They restore the live usage cache on exit,
and remove the fixture outright when there was no cache to restore — otherwise
a test run would leave fabricated usage figures live for an hour.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-27 17:55:35 +01:00
|
|
|
|
# The `padding` setting indents the status line on both sides, and Claude Code
|
|
|
|
|
|
# applies it on top of COLUMNS rather than deducting it first — measured: with
|
|
|
|
|
|
# COLUMNS=121 and padding 2, a 121-column line is clipped with an ellipsis. So
|
|
|
|
|
|
# deduct it here, plus a column of margin.
|
|
|
|
|
|
#
|
|
|
|
|
|
# To override detection entirely, set TERM_WIDTH in settings.json:
|
2026-08-08 11:54:05 +01:00
|
|
|
|
# "command": "TERM_WIDTH=160 ~/.claude/statusline.sh"
|
Show the per-model usage limit, and size the status line by measurement
/usage reports a weekly limit scoped to a single model (Fable) that the
status line did not surface. It is absent from both the status line's stdin
JSON and the legacy seven_day_opus/seven_day_sonnet fields, which are always
null now; it lives in the usage API's .limits[] under kind "weekly_scoped".
Render it labelled by scope.model.display_name so it follows whichever model
the limit applies to, inside the 7d segment since both are weekly limits
sharing one reset time.
Fitting it exposed a problem with sizing by width tier. Tiers only know the
terminal width, so content that varies with session state — branch name, cwd,
model display name — could push a line past the edge and be clipped by the
renderer. Measure the assembled line instead (vis_len) and emit the richest of
four rate-limit variants that fits, on one line or two: with reset times, with
extra-usage credits, bars only, or bars without the per-model segment. Every
branch is fit-checked, including with wrapping disabled. Two lines render
correctly in the status line.
Cap the cwd basename as well: nothing else shortened line one, so a long
project directory could overflow it on its own.
Correct the usable width. Claude Code exports COLUMNS but applies the `padding`
setting on top of it rather than deducting it first, so a line sized to COLUMNS
is clipped; deduct padding on both sides plus a column of margin.
Sanitise payload data before rendering. printf %b interprets backslash escapes,
which is how the colour variables work, so a cwd or model name containing a
literal \n — or a real newline, which jq decodes from the JSON — would split
the line and break both the width measurement and the two-line guarantee.
Parse the usage payload in one jq pass rather than ten. The status line runs on
every redraw and per-field parsing had become the dominant cost; this brings a
render back to roughly what it was before (~155ms vs ~115ms parent), the
remainder being the reset times the tiered version did not show at this width.
Fields are read one per line rather than via @tsv: tab is IFS whitespace, so
`read` collapses runs of it and an empty field — no scoped limit, which is the
common case — silently shifted every later field along by one.
Tests cover the API payload shapes including malformed and hostile input, the
width invariants (never exceed the usable width, never more than two lines,
never wrap unnecessarily), and a sweep over widths, branch lengths, model
names, cwd lengths and extra-usage. They restore the live usage cache on exit,
and remove the fixture outright when there was no cache to restore — otherwise
a test run would leave fabricated usage figures live for an hour.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-27 17:55:35 +01:00
|
|
|
|
# Read both settings this script needs in one jq pass.
|
|
|
|
|
|
# Only the user-level file: Claude Code also merges project .claude/settings.json,
|
|
|
|
|
|
# .claude/settings.local.json and managed policy, so a statusLine.padding set at
|
|
|
|
|
|
# project level would be applied by the renderer but missed here.
|
|
|
|
|
|
sl_padding=0
|
|
|
|
|
|
sl_thinking=false
|
|
|
|
|
|
if [ -f "$HOME/.claude/settings.json" ]; then
|
|
|
|
|
|
{ IFS= read -r sl_padding; IFS= read -r sl_thinking; } <<EOF
|
|
|
|
|
|
$(jq -r '[(.statusLine.padding // 0), (.alwaysThinkingEnabled // false)] | .[] | tostring' "$HOME/.claude/settings.json" 2>/dev/null)
|
|
|
|
|
|
EOF
|
|
|
|
|
|
[ "${sl_padding:-0}" -ge 0 ] 2>/dev/null || sl_padding=0
|
|
|
|
|
|
fi
|
|
|
|
|
|
|
2026-08-08 11:54:05 +01:00
|
|
|
|
if [ "${TERM_WIDTH:-0}" -le 0 ] 2>/dev/null; then
|
|
|
|
|
|
if [ "${COLUMNS:-0}" -gt 0 ] 2>/dev/null; then
|
|
|
|
|
|
TERM_WIDTH=$COLUMNS
|
|
|
|
|
|
else
|
|
|
|
|
|
_w=$(stty size </dev/tty 2>/dev/null | awk '{print $2}')
|
|
|
|
|
|
[ "${_w:-0}" -gt 0 ] 2>/dev/null && TERM_WIDTH=$_w || TERM_WIDTH=80
|
|
|
|
|
|
unset _w
|
|
|
|
|
|
fi
|
|
|
|
|
|
fi
|
|
|
|
|
|
|
Show the per-model usage limit, and size the status line by measurement
/usage reports a weekly limit scoped to a single model (Fable) that the
status line did not surface. It is absent from both the status line's stdin
JSON and the legacy seven_day_opus/seven_day_sonnet fields, which are always
null now; it lives in the usage API's .limits[] under kind "weekly_scoped".
Render it labelled by scope.model.display_name so it follows whichever model
the limit applies to, inside the 7d segment since both are weekly limits
sharing one reset time.
Fitting it exposed a problem with sizing by width tier. Tiers only know the
terminal width, so content that varies with session state — branch name, cwd,
model display name — could push a line past the edge and be clipped by the
renderer. Measure the assembled line instead (vis_len) and emit the richest of
four rate-limit variants that fits, on one line or two: with reset times, with
extra-usage credits, bars only, or bars without the per-model segment. Every
branch is fit-checked, including with wrapping disabled. Two lines render
correctly in the status line.
Cap the cwd basename as well: nothing else shortened line one, so a long
project directory could overflow it on its own.
Correct the usable width. Claude Code exports COLUMNS but applies the `padding`
setting on top of it rather than deducting it first, so a line sized to COLUMNS
is clipped; deduct padding on both sides plus a column of margin.
Sanitise payload data before rendering. printf %b interprets backslash escapes,
which is how the colour variables work, so a cwd or model name containing a
literal \n — or a real newline, which jq decodes from the JSON — would split
the line and break both the width measurement and the two-line guarantee.
Parse the usage payload in one jq pass rather than ten. The status line runs on
every redraw and per-field parsing had become the dominant cost; this brings a
render back to roughly what it was before (~155ms vs ~115ms parent), the
remainder being the reset times the tiered version did not show at this width.
Fields are read one per line rather than via @tsv: tab is IFS whitespace, so
`read` collapses runs of it and an empty field — no scoped limit, which is the
common case — silently shifted every later field along by one.
Tests cover the API payload shapes including malformed and hostile input, the
width invariants (never exceed the usable width, never more than two lines,
never wrap unnecessarily), and a sweep over widths, branch lengths, model
names, cwd lengths and extra-usage. They restore the live usage cache on exit,
and remove the fixture outright when there was no cache to restore — otherwise
a test run would leave fabricated usage figures live for an hour.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-27 17:55:35 +01:00
|
|
|
|
# Usable width: minus padding on both sides, minus a column of margin.
|
|
|
|
|
|
USABLE_WIDTH=$(( TERM_WIDTH - 2 * sl_padding - 1 ))
|
|
|
|
|
|
[ "$USABLE_WIDTH" -lt 20 ] && USABLE_WIDTH=20
|
|
|
|
|
|
|
|
|
|
|
|
# ${#str} counts characters only under a UTF-8 locale; under LC_ALL=C it counts
|
|
|
|
|
|
# bytes, which would mis-measure the box/block glyphs and break line wrapping.
|
|
|
|
|
|
case "${LC_ALL:-${LC_CTYPE:-${LANG:-}}}" in
|
|
|
|
|
|
*UTF-8*|*utf8*|*UTF8*) ;;
|
|
|
|
|
|
*)
|
|
|
|
|
|
# C.UTF-8 on glibc, en_US.UTF-8 on macOS (which has no C.UTF-8).
|
|
|
|
|
|
# Picking one that does not exist makes bash warn on every render and
|
|
|
|
|
|
# silently fall back to byte counting, so verify before committing.
|
|
|
|
|
|
for _loc in C.UTF-8 en_US.UTF-8; do
|
|
|
|
|
|
if LC_ALL="$_loc" locale charmap 2>/dev/null | grep -qi utf; then
|
|
|
|
|
|
export LC_ALL="$_loc"; break
|
|
|
|
|
|
fi
|
|
|
|
|
|
done
|
|
|
|
|
|
unset _loc
|
|
|
|
|
|
;;
|
|
|
|
|
|
esac
|
|
|
|
|
|
|
2026-08-08 11:54:05 +01:00
|
|
|
|
input=$(cat)
|
|
|
|
|
|
[ -z "$input" ] && printf "Claude" && exit 0
|
|
|
|
|
|
|
Wrap the status line rather than drop the per-model limit
The measured ladder tried every single-line variant before considering a
second line, so whenever line one plus the full group did not fit — from
around 92 usable columns upwards, depending on how long the branch, cwd and
model names are — it emitted rl_bare (5h and 7d only) and silently dropped
the per-model (Fable) bar that is the main reason the group is worth
rendering. On a 110-column laptop the limit was invisible.
Reorder so a second line beats losing that bar: one line rich/mid/lean, then
wrap, and rl_bare only when WRAP_NARROW is false. Reset times and
extra-usage credits are still given up rather than wrapped for, which makes
the rendered content non-monotone in width; the comment on the ladder spells
that out.
Also stop the tests writing fixtures to the caches the live status line
reads. ~/.claude/statusline.sh is a symlink to the script, so a test run was
visibly rendering fabricated usage — a Fable bar at 10%, extra-usage credits
of $1234.56/$2000.00 — in whatever session happened to be open, and a run
killed before its trap fired would have left that in place for an hour. The
cache directory is now $STATUSLINE_CACHE_DIR (default /tmp/claude) and each
suite points it at a temporary directory, which also removes the
backup/restore dance and the deletion of the live git-status cache.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-31 17:28:55 +01:00
|
|
|
|
mkdir -p "$CACHE_DIR"
|
2026-08-08 11:54:05 +01:00
|
|
|
|
|
|
|
|
|
|
# ===== Colors =====
|
Make the Claude Code statusline follow the terminal palette
The nine colours were hardcoded One Dark RGB via 38;2;R;G;B, so they did not
follow the light/dark switch. Measured against the tangere backgrounds they are
fine on dark (4.25-10.80:1) but collapse on light: orange 1.78, amber 1.70,
yellow 1.63, and 'white' — used for the token total — at 1.35:1, effectively
invisible.
Switched to palette indices, which resolve through the terminal's own palette and
so follow the ghostty tangere-light/tangere-dark swap with no detection and no
cost per render. 'white' becomes 39, default foreground, since it means primary
text. All seven colours actually used get distinct indices; blue and yellow are
unreferenced but converted too, so no RGB is left behind.
Contrast now clears WCAG AA in both modes: 5.77-14.26:1 on dark, 6.32-20.61:1 on
light.
Trade-off: tangere has no orange and its palette 2 is teal rather than green, so
'green' reads teal and the warm colours sit closer together than the One Dark
originals did.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-09 14:02:01 +01:00
|
|
|
|
# Palette indices rather than RGB, so these resolve through the terminal's own
|
|
|
|
|
|
# palette and follow the ghostty tangere-light/tangere-dark swap for free. No
|
|
|
|
|
|
# appearance detection, so no cost per statusline render.
|
|
|
|
|
|
#
|
|
|
|
|
|
# The previous values were One Dark RGB: fine on dark (4.3-10.8:1) but 1.35-3.4:1
|
|
|
|
|
|
# on a light background, with 'white' effectively invisible. These clear WCAG AA
|
|
|
|
|
|
# in both modes (5.8-14.3:1 on dark, 6.3-20.6:1 on light).
|
|
|
|
|
|
blue='\033[38;5;4m'
|
|
|
|
|
|
orange='\033[38;5;3m'
|
|
|
|
|
|
amber='\033[38;5;11m'
|
|
|
|
|
|
green='\033[38;5;2m'
|
|
|
|
|
|
cyan='\033[38;5;14m'
|
|
|
|
|
|
red='\033[38;5;1m'
|
|
|
|
|
|
yellow='\033[38;5;3m'
|
|
|
|
|
|
white='\033[39m'
|
|
|
|
|
|
magenta='\033[38;5;5m'
|
2026-08-08 11:54:05 +01:00
|
|
|
|
dim='\033[2m'
|
|
|
|
|
|
reset='\033[0m'
|
|
|
|
|
|
|
|
|
|
|
|
sep=" ${dim}│${reset} "
|
|
|
|
|
|
|
|
|
|
|
|
# ===== Helpers =====
|
|
|
|
|
|
|
|
|
|
|
|
format_tokens() {
|
|
|
|
|
|
local num=$1
|
|
|
|
|
|
if [ "$num" -ge 1000000 ]; then
|
|
|
|
|
|
awk "BEGIN {printf \"%.1fm\", $num / 1000000}"
|
|
|
|
|
|
elif [ "$num" -ge 1000 ]; then
|
|
|
|
|
|
awk "BEGIN {printf \"%.0fk\", $num / 1000}"
|
|
|
|
|
|
else
|
|
|
|
|
|
printf "%d" "$num"
|
|
|
|
|
|
fi
|
|
|
|
|
|
}
|
|
|
|
|
|
|
Show the per-model usage limit, and size the status line by measurement
/usage reports a weekly limit scoped to a single model (Fable) that the
status line did not surface. It is absent from both the status line's stdin
JSON and the legacy seven_day_opus/seven_day_sonnet fields, which are always
null now; it lives in the usage API's .limits[] under kind "weekly_scoped".
Render it labelled by scope.model.display_name so it follows whichever model
the limit applies to, inside the 7d segment since both are weekly limits
sharing one reset time.
Fitting it exposed a problem with sizing by width tier. Tiers only know the
terminal width, so content that varies with session state — branch name, cwd,
model display name — could push a line past the edge and be clipped by the
renderer. Measure the assembled line instead (vis_len) and emit the richest of
four rate-limit variants that fits, on one line or two: with reset times, with
extra-usage credits, bars only, or bars without the per-model segment. Every
branch is fit-checked, including with wrapping disabled. Two lines render
correctly in the status line.
Cap the cwd basename as well: nothing else shortened line one, so a long
project directory could overflow it on its own.
Correct the usable width. Claude Code exports COLUMNS but applies the `padding`
setting on top of it rather than deducting it first, so a line sized to COLUMNS
is clipped; deduct padding on both sides plus a column of margin.
Sanitise payload data before rendering. printf %b interprets backslash escapes,
which is how the colour variables work, so a cwd or model name containing a
literal \n — or a real newline, which jq decodes from the JSON — would split
the line and break both the width measurement and the two-line guarantee.
Parse the usage payload in one jq pass rather than ten. The status line runs on
every redraw and per-field parsing had become the dominant cost; this brings a
render back to roughly what it was before (~155ms vs ~115ms parent), the
remainder being the reset times the tiered version did not show at this width.
Fields are read one per line rather than via @tsv: tab is IFS whitespace, so
`read` collapses runs of it and an empty field — no scoped limit, which is the
common case — silently shifted every later field along by one.
Tests cover the API payload shapes including malformed and hostile input, the
width invariants (never exceed the usable width, never more than two lines,
never wrap unnecessarily), and a sweep over widths, branch lengths, model
names, cwd lengths and extra-usage. They restore the live usage cache on exit,
and remove the fixture outright when there was no cache to restore — otherwise
a test run would leave fabricated usage figures live for an hour.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-27 17:55:35 +01:00
|
|
|
|
# Display width of a string, ignoring ANSI colour escapes. Every glyph the
|
|
|
|
|
|
# statusline uses is single-width, so a character count is the display width.
|
|
|
|
|
|
#
|
|
|
|
|
|
# The colour vars hold the escapes as literal backslash-033 text (they are
|
|
|
|
|
|
# single-quoted, and printf %b only interprets them when the line is finally
|
|
|
|
|
|
# emitted), so strip that literal form as well as a real ESC byte.
|
|
|
|
|
|
vis_len() {
|
|
|
|
|
|
local plain
|
|
|
|
|
|
plain=$(printf '%s' "$1" | sed -e 's/\\033\[[0-9;]*m//g' -e $'s/\033\[[0-9;]*m//g' -e 's/\\\\/\\/g')
|
|
|
|
|
|
printf '%d' "${#plain}"
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
# printf %b interprets backslash escapes -- that is how the colour variables
|
|
|
|
|
|
# above are applied. Data from the API or the filesystem must NOT be
|
|
|
|
|
|
# interpreted: a directory or model name containing a literal \n would split
|
|
|
|
|
|
# the line, breaking both the width measurement and the two-line guarantee.
|
|
|
|
|
|
# Double the backslashes so %b renders them literally.
|
|
|
|
|
|
esc_data() {
|
|
|
|
|
|
local v="$1"
|
|
|
|
|
|
# Real control characters (jq decodes \n in the JSON payload into an actual
|
|
|
|
|
|
# newline) would split the line regardless of escaping, so drop them first.
|
|
|
|
|
|
v="${v//$'\n'/ }"; v="${v//$'\r'/ }"; v="${v//$'\t'/ }"; v="${v//$'\033'/}"
|
|
|
|
|
|
printf '%s' "${v//\\/\\\\}"
|
|
|
|
|
|
}
|
|
|
|
|
|
|
2026-08-08 11:54:05 +01:00
|
|
|
|
truncate_str() {
|
|
|
|
|
|
local str="$1" max="$2"
|
|
|
|
|
|
[ "${#str}" -gt "$max" ] \
|
|
|
|
|
|
&& printf "%s…" "${str:0:$((max-1))}" \
|
|
|
|
|
|
|| printf "%s" "$str"
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
# Colored progress bar using block chars
|
|
|
|
|
|
build_bar() {
|
|
|
|
|
|
local pct=$1 width=$2
|
|
|
|
|
|
[ "$pct" -lt 0 ] 2>/dev/null && pct=0
|
|
|
|
|
|
[ "$pct" -gt 100 ] 2>/dev/null && pct=100
|
|
|
|
|
|
local filled=$(( pct * width / 100 ))
|
|
|
|
|
|
local empty=$(( width - filled ))
|
|
|
|
|
|
local bar_color
|
|
|
|
|
|
if [ "$pct" -ge 90 ]; then bar_color="$red"
|
|
|
|
|
|
elif [ "$pct" -ge 70 ]; then bar_color="$yellow"
|
|
|
|
|
|
elif [ "$pct" -ge 50 ]; then bar_color="$orange"
|
|
|
|
|
|
else bar_color="$green"
|
|
|
|
|
|
fi
|
|
|
|
|
|
local f="" e=""
|
|
|
|
|
|
for ((i=0; i<filled; i++)); do f+="█"; done
|
|
|
|
|
|
for ((i=0; i<empty; i++)); do e+="░"; done
|
|
|
|
|
|
printf "${bar_color}${f}${dim}${e}${reset}"
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
# ===== Git info with per-directory caching =====
|
|
|
|
|
|
get_git_info() {
|
|
|
|
|
|
local dir="$1"
|
|
|
|
|
|
[ -z "$dir" ] && return
|
|
|
|
|
|
|
|
|
|
|
|
# Stable cache key per directory path
|
|
|
|
|
|
local dir_hash
|
|
|
|
|
|
dir_hash=$(printf '%s' "$dir" | cksum | awk '{print $1}')
|
Wrap the status line rather than drop the per-model limit
The measured ladder tried every single-line variant before considering a
second line, so whenever line one plus the full group did not fit — from
around 92 usable columns upwards, depending on how long the branch, cwd and
model names are — it emitted rl_bare (5h and 7d only) and silently dropped
the per-model (Fable) bar that is the main reason the group is worth
rendering. On a 110-column laptop the limit was invisible.
Reorder so a second line beats losing that bar: one line rich/mid/lean, then
wrap, and rl_bare only when WRAP_NARROW is false. Reset times and
extra-usage credits are still given up rather than wrapped for, which makes
the rendered content non-monotone in width; the comment on the ladder spells
that out.
Also stop the tests writing fixtures to the caches the live status line
reads. ~/.claude/statusline.sh is a symlink to the script, so a test run was
visibly rendering fabricated usage — a Fable bar at 10%, extra-usage credits
of $1234.56/$2000.00 — in whatever session happened to be open, and a run
killed before its trap fired would have left that in place for an hour. The
cache directory is now $STATUSLINE_CACHE_DIR (default /tmp/claude) and each
suite points it at a temporary directory, which also removes the
backup/restore dance and the deletion of the live git-status cache.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-31 17:28:55 +01:00
|
|
|
|
local cache_file="$CACHE_DIR/git-${dir_hash}"
|
2026-08-08 11:54:05 +01:00
|
|
|
|
|
|
|
|
|
|
local needs_refresh=true
|
|
|
|
|
|
if [ -f "$cache_file" ]; then
|
|
|
|
|
|
local mtime now age
|
|
|
|
|
|
mtime=$(stat -f %m "$cache_file" 2>/dev/null || stat -c %Y "$cache_file" 2>/dev/null)
|
|
|
|
|
|
now=$(date +%s)
|
|
|
|
|
|
age=$(( now - mtime ))
|
|
|
|
|
|
[ "$age" -lt "$GIT_CACHE_SECS" ] && needs_refresh=false
|
|
|
|
|
|
fi
|
|
|
|
|
|
|
|
|
|
|
|
if $needs_refresh; then
|
|
|
|
|
|
# Branch name (or short hash for detached HEAD)
|
|
|
|
|
|
local branch
|
|
|
|
|
|
branch=$(git -C "$dir" symbolic-ref --short HEAD 2>/dev/null)
|
|
|
|
|
|
if [ -z "$branch" ]; then
|
|
|
|
|
|
branch=$(git -C "$dir" rev-parse --short HEAD 2>/dev/null)
|
|
|
|
|
|
if [ -z "$branch" ]; then
|
|
|
|
|
|
: > "$cache_file" # not a git repo — write empty cache
|
|
|
|
|
|
return
|
|
|
|
|
|
fi
|
|
|
|
|
|
branch="(${branch})"
|
|
|
|
|
|
fi
|
|
|
|
|
|
|
|
|
|
|
|
# Dirty: any staged or unstaged changes
|
|
|
|
|
|
local dirty=""
|
|
|
|
|
|
[ -n "$(git -C "$dir" status --porcelain 2>/dev/null)" ] && dirty="dirty"
|
|
|
|
|
|
|
|
|
|
|
|
# Ahead / behind upstream (skip if no upstream set)
|
|
|
|
|
|
local ahead=0 behind=0
|
|
|
|
|
|
local upstream
|
|
|
|
|
|
upstream=$(git -C "$dir" rev-parse --abbrev-ref "@{upstream}" 2>/dev/null)
|
|
|
|
|
|
if [ -n "$upstream" ]; then
|
|
|
|
|
|
local ab
|
|
|
|
|
|
ab=$(git -C "$dir" rev-list --left-right --count "HEAD...${upstream}" 2>/dev/null)
|
|
|
|
|
|
ahead=$(echo "$ab" | awk '{print $1}')
|
|
|
|
|
|
behind=$(echo "$ab" | awk '{print $2}')
|
|
|
|
|
|
fi
|
|
|
|
|
|
|
|
|
|
|
|
# Tab-delimited cache (branch names can't contain tabs per git spec)
|
|
|
|
|
|
printf '%s\t%s\t%s\t%s' "$branch" "$dirty" "${ahead:-0}" "${behind:-0}" > "$cache_file"
|
|
|
|
|
|
fi
|
|
|
|
|
|
|
|
|
|
|
|
cat "$cache_file" 2>/dev/null
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
# ===== OAuth token resolution =====
|
|
|
|
|
|
get_oauth_token() {
|
|
|
|
|
|
[ -n "$CLAUDE_CODE_OAUTH_TOKEN" ] && echo "$CLAUDE_CODE_OAUTH_TOKEN" && return 0
|
|
|
|
|
|
|
|
|
|
|
|
if command -v security >/dev/null 2>&1; then
|
|
|
|
|
|
local blob token
|
|
|
|
|
|
blob=$(security find-generic-password -s "Claude Code-credentials" -w 2>/dev/null)
|
|
|
|
|
|
if [ -n "$blob" ]; then
|
|
|
|
|
|
token=$(echo "$blob" | jq -r '.claudeAiOauth.accessToken // empty' 2>/dev/null)
|
|
|
|
|
|
[ -n "$token" ] && [ "$token" != "null" ] && echo "$token" && return 0
|
|
|
|
|
|
fi
|
|
|
|
|
|
fi
|
|
|
|
|
|
|
|
|
|
|
|
local creds_file="${HOME}/.claude/.credentials.json"
|
|
|
|
|
|
if [ -f "$creds_file" ]; then
|
|
|
|
|
|
local token
|
|
|
|
|
|
token=$(jq -r '.claudeAiOauth.accessToken // empty' "$creds_file" 2>/dev/null)
|
|
|
|
|
|
[ -n "$token" ] && [ "$token" != "null" ] && echo "$token" && return 0
|
|
|
|
|
|
fi
|
|
|
|
|
|
|
|
|
|
|
|
if command -v secret-tool >/dev/null 2>&1; then
|
|
|
|
|
|
local blob token
|
|
|
|
|
|
blob=$(timeout 2 secret-tool lookup service "Claude Code-credentials" 2>/dev/null)
|
|
|
|
|
|
if [ -n "$blob" ]; then
|
|
|
|
|
|
token=$(echo "$blob" | jq -r '.claudeAiOauth.accessToken // empty' 2>/dev/null)
|
|
|
|
|
|
[ -n "$token" ] && [ "$token" != "null" ] && echo "$token" && return 0
|
|
|
|
|
|
fi
|
|
|
|
|
|
fi
|
|
|
|
|
|
|
|
|
|
|
|
echo ""
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
# ===== ISO 8601 → epoch (cross-platform) =====
|
|
|
|
|
|
iso_to_epoch() {
|
|
|
|
|
|
local iso_str="$1"
|
|
|
|
|
|
local epoch
|
|
|
|
|
|
|
|
|
|
|
|
epoch=$(date -d "${iso_str}" +%s 2>/dev/null)
|
|
|
|
|
|
[ -n "$epoch" ] && echo "$epoch" && return 0
|
|
|
|
|
|
|
|
|
|
|
|
local stripped="${iso_str%%.*}"
|
|
|
|
|
|
stripped="${stripped%%Z}"
|
|
|
|
|
|
stripped="${stripped%%+*}"
|
|
|
|
|
|
stripped="${stripped%%-[0-9][0-9]:[0-9][0-9]}"
|
|
|
|
|
|
|
|
|
|
|
|
if [[ "$iso_str" == *"Z"* ]] || [[ "$iso_str" == *"+00:00"* ]] || [[ "$iso_str" == *"-00:00"* ]]; then
|
|
|
|
|
|
epoch=$(env TZ=UTC date -j -f "%Y-%m-%dT%H:%M:%S" "$stripped" +%s 2>/dev/null)
|
|
|
|
|
|
else
|
|
|
|
|
|
epoch=$(date -j -f "%Y-%m-%dT%H:%M:%S" "$stripped" +%s 2>/dev/null)
|
|
|
|
|
|
fi
|
|
|
|
|
|
|
|
|
|
|
|
[ -n "$epoch" ] && echo "$epoch" && return 0
|
|
|
|
|
|
return 1
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
format_reset_time() {
|
|
|
|
|
|
local iso_str="$1" style="$2"
|
|
|
|
|
|
[ -z "$iso_str" ] || [ "$iso_str" = "null" ] && return
|
|
|
|
|
|
local epoch
|
|
|
|
|
|
epoch=$(iso_to_epoch "$iso_str")
|
|
|
|
|
|
[ -z "$epoch" ] && return
|
|
|
|
|
|
case "$style" in
|
|
|
|
|
|
time)
|
|
|
|
|
|
date -j -r "$epoch" +"%l:%M%p" 2>/dev/null | sed 's/^ //' | tr '[:upper:]' '[:lower:]' ||
|
|
|
|
|
|
date -d "@$epoch" +"%l:%M%P" 2>/dev/null | sed 's/^ //'
|
|
|
|
|
|
;;
|
|
|
|
|
|
datetime)
|
|
|
|
|
|
date -j -r "$epoch" +"%b %-d, %l:%M%p" 2>/dev/null | sed 's/ / /g; s/^ //' | tr '[:upper:]' '[:lower:]' ||
|
|
|
|
|
|
date -d "@$epoch" +"%b %-d, %l:%M%P" 2>/dev/null | sed 's/ / /g; s/^ //'
|
|
|
|
|
|
;;
|
|
|
|
|
|
esac
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
# ===== Extract JSON =====
|
|
|
|
|
|
model_name=$(echo "$input" | jq -r '.model.display_name // "Claude"')
|
|
|
|
|
|
cwd=$(echo "$input" | jq -r '.cwd // empty')
|
|
|
|
|
|
cost_usd=$(echo "$input" | jq -r '.cost.total_cost_usd // empty')
|
Measure line one's branch budget instead of assuming 56 columns
In wrap mode the branch is truncated to whatever the rest of line one leaves,
and "the rest" was a flat 56 columns. It is not flat: the token counts and the
cost figure vary with the session, and with a four-digit cost the real width is
57, so line one came out exactly one column over and the renderer clipped it.
Reachable at COLUMNS 73-81 with padding 2 — a narrow split pane.
Add up the segments that follow the branch instead, from the same values that
render them, and count the ahead/behind markers too, which the constant also
ignored. That needs the cost formatted and the token bar width chosen before
the git segment is built, so both move up; neither depends on anything in
between.
The budget is now exact rather than approximate, so the branch also gets the
columns the old constant was over-reserving when the cost was short.
Tests: the large-cost sweep only used the short branch 'main', where the
overflow cannot show. Repeat it with the long branch, which fails on the old
budget at usable 68, 69, 70 and 75 by exactly one column.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-31 19:01:24 +01:00
|
|
|
|
# Formatted here rather than where it is rendered: the wrap-mode branch budget
|
|
|
|
|
|
# needs its width, and it is not fixed ("$4.61" vs "$1234.56").
|
|
|
|
|
|
cost_fmt=""
|
|
|
|
|
|
[ -n "$cost_usd" ] && cost_fmt=$(printf '%.2f' "$cost_usd" 2>/dev/null)
|
2026-08-08 11:54:05 +01:00
|
|
|
|
|
|
|
|
|
|
size=$(echo "$input" | jq -r '.context_window.context_window_size // 200000')
|
|
|
|
|
|
[ "$size" -eq 0 ] 2>/dev/null && size=200000
|
|
|
|
|
|
|
|
|
|
|
|
input_tokens=$(echo "$input" | jq -r '.context_window.current_usage.input_tokens // 0')
|
|
|
|
|
|
cache_create=$(echo "$input" | jq -r '.context_window.current_usage.cache_creation_input_tokens // 0')
|
|
|
|
|
|
cache_read=$(echo "$input" | jq -r '.context_window.current_usage.cache_read_input_tokens // 0')
|
|
|
|
|
|
current=$(( input_tokens + cache_create + cache_read ))
|
|
|
|
|
|
|
|
|
|
|
|
used_tokens=$(format_tokens $current)
|
|
|
|
|
|
total_tokens=$(format_tokens $size)
|
|
|
|
|
|
pct_used=$(( size > 0 ? current * 100 / size : 0 ))
|
|
|
|
|
|
|
|
|
|
|
|
thinking_on=false
|
Show the per-model usage limit, and size the status line by measurement
/usage reports a weekly limit scoped to a single model (Fable) that the
status line did not surface. It is absent from both the status line's stdin
JSON and the legacy seven_day_opus/seven_day_sonnet fields, which are always
null now; it lives in the usage API's .limits[] under kind "weekly_scoped".
Render it labelled by scope.model.display_name so it follows whichever model
the limit applies to, inside the 7d segment since both are weekly limits
sharing one reset time.
Fitting it exposed a problem with sizing by width tier. Tiers only know the
terminal width, so content that varies with session state — branch name, cwd,
model display name — could push a line past the edge and be clipped by the
renderer. Measure the assembled line instead (vis_len) and emit the richest of
four rate-limit variants that fits, on one line or two: with reset times, with
extra-usage credits, bars only, or bars without the per-model segment. Every
branch is fit-checked, including with wrapping disabled. Two lines render
correctly in the status line.
Cap the cwd basename as well: nothing else shortened line one, so a long
project directory could overflow it on its own.
Correct the usable width. Claude Code exports COLUMNS but applies the `padding`
setting on top of it rather than deducting it first, so a line sized to COLUMNS
is clipped; deduct padding on both sides plus a column of margin.
Sanitise payload data before rendering. printf %b interprets backslash escapes,
which is how the colour variables work, so a cwd or model name containing a
literal \n — or a real newline, which jq decodes from the JSON — would split
the line and break both the width measurement and the two-line guarantee.
Parse the usage payload in one jq pass rather than ten. The status line runs on
every redraw and per-field parsing had become the dominant cost; this brings a
render back to roughly what it was before (~155ms vs ~115ms parent), the
remainder being the reset times the tiered version did not show at this width.
Fields are read one per line rather than via @tsv: tab is IFS whitespace, so
`read` collapses runs of it and an empty field — no scoped limit, which is the
common case — silently shifted every later field along by one.
Tests cover the API payload shapes including malformed and hostile input, the
width invariants (never exceed the usable width, never more than two lines,
never wrap unnecessarily), and a sweep over widths, branch lengths, model
names, cwd lengths and extra-usage. They restore the live usage cache on exit,
and remove the fixture outright when there was no cache to restore — otherwise
a test run would leave fabricated usage figures live for an hour.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-27 17:55:35 +01:00
|
|
|
|
[ "$sl_thinking" = "true" ] && thinking_on=true
|
2026-08-08 11:54:05 +01:00
|
|
|
|
|
|
|
|
|
|
# ===== Adaptive width tiers =====
|
|
|
|
|
|
#
|
Show the per-model usage limit, and size the status line by measurement
/usage reports a weekly limit scoped to a single model (Fable) that the
status line did not surface. It is absent from both the status line's stdin
JSON and the legacy seven_day_opus/seven_day_sonnet fields, which are always
null now; it lives in the usage API's .limits[] under kind "weekly_scoped".
Render it labelled by scope.model.display_name so it follows whichever model
the limit applies to, inside the 7d segment since both are weekly limits
sharing one reset time.
Fitting it exposed a problem with sizing by width tier. Tiers only know the
terminal width, so content that varies with session state — branch name, cwd,
model display name — could push a line past the edge and be clipped by the
renderer. Measure the assembled line instead (vis_len) and emit the richest of
four rate-limit variants that fits, on one line or two: with reset times, with
extra-usage credits, bars only, or bars without the per-model segment. Every
branch is fit-checked, including with wrapping disabled. Two lines render
correctly in the status line.
Cap the cwd basename as well: nothing else shortened line one, so a long
project directory could overflow it on its own.
Correct the usable width. Claude Code exports COLUMNS but applies the `padding`
setting on top of it rather than deducting it first, so a line sized to COLUMNS
is clipped; deduct padding on both sides plus a column of margin.
Sanitise payload data before rendering. printf %b interprets backslash escapes,
which is how the colour variables work, so a cwd or model name containing a
literal \n — or a real newline, which jq decodes from the JSON — would split
the line and break both the width measurement and the two-line guarantee.
Parse the usage payload in one jq pass rather than ten. The status line runs on
every redraw and per-field parsing had become the dominant cost; this brings a
render back to roughly what it was before (~155ms vs ~115ms parent), the
remainder being the reset times the tiered version did not show at this width.
Fields are read one per line rather than via @tsv: tab is IFS whitespace, so
`read` collapses runs of it and an empty field — no scoped limit, which is the
common case — silently shifted every later field along by one.
Tests cover the API payload shapes including malformed and hostile input, the
width invariants (never exceed the usable width, never more than two lines,
never wrap unnecessarily), and a sweep over widths, branch lengths, model
names, cwd lengths and extra-usage. They restore the live usage cache on exit,
and remove the fixture outright when there was no cache to restore — otherwise
a test run would leave fabricated usage figures live for an hour.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-27 17:55:35 +01:00
|
|
|
|
# Tiers now choose only how much of LINE ONE to show. What the rate-limit
|
|
|
|
|
|
# group contains, and whether it wraps, is decided by measurement at the end of
|
|
|
|
|
|
# the script, not here -- tiers cannot see how long the branch, cwd or model
|
|
|
|
|
|
# name is, which is how content used to end up clipped.
|
2026-08-08 11:54:05 +01:00
|
|
|
|
#
|
Show the per-model usage limit, and size the status line by measurement
/usage reports a weekly limit scoped to a single model (Fable) that the
status line did not surface. It is absent from both the status line's stdin
JSON and the legacy seven_day_opus/seven_day_sonnet fields, which are always
null now; it lives in the usage API's .limits[] under kind "weekly_scoped".
Render it labelled by scope.model.display_name so it follows whichever model
the limit applies to, inside the 7d segment since both are weekly limits
sharing one reset time.
Fitting it exposed a problem with sizing by width tier. Tiers only know the
terminal width, so content that varies with session state — branch name, cwd,
model display name — could push a line past the edge and be clipped by the
renderer. Measure the assembled line instead (vis_len) and emit the richest of
four rate-limit variants that fits, on one line or two: with reset times, with
extra-usage credits, bars only, or bars without the per-model segment. Every
branch is fit-checked, including with wrapping disabled. Two lines render
correctly in the status line.
Cap the cwd basename as well: nothing else shortened line one, so a long
project directory could overflow it on its own.
Correct the usable width. Claude Code exports COLUMNS but applies the `padding`
setting on top of it rather than deducting it first, so a line sized to COLUMNS
is clipped; deduct padding on both sides plus a column of margin.
Sanitise payload data before rendering. printf %b interprets backslash escapes,
which is how the colour variables work, so a cwd or model name containing a
literal \n — or a real newline, which jq decodes from the JSON — would split
the line and break both the width measurement and the two-line guarantee.
Parse the usage payload in one jq pass rather than ten. The status line runs on
every redraw and per-field parsing had become the dominant cost; this brings a
render back to roughly what it was before (~155ms vs ~115ms parent), the
remainder being the reset times the tiered version did not show at this width.
Fields are read one per line rather than via @tsv: tab is IFS whitespace, so
`read` collapses runs of it and an empty field — no scoped limit, which is the
common case — silently shifted every later field along by one.
Tests cover the API payload shapes including malformed and hostile input, the
width invariants (never exceed the usable width, never more than two lines,
never wrap unnecessarily), and a sweep over widths, branch lengths, model
names, cwd lengths and extra-usage. They restore the live usage cache on exit,
and remove the fixture outright when there was no cache to restore — otherwise
a test run would leave fabricated usage figures live for an hour.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-27 17:55:35 +01:00
|
|
|
|
# full (≥150): CWD, ahead/behind, "◆ thinking", cost
|
|
|
|
|
|
# wide (100–149): ahead/behind, "◆ thinking", cost
|
|
|
|
|
|
# split (76–99): short model, ahead/behind, "◆" symbol
|
|
|
|
|
|
# (reachable only with WRAP_NARROW=false; otherwise the wrap
|
|
|
|
|
|
# band below overrides this range to wide)
|
|
|
|
|
|
# narrow (<76): short model + branch + token only; no rate limits at all
|
2026-08-08 11:54:05 +01:00
|
|
|
|
#
|
Show the per-model usage limit, and size the status line by measurement
/usage reports a weekly limit scoped to a single model (Fable) that the
status line did not surface. It is absent from both the status line's stdin
JSON and the legacy seven_day_opus/seven_day_sonnet fields, which are always
null now; it lives in the usage API's .limits[] under kind "weekly_scoped".
Render it labelled by scope.model.display_name so it follows whichever model
the limit applies to, inside the 7d segment since both are weekly limits
sharing one reset time.
Fitting it exposed a problem with sizing by width tier. Tiers only know the
terminal width, so content that varies with session state — branch name, cwd,
model display name — could push a line past the edge and be clipped by the
renderer. Measure the assembled line instead (vis_len) and emit the richest of
four rate-limit variants that fits, on one line or two: with reset times, with
extra-usage credits, bars only, or bars without the per-model segment. Every
branch is fit-checked, including with wrapping disabled. Two lines render
correctly in the status line.
Cap the cwd basename as well: nothing else shortened line one, so a long
project directory could overflow it on its own.
Correct the usable width. Claude Code exports COLUMNS but applies the `padding`
setting on top of it rather than deducting it first, so a line sized to COLUMNS
is clipped; deduct padding on both sides plus a column of margin.
Sanitise payload data before rendering. printf %b interprets backslash escapes,
which is how the colour variables work, so a cwd or model name containing a
literal \n — or a real newline, which jq decodes from the JSON — would split
the line and break both the width measurement and the two-line guarantee.
Parse the usage payload in one jq pass rather than ten. The status line runs on
every redraw and per-field parsing had become the dominant cost; this brings a
render back to roughly what it was before (~155ms vs ~115ms parent), the
remainder being the reset times the tiered version did not show at this width.
Fields are read one per line rather than via @tsv: tab is IFS whitespace, so
`read` collapses runs of it and an empty field — no scoped limit, which is the
common case — silently shifted every later field along by one.
Tests cover the API payload shapes including malformed and hostile input, the
width invariants (never exceed the usable width, never more than two lines,
never wrap unnecessarily), and a sweep over widths, branch lengths, model
names, cwd lengths and extra-usage. They restore the live usage cache on exit,
and remove the fixture outright when there was no cache to restore — otherwise
a test run would leave fabricated usage figures live for an hour.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-27 17:55:35 +01:00
|
|
|
|
if [ "$USABLE_WIDTH" -ge 150 ] 2>/dev/null; then width_tier="full"
|
|
|
|
|
|
elif [ "$USABLE_WIDTH" -ge 100 ] 2>/dev/null; then width_tier="wide"
|
|
|
|
|
|
elif [ "$USABLE_WIDTH" -ge 76 ] 2>/dev/null; then width_tier="split"
|
2026-08-08 11:54:05 +01:00
|
|
|
|
else width_tier="narrow"
|
|
|
|
|
|
fi
|
|
|
|
|
|
|
Show the per-model usage limit, and size the status line by measurement
/usage reports a weekly limit scoped to a single model (Fable) that the
status line did not surface. It is absent from both the status line's stdin
JSON and the legacy seven_day_opus/seven_day_sonnet fields, which are always
null now; it lives in the usage API's .limits[] under kind "weekly_scoped".
Render it labelled by scope.model.display_name so it follows whichever model
the limit applies to, inside the 7d segment since both are weekly limits
sharing one reset time.
Fitting it exposed a problem with sizing by width tier. Tiers only know the
terminal width, so content that varies with session state — branch name, cwd,
model display name — could push a line past the edge and be clipped by the
renderer. Measure the assembled line instead (vis_len) and emit the richest of
four rate-limit variants that fits, on one line or two: with reset times, with
extra-usage credits, bars only, or bars without the per-model segment. Every
branch is fit-checked, including with wrapping disabled. Two lines render
correctly in the status line.
Cap the cwd basename as well: nothing else shortened line one, so a long
project directory could overflow it on its own.
Correct the usable width. Claude Code exports COLUMNS but applies the `padding`
setting on top of it rather than deducting it first, so a line sized to COLUMNS
is clipped; deduct padding on both sides plus a column of margin.
Sanitise payload data before rendering. printf %b interprets backslash escapes,
which is how the colour variables work, so a cwd or model name containing a
literal \n — or a real newline, which jq decodes from the JSON — would split
the line and break both the width measurement and the two-line guarantee.
Parse the usage payload in one jq pass rather than ten. The status line runs on
every redraw and per-field parsing had become the dominant cost; this brings a
render back to roughly what it was before (~155ms vs ~115ms parent), the
remainder being the reset times the tiered version did not show at this width.
Fields are read one per line rather than via @tsv: tab is IFS whitespace, so
`read` collapses runs of it and an empty field — no scoped limit, which is the
common case — silently shifted every later field along by one.
Tests cover the API payload shapes including malformed and hostile input, the
width invariants (never exceed the usable width, never more than two lines,
never wrap unnecessarily), and a sweep over widths, branch lengths, model
names, cwd lengths and extra-usage. They restore the live usage cache on exit,
and remove the fixture outright when there was no cache to restore — otherwise
a test run would leave fabricated usage figures live for an hour.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-27 17:55:35 +01:00
|
|
|
|
# Two-line mode. Between WRAP_FLOOR and WRAP_MIN_WIDTH there isn't room for
|
|
|
|
|
|
# everything on one line, but there IS room across two — so instead of dropping
|
|
|
|
|
|
# the rate-limit group we break before it and render the wide-tier content.
|
|
|
|
|
|
# Below WRAP_FLOOR the narrow tier applies instead, which drops the rate-limit
|
|
|
|
|
|
# group entirely -- so there is nothing to wrap and no second line to put it on.
|
|
|
|
|
|
WRAP_FLOOR=68
|
|
|
|
|
|
wrap_mode=false
|
|
|
|
|
|
if $WRAP_NARROW \
|
|
|
|
|
|
&& [ "$USABLE_WIDTH" -lt "$WRAP_MIN_WIDTH" ] 2>/dev/null \
|
|
|
|
|
|
&& [ "$USABLE_WIDTH" -ge "$WRAP_FLOOR" ] 2>/dev/null; then
|
|
|
|
|
|
wrap_mode=true
|
|
|
|
|
|
width_tier="wide"
|
|
|
|
|
|
fi
|
|
|
|
|
|
|
2026-08-08 11:54:05 +01:00
|
|
|
|
# Shorten model name for tight spaces
|
|
|
|
|
|
short_model() {
|
|
|
|
|
|
case "$1" in
|
|
|
|
|
|
*Opus*) echo "Opus" ;;
|
|
|
|
|
|
*Sonnet*) echo "Sonnet" ;;
|
|
|
|
|
|
*Haiku*) echo "Haiku" ;;
|
|
|
|
|
|
*) echo "$1" | awk '{print $NF}' ;; # last word
|
|
|
|
|
|
esac
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
# ===== Build output =====
|
Show the per-model usage limit, and size the status line by measurement
/usage reports a weekly limit scoped to a single model (Fable) that the
status line did not surface. It is absent from both the status line's stdin
JSON and the legacy seven_day_opus/seven_day_sonnet fields, which are always
null now; it lives in the usage API's .limits[] under kind "weekly_scoped".
Render it labelled by scope.model.display_name so it follows whichever model
the limit applies to, inside the 7d segment since both are weekly limits
sharing one reset time.
Fitting it exposed a problem with sizing by width tier. Tiers only know the
terminal width, so content that varies with session state — branch name, cwd,
model display name — could push a line past the edge and be clipped by the
renderer. Measure the assembled line instead (vis_len) and emit the richest of
four rate-limit variants that fits, on one line or two: with reset times, with
extra-usage credits, bars only, or bars without the per-model segment. Every
branch is fit-checked, including with wrapping disabled. Two lines render
correctly in the status line.
Cap the cwd basename as well: nothing else shortened line one, so a long
project directory could overflow it on its own.
Correct the usable width. Claude Code exports COLUMNS but applies the `padding`
setting on top of it rather than deducting it first, so a line sized to COLUMNS
is clipped; deduct padding on both sides plus a column of margin.
Sanitise payload data before rendering. printf %b interprets backslash escapes,
which is how the colour variables work, so a cwd or model name containing a
literal \n — or a real newline, which jq decodes from the JSON — would split
the line and break both the width measurement and the two-line guarantee.
Parse the usage payload in one jq pass rather than ten. The status line runs on
every redraw and per-field parsing had become the dominant cost; this brings a
render back to roughly what it was before (~155ms vs ~115ms parent), the
remainder being the reset times the tiered version did not show at this width.
Fields are read one per line rather than via @tsv: tab is IFS whitespace, so
`read` collapses runs of it and an empty field — no scoped limit, which is the
common case — silently shifted every later field along by one.
Tests cover the API payload shapes including malformed and hostile input, the
width invariants (never exceed the usable width, never more than two lines,
never wrap unnecessarily), and a sweep over widths, branch lengths, model
names, cwd lengths and extra-usage. They restore the live usage cache on exit,
and remove the fixture outright when there was no cache to restore — otherwise
a test run would leave fabricated usage figures live for an hour.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-27 17:55:35 +01:00
|
|
|
|
# Explicitly empty: bash imports same-named environment variables, and a
|
|
|
|
|
|
# stray $rl_lean would otherwise leak into the rendered line.
|
2026-08-08 11:54:05 +01:00
|
|
|
|
out=""
|
Show the per-model usage limit, and size the status line by measurement
/usage reports a weekly limit scoped to a single model (Fable) that the
status line did not surface. It is absent from both the status line's stdin
JSON and the legacy seven_day_opus/seven_day_sonnet fields, which are always
null now; it lives in the usage API's .limits[] under kind "weekly_scoped".
Render it labelled by scope.model.display_name so it follows whichever model
the limit applies to, inside the 7d segment since both are weekly limits
sharing one reset time.
Fitting it exposed a problem with sizing by width tier. Tiers only know the
terminal width, so content that varies with session state — branch name, cwd,
model display name — could push a line past the edge and be clipped by the
renderer. Measure the assembled line instead (vis_len) and emit the richest of
four rate-limit variants that fits, on one line or two: with reset times, with
extra-usage credits, bars only, or bars without the per-model segment. Every
branch is fit-checked, including with wrapping disabled. Two lines render
correctly in the status line.
Cap the cwd basename as well: nothing else shortened line one, so a long
project directory could overflow it on its own.
Correct the usable width. Claude Code exports COLUMNS but applies the `padding`
setting on top of it rather than deducting it first, so a line sized to COLUMNS
is clipped; deduct padding on both sides plus a column of margin.
Sanitise payload data before rendering. printf %b interprets backslash escapes,
which is how the colour variables work, so a cwd or model name containing a
literal \n — or a real newline, which jq decodes from the JSON — would split
the line and break both the width measurement and the two-line guarantee.
Parse the usage payload in one jq pass rather than ten. The status line runs on
every redraw and per-field parsing had become the dominant cost; this brings a
render back to roughly what it was before (~155ms vs ~115ms parent), the
remainder being the reset times the tiered version did not show at this width.
Fields are read one per line rather than via @tsv: tab is IFS whitespace, so
`read` collapses runs of it and an empty field — no scoped limit, which is the
common case — silently shifted every later field along by one.
Tests cover the API payload shapes including malformed and hostile input, the
width invariants (never exceed the usable width, never more than two lines,
never wrap unnecessarily), and a sweep over widths, branch lengths, model
names, cwd lengths and extra-usage. They restore the live usage cache on exit,
and remove the fixture outright when there was no cache to restore — otherwise
a test run would leave fabricated usage figures live for an hour.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-27 17:55:35 +01:00
|
|
|
|
rl_bare=""
|
|
|
|
|
|
rl_lean=""
|
|
|
|
|
|
rl_mid=""
|
|
|
|
|
|
rl_rich=""
|
2026-08-08 11:54:05 +01:00
|
|
|
|
|
|
|
|
|
|
# Model — color by family
|
|
|
|
|
|
model_color="$blue"
|
|
|
|
|
|
case "$model_name" in
|
|
|
|
|
|
*Opus*) model_color="$amber" ;;
|
|
|
|
|
|
*Haiku*) model_color="$cyan" ;;
|
|
|
|
|
|
esac
|
|
|
|
|
|
|
|
|
|
|
|
display_model="$model_name"
|
Show the per-model usage limit, and size the status line by measurement
/usage reports a weekly limit scoped to a single model (Fable) that the
status line did not surface. It is absent from both the status line's stdin
JSON and the legacy seven_day_opus/seven_day_sonnet fields, which are always
null now; it lives in the usage API's .limits[] under kind "weekly_scoped".
Render it labelled by scope.model.display_name so it follows whichever model
the limit applies to, inside the 7d segment since both are weekly limits
sharing one reset time.
Fitting it exposed a problem with sizing by width tier. Tiers only know the
terminal width, so content that varies with session state — branch name, cwd,
model display name — could push a line past the edge and be clipped by the
renderer. Measure the assembled line instead (vis_len) and emit the richest of
four rate-limit variants that fits, on one line or two: with reset times, with
extra-usage credits, bars only, or bars without the per-model segment. Every
branch is fit-checked, including with wrapping disabled. Two lines render
correctly in the status line.
Cap the cwd basename as well: nothing else shortened line one, so a long
project directory could overflow it on its own.
Correct the usable width. Claude Code exports COLUMNS but applies the `padding`
setting on top of it rather than deducting it first, so a line sized to COLUMNS
is clipped; deduct padding on both sides plus a column of margin.
Sanitise payload data before rendering. printf %b interprets backslash escapes,
which is how the colour variables work, so a cwd or model name containing a
literal \n — or a real newline, which jq decodes from the JSON — would split
the line and break both the width measurement and the two-line guarantee.
Parse the usage payload in one jq pass rather than ten. The status line runs on
every redraw and per-field parsing had become the dominant cost; this brings a
render back to roughly what it was before (~155ms vs ~115ms parent), the
remainder being the reset times the tiered version did not show at this width.
Fields are read one per line rather than via @tsv: tab is IFS whitespace, so
`read` collapses runs of it and an empty field — no scoped limit, which is the
common case — silently shifted every later field along by one.
Tests cover the API payload shapes including malformed and hostile input, the
width invariants (never exceed the usable width, never more than two lines,
never wrap unnecessarily), and a sweep over widths, branch lengths, model
names, cwd lengths and extra-usage. They restore the live usage cache on exit,
and remove the fixture outright when there was no cache to restore — otherwise
a test run would leave fabricated usage figures live for an hour.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-27 17:55:35 +01:00
|
|
|
|
# split/narrow: shorten so the 5h + 7d bars still fit on one line.
|
|
|
|
|
|
# wrap mode too: it renders wide-tier content at a split-tier width, and a long
|
|
|
|
|
|
# display name ("Opus 5 (1M context)" is 19 cols) would overflow line one.
|
|
|
|
|
|
if [ "$width_tier" = "split" -o "$width_tier" = "narrow" ] || $wrap_mode; then
|
|
|
|
|
|
display_model=$(short_model "$model_name")
|
|
|
|
|
|
fi
|
|
|
|
|
|
out+="${model_color}$(esc_data "$display_model")${reset}"
|
2026-08-08 11:54:05 +01:00
|
|
|
|
|
|
|
|
|
|
# CWD — full tier only (branch name gives enough context below that)
|
|
|
|
|
|
if [ "$width_tier" = "full" ] && [ -n "$cwd" ]; then
|
Show the per-model usage limit, and size the status line by measurement
/usage reports a weekly limit scoped to a single model (Fable) that the
status line did not surface. It is absent from both the status line's stdin
JSON and the legacy seven_day_opus/seven_day_sonnet fields, which are always
null now; it lives in the usage API's .limits[] under kind "weekly_scoped".
Render it labelled by scope.model.display_name so it follows whichever model
the limit applies to, inside the 7d segment since both are weekly limits
sharing one reset time.
Fitting it exposed a problem with sizing by width tier. Tiers only know the
terminal width, so content that varies with session state — branch name, cwd,
model display name — could push a line past the edge and be clipped by the
renderer. Measure the assembled line instead (vis_len) and emit the richest of
four rate-limit variants that fits, on one line or two: with reset times, with
extra-usage credits, bars only, or bars without the per-model segment. Every
branch is fit-checked, including with wrapping disabled. Two lines render
correctly in the status line.
Cap the cwd basename as well: nothing else shortened line one, so a long
project directory could overflow it on its own.
Correct the usable width. Claude Code exports COLUMNS but applies the `padding`
setting on top of it rather than deducting it first, so a line sized to COLUMNS
is clipped; deduct padding on both sides plus a column of margin.
Sanitise payload data before rendering. printf %b interprets backslash escapes,
which is how the colour variables work, so a cwd or model name containing a
literal \n — or a real newline, which jq decodes from the JSON — would split
the line and break both the width measurement and the two-line guarantee.
Parse the usage payload in one jq pass rather than ten. The status line runs on
every redraw and per-field parsing had become the dominant cost; this brings a
render back to roughly what it was before (~155ms vs ~115ms parent), the
remainder being the reset times the tiered version did not show at this width.
Fields are read one per line rather than via @tsv: tab is IFS whitespace, so
`read` collapses runs of it and an empty field — no scoped limit, which is the
common case — silently shifted every later field along by one.
Tests cover the API payload shapes including malformed and hostile input, the
width invariants (never exceed the usable width, never more than two lines,
never wrap unnecessarily), and a sweep over widths, branch lengths, model
names, cwd lengths and extra-usage. They restore the live usage cache on exit,
and remove the fixture outright when there was no cache to restore — otherwise
a test run would leave fabricated usage figures live for an hour.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-27 17:55:35 +01:00
|
|
|
|
# Capped like the branch name: nothing else shortens line one, so an
|
|
|
|
|
|
# unusually long project directory would push it past the terminal edge.
|
|
|
|
|
|
display_dir=$(truncate_str "${cwd##*/}" "$CWD_MAX_LEN")
|
|
|
|
|
|
out+="${sep}${dim}$(esc_data "$display_dir")${reset}"
|
2026-08-08 11:54:05 +01:00
|
|
|
|
fi
|
|
|
|
|
|
|
Measure line one's branch budget instead of assuming 56 columns
In wrap mode the branch is truncated to whatever the rest of line one leaves,
and "the rest" was a flat 56 columns. It is not flat: the token counts and the
cost figure vary with the session, and with a four-digit cost the real width is
57, so line one came out exactly one column over and the renderer clipped it.
Reachable at COLUMNS 73-81 with padding 2 — a narrow split pane.
Add up the segments that follow the branch instead, from the same values that
render them, and count the ahead/behind markers too, which the constant also
ignored. That needs the cost formatted and the token bar width chosen before
the git segment is built, so both move up; neither depends on anything in
between.
The budget is now exact rather than approximate, so the branch also gets the
columns the old constant was over-reserving when the cost was short.
Tests: the large-cost sweep only used the short branch 'main', where the
overflow cannot show. Repeat it with the long branch, which fails on the old
budget at usable 68, 69, 70 and 75 by exactly one column.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-31 19:01:24 +01:00
|
|
|
|
bar_w="$TOKEN_BAR_WIDTH"
|
|
|
|
|
|
[ "$width_tier" = "wide" ] && bar_w=6
|
|
|
|
|
|
[ "$width_tier" = "split" ] && bar_w=5
|
|
|
|
|
|
[ "$width_tier" = "narrow" ] && bar_w=4
|
|
|
|
|
|
|
2026-08-08 11:54:05 +01:00
|
|
|
|
# Git branch + dirty + ahead/behind
|
|
|
|
|
|
if $SHOW_GIT && [ -n "$cwd" ]; then
|
|
|
|
|
|
git_info=$(get_git_info "$cwd")
|
|
|
|
|
|
if [ -n "$git_info" ]; then
|
|
|
|
|
|
IFS=$'\t' read -r g_branch g_dirty g_ahead g_behind <<< "$git_info"
|
|
|
|
|
|
|
|
|
|
|
|
# Progressively tighten branch truncation
|
|
|
|
|
|
local_max="$BRANCH_MAX_LEN"
|
|
|
|
|
|
[ "$width_tier" = "wide" ] && local_max=24
|
|
|
|
|
|
[ "$width_tier" = "split" ] && local_max=18
|
|
|
|
|
|
[ "$width_tier" = "narrow" ] && local_max=12
|
Measure line one's branch budget instead of assuming 56 columns
In wrap mode the branch is truncated to whatever the rest of line one leaves,
and "the rest" was a flat 56 columns. It is not flat: the token counts and the
cost figure vary with the session, and with a four-digit cost the real width is
57, so line one came out exactly one column over and the renderer clipped it.
Reachable at COLUMNS 73-81 with padding 2 — a narrow split pane.
Add up the segments that follow the branch instead, from the same values that
render them, and count the ahead/behind markers too, which the constant also
ignored. That needs the cost formatted and the token bar width chosen before
the git segment is built, so both move up; neither depends on anything in
between.
The budget is now exact rather than approximate, so the branch also gets the
columns the old constant was over-reserving when the cost was short.
Tests: the large-cost sweep only used the short branch 'main', where the
overflow cannot show. Repeat it with the long branch, which fails on the old
budget at usable 68, 69, 70 and 75 by exactly one column.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-31 19:01:24 +01:00
|
|
|
|
# In wrap mode line one is
|
|
|
|
|
|
# model │ ⎇ branch ✔ ↑1 │ <bar> used/total pct% │ ◇ thinking │ $cost
|
|
|
|
|
|
# and the branch gets whatever the rest of it leaves. Everything after
|
|
|
|
|
|
# the branch is appended below, so its width is added up here rather
|
|
|
|
|
|
# than assumed: a flat 56 columns was three short of a four-digit cost,
|
|
|
|
|
|
# which pushed line one past the edge to be clipped by the renderer.
|
Show the per-model usage limit, and size the status line by measurement
/usage reports a weekly limit scoped to a single model (Fable) that the
status line did not surface. It is absent from both the status line's stdin
JSON and the legacy seven_day_opus/seven_day_sonnet fields, which are always
null now; it lives in the usage API's .limits[] under kind "weekly_scoped".
Render it labelled by scope.model.display_name so it follows whichever model
the limit applies to, inside the 7d segment since both are weekly limits
sharing one reset time.
Fitting it exposed a problem with sizing by width tier. Tiers only know the
terminal width, so content that varies with session state — branch name, cwd,
model display name — could push a line past the edge and be clipped by the
renderer. Measure the assembled line instead (vis_len) and emit the richest of
four rate-limit variants that fits, on one line or two: with reset times, with
extra-usage credits, bars only, or bars without the per-model segment. Every
branch is fit-checked, including with wrapping disabled. Two lines render
correctly in the status line.
Cap the cwd basename as well: nothing else shortened line one, so a long
project directory could overflow it on its own.
Correct the usable width. Claude Code exports COLUMNS but applies the `padding`
setting on top of it rather than deducting it first, so a line sized to COLUMNS
is clipped; deduct padding on both sides plus a column of margin.
Sanitise payload data before rendering. printf %b interprets backslash escapes,
which is how the colour variables work, so a cwd or model name containing a
literal \n — or a real newline, which jq decodes from the JSON — would split
the line and break both the width measurement and the two-line guarantee.
Parse the usage payload in one jq pass rather than ten. The status line runs on
every redraw and per-field parsing had become the dominant cost; this brings a
render back to roughly what it was before (~155ms vs ~115ms parent), the
remainder being the reset times the tiered version did not show at this width.
Fields are read one per line rather than via @tsv: tab is IFS whitespace, so
`read` collapses runs of it and an empty field — no scoped limit, which is the
common case — silently shifted every later field along by one.
Tests cover the API payload shapes including malformed and hostile input, the
width invariants (never exceed the usable width, never more than two lines,
never wrap unnecessarily), and a sweep over widths, branch lengths, model
names, cwd lengths and extra-usage. They restore the live usage cache on exit,
and remove the fixture outright when there was no cache to restore — otherwise
a test run would leave fabricated usage figures live for an hour.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-27 17:55:35 +01:00
|
|
|
|
if $wrap_mode; then
|
Measure line one's branch budget instead of assuming 56 columns
In wrap mode the branch is truncated to whatever the rest of line one leaves,
and "the rest" was a flat 56 columns. It is not flat: the token counts and the
cost figure vary with the session, and with a four-digit cost the real width is
57, so line one came out exactly one column over and the renderer clipped it.
Reachable at COLUMNS 73-81 with padding 2 — a narrow split pane.
Add up the segments that follow the branch instead, from the same values that
render them, and count the ahead/behind markers too, which the constant also
ignored. That needs the cost formatted and the token bar width chosen before
the git segment is built, so both move up; neither depends on anything in
between.
The budget is now exact rather than approximate, so the branch also gets the
columns the old constant was over-reserving when the cost was short.
Tests: the large-cost sweep only used the short branch 'main', where the
overflow cannot show. Repeat it with the long branch, which fails on the old
budget at usable 68, 69, 70 and 75 by exactly one column.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-31 19:01:24 +01:00
|
|
|
|
tail_len=$(( 5 + 2 )) # "│ ⎇ " and the dirty mark
|
|
|
|
|
|
[ "${g_ahead:-0}" -gt 0 ] && tail_len=$(( tail_len + 2 + ${#g_ahead} ))
|
|
|
|
|
|
[ "${g_behind:-0}" -gt 0 ] && tail_len=$(( tail_len + 2 + ${#g_behind} ))
|
|
|
|
|
|
$SHOW_TOKENS && tail_len=$(( tail_len + 3 + bar_w + 1 \
|
|
|
|
|
|
+ ${#used_tokens} + 1 + ${#total_tokens} + 1 + ${#pct_used} + 1 ))
|
|
|
|
|
|
$SHOW_THINKING && tail_len=$(( tail_len + 3 + 10 ))
|
|
|
|
|
|
[ -n "$cost_fmt" ] && tail_len=$(( tail_len + 3 + 1 + ${#cost_fmt} ))
|
|
|
|
|
|
local_max=$(( USABLE_WIDTH - $(vis_len "$out") - tail_len ))
|
Show the per-model usage limit, and size the status line by measurement
/usage reports a weekly limit scoped to a single model (Fable) that the
status line did not surface. It is absent from both the status line's stdin
JSON and the legacy seven_day_opus/seven_day_sonnet fields, which are always
null now; it lives in the usage API's .limits[] under kind "weekly_scoped".
Render it labelled by scope.model.display_name so it follows whichever model
the limit applies to, inside the 7d segment since both are weekly limits
sharing one reset time.
Fitting it exposed a problem with sizing by width tier. Tiers only know the
terminal width, so content that varies with session state — branch name, cwd,
model display name — could push a line past the edge and be clipped by the
renderer. Measure the assembled line instead (vis_len) and emit the richest of
four rate-limit variants that fits, on one line or two: with reset times, with
extra-usage credits, bars only, or bars without the per-model segment. Every
branch is fit-checked, including with wrapping disabled. Two lines render
correctly in the status line.
Cap the cwd basename as well: nothing else shortened line one, so a long
project directory could overflow it on its own.
Correct the usable width. Claude Code exports COLUMNS but applies the `padding`
setting on top of it rather than deducting it first, so a line sized to COLUMNS
is clipped; deduct padding on both sides plus a column of margin.
Sanitise payload data before rendering. printf %b interprets backslash escapes,
which is how the colour variables work, so a cwd or model name containing a
literal \n — or a real newline, which jq decodes from the JSON — would split
the line and break both the width measurement and the two-line guarantee.
Parse the usage payload in one jq pass rather than ten. The status line runs on
every redraw and per-field parsing had become the dominant cost; this brings a
render back to roughly what it was before (~155ms vs ~115ms parent), the
remainder being the reset times the tiered version did not show at this width.
Fields are read one per line rather than via @tsv: tab is IFS whitespace, so
`read` collapses runs of it and an empty field — no scoped limit, which is the
common case — silently shifted every later field along by one.
Tests cover the API payload shapes including malformed and hostile input, the
width invariants (never exceed the usable width, never more than two lines,
never wrap unnecessarily), and a sweep over widths, branch lengths, model
names, cwd lengths and extra-usage. They restore the live usage cache on exit,
and remove the fixture outright when there was no cache to restore — otherwise
a test run would leave fabricated usage figures live for an hour.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-27 17:55:35 +01:00
|
|
|
|
[ "$local_max" -lt 8 ] && local_max=8
|
|
|
|
|
|
[ "$local_max" -gt 24 ] && local_max=24
|
|
|
|
|
|
fi
|
2026-08-08 11:54:05 +01:00
|
|
|
|
g_branch_display=$(truncate_str "$g_branch" "$local_max")
|
|
|
|
|
|
|
Show the per-model usage limit, and size the status line by measurement
/usage reports a weekly limit scoped to a single model (Fable) that the
status line did not surface. It is absent from both the status line's stdin
JSON and the legacy seven_day_opus/seven_day_sonnet fields, which are always
null now; it lives in the usage API's .limits[] under kind "weekly_scoped".
Render it labelled by scope.model.display_name so it follows whichever model
the limit applies to, inside the 7d segment since both are weekly limits
sharing one reset time.
Fitting it exposed a problem with sizing by width tier. Tiers only know the
terminal width, so content that varies with session state — branch name, cwd,
model display name — could push a line past the edge and be clipped by the
renderer. Measure the assembled line instead (vis_len) and emit the richest of
four rate-limit variants that fits, on one line or two: with reset times, with
extra-usage credits, bars only, or bars without the per-model segment. Every
branch is fit-checked, including with wrapping disabled. Two lines render
correctly in the status line.
Cap the cwd basename as well: nothing else shortened line one, so a long
project directory could overflow it on its own.
Correct the usable width. Claude Code exports COLUMNS but applies the `padding`
setting on top of it rather than deducting it first, so a line sized to COLUMNS
is clipped; deduct padding on both sides plus a column of margin.
Sanitise payload data before rendering. printf %b interprets backslash escapes,
which is how the colour variables work, so a cwd or model name containing a
literal \n — or a real newline, which jq decodes from the JSON — would split
the line and break both the width measurement and the two-line guarantee.
Parse the usage payload in one jq pass rather than ten. The status line runs on
every redraw and per-field parsing had become the dominant cost; this brings a
render back to roughly what it was before (~155ms vs ~115ms parent), the
remainder being the reset times the tiered version did not show at this width.
Fields are read one per line rather than via @tsv: tab is IFS whitespace, so
`read` collapses runs of it and an empty field — no scoped limit, which is the
common case — silently shifted every later field along by one.
Tests cover the API payload shapes including malformed and hostile input, the
width invariants (never exceed the usable width, never more than two lines,
never wrap unnecessarily), and a sweep over widths, branch lengths, model
names, cwd lengths and extra-usage. They restore the live usage cache on exit,
and remove the fixture outright when there was no cache to restore — otherwise
a test run would leave fabricated usage figures live for an hour.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-27 17:55:35 +01:00
|
|
|
|
out+="${sep}${dim}⎇${reset} ${magenta}$(esc_data "$g_branch_display")${reset}"
|
2026-08-08 11:54:05 +01:00
|
|
|
|
|
|
|
|
|
|
if [ "$g_dirty" = "dirty" ]; then
|
|
|
|
|
|
out+=" ${red}✗${reset}"
|
|
|
|
|
|
else
|
|
|
|
|
|
out+=" ${green}✔${reset}"
|
|
|
|
|
|
fi
|
|
|
|
|
|
|
|
|
|
|
|
# Ahead/behind: shown in all tiers except narrow (only if non-zero)
|
|
|
|
|
|
if [ "$width_tier" != "narrow" ]; then
|
|
|
|
|
|
[ "${g_ahead:-0}" -gt 0 ] && out+=" ${green}↑${g_ahead}${reset}"
|
|
|
|
|
|
[ "${g_behind:-0}" -gt 0 ] && out+=" ${orange}↓${g_behind}${reset}"
|
|
|
|
|
|
fi
|
|
|
|
|
|
fi
|
|
|
|
|
|
fi
|
|
|
|
|
|
|
|
|
|
|
|
# Token bar
|
|
|
|
|
|
if $SHOW_TOKENS; then
|
|
|
|
|
|
token_bar=$(build_bar "$pct_used" "$bar_w")
|
|
|
|
|
|
out+="${sep}${token_bar} ${orange}${used_tokens}${dim}/${reset}${white}${total_tokens}${reset} ${dim}${pct_used}%${reset}"
|
|
|
|
|
|
fi
|
|
|
|
|
|
|
|
|
|
|
|
# Thinking:
|
|
|
|
|
|
# full/wide → "◆ thinking" / "◇ thinking" (label)
|
|
|
|
|
|
# split → "◆" / "◇" (symbol only, saves ~9 chars)
|
|
|
|
|
|
# narrow → hidden
|
|
|
|
|
|
if $SHOW_THINKING && [ "$width_tier" != "narrow" ]; then
|
|
|
|
|
|
out+="${sep}"
|
|
|
|
|
|
if $thinking_on; then
|
|
|
|
|
|
if [ "$width_tier" = "split" ]; then out+="${amber}◆${reset}"
|
|
|
|
|
|
else out+="${amber}◆ thinking${reset}"; fi
|
|
|
|
|
|
else
|
|
|
|
|
|
if [ "$width_tier" = "split" ]; then out+="${dim}◇${reset}"
|
|
|
|
|
|
else out+="${dim}◇ thinking${reset}"; fi
|
|
|
|
|
|
fi
|
|
|
|
|
|
fi
|
|
|
|
|
|
|
|
|
|
|
|
# Session cost — wide/full only
|
Measure line one's branch budget instead of assuming 56 columns
In wrap mode the branch is truncated to whatever the rest of line one leaves,
and "the rest" was a flat 56 columns. It is not flat: the token counts and the
cost figure vary with the session, and with a four-digit cost the real width is
57, so line one came out exactly one column over and the renderer clipped it.
Reachable at COLUMNS 73-81 with padding 2 — a narrow split pane.
Add up the segments that follow the branch instead, from the same values that
render them, and count the ahead/behind markers too, which the constant also
ignored. That needs the cost formatted and the token bar width chosen before
the git segment is built, so both move up; neither depends on anything in
between.
The budget is now exact rather than approximate, so the branch also gets the
columns the old constant was over-reserving when the cost was short.
Tests: the large-cost sweep only used the short branch 'main', where the
overflow cannot show. Repeat it with the long branch, which fails on the old
budget at usable 68, 69, 70 and 75 by exactly one column.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-31 19:01:24 +01:00
|
|
|
|
if [ -n "$cost_fmt" ] && [ "$width_tier" = "wide" -o "$width_tier" = "full" ]; then
|
|
|
|
|
|
out+="${sep}${dim}\$${cost_fmt}${reset}"
|
2026-08-08 11:54:05 +01:00
|
|
|
|
fi
|
|
|
|
|
|
|
2026-09-02 13:01:12 +01:00
|
|
|
|
# ===== Rate limits (API, cached USAGE_CACHE_SECS) =====
|
Show the per-model usage limit, and size the status line by measurement
/usage reports a weekly limit scoped to a single model (Fable) that the
status line did not surface. It is absent from both the status line's stdin
JSON and the legacy seven_day_opus/seven_day_sonnet fields, which are always
null now; it lives in the usage API's .limits[] under kind "weekly_scoped".
Render it labelled by scope.model.display_name so it follows whichever model
the limit applies to, inside the 7d segment since both are weekly limits
sharing one reset time.
Fitting it exposed a problem with sizing by width tier. Tiers only know the
terminal width, so content that varies with session state — branch name, cwd,
model display name — could push a line past the edge and be clipped by the
renderer. Measure the assembled line instead (vis_len) and emit the richest of
four rate-limit variants that fits, on one line or two: with reset times, with
extra-usage credits, bars only, or bars without the per-model segment. Every
branch is fit-checked, including with wrapping disabled. Two lines render
correctly in the status line.
Cap the cwd basename as well: nothing else shortened line one, so a long
project directory could overflow it on its own.
Correct the usable width. Claude Code exports COLUMNS but applies the `padding`
setting on top of it rather than deducting it first, so a line sized to COLUMNS
is clipped; deduct padding on both sides plus a column of margin.
Sanitise payload data before rendering. printf %b interprets backslash escapes,
which is how the colour variables work, so a cwd or model name containing a
literal \n — or a real newline, which jq decodes from the JSON — would split
the line and break both the width measurement and the two-line guarantee.
Parse the usage payload in one jq pass rather than ten. The status line runs on
every redraw and per-field parsing had become the dominant cost; this brings a
render back to roughly what it was before (~155ms vs ~115ms parent), the
remainder being the reset times the tiered version did not show at this width.
Fields are read one per line rather than via @tsv: tab is IFS whitespace, so
`read` collapses runs of it and an empty field — no scoped limit, which is the
common case — silently shifted every later field along by one.
Tests cover the API payload shapes including malformed and hostile input, the
width invariants (never exceed the usable width, never more than two lines,
never wrap unnecessarily), and a sweep over widths, branch lengths, model
names, cwd lengths and extra-usage. They restore the live usage cache on exit,
and remove the fixture outright when there was no cache to restore — otherwise
a test run would leave fabricated usage figures live for an hour.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-27 17:55:35 +01:00
|
|
|
|
# Built at every tier except narrow. Which variant is actually emitted,
|
|
|
|
|
|
# and on how many lines, is decided by the measured ladder at the end.
|
2026-08-08 11:54:05 +01:00
|
|
|
|
if $SHOW_RATE_LIMITS && [ "$width_tier" != "narrow" ]; then
|
Wrap the status line rather than drop the per-model limit
The measured ladder tried every single-line variant before considering a
second line, so whenever line one plus the full group did not fit — from
around 92 usable columns upwards, depending on how long the branch, cwd and
model names are — it emitted rl_bare (5h and 7d only) and silently dropped
the per-model (Fable) bar that is the main reason the group is worth
rendering. On a 110-column laptop the limit was invisible.
Reorder so a second line beats losing that bar: one line rich/mid/lean, then
wrap, and rl_bare only when WRAP_NARROW is false. Reset times and
extra-usage credits are still given up rather than wrapped for, which makes
the rendered content non-monotone in width; the comment on the ladder spells
that out.
Also stop the tests writing fixtures to the caches the live status line
reads. ~/.claude/statusline.sh is a symlink to the script, so a test run was
visibly rendering fabricated usage — a Fable bar at 10%, extra-usage credits
of $1234.56/$2000.00 — in whatever session happened to be open, and a run
killed before its trap fired would have left that in place for an hour. The
cache directory is now $STATUSLINE_CACHE_DIR (default /tmp/claude) and each
suite points it at a temporary directory, which also removes the
backup/restore dance and the deletion of the live git-status cache.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-31 17:28:55 +01:00
|
|
|
|
api_cache="$CACHE_DIR/statusline-usage-cache.json"
|
Back off after a failed usage fetch instead of retrying every redraw
A failed fetch left the cache mtime untouched, so with no network every
redraw — and a redraw happens on every keystroke — tried curl again and
could wait --max-time 10 for it. Cutting the TTL from an hour to five
minutes made that start 55 minutes sooner, so it is worth fixing now.
The marker is a separate file rather than a touch of the cached
response: a cold /tmp has no response to touch, and that is exactly the
case with no stale data to fall back on, where the timeout is paid in
full with nothing to show for it.
It goes down before the request and is cleared when that lands, not
written afterwards. A fetch is in flight for as long as curl takes to
give up, and this cache dir is shared by every session, so redraws
starting inside that window are real rather than hypothetical: they now
serve stale data instead of each launching their own doomed request.
Every attempt writes it, so a repeated failure restarts the backoff, and
only an attempt does, so the redraws it suppresses cannot keep
re-stamping it and it always lapses.
Missing credentials stays outside the backoff — it costs no timeout, and
the redraw after a login should show the bars rather than wait out a
retry window it had no part in earning.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-09-02 13:12:45 +01:00
|
|
|
|
fail_marker="$CACHE_DIR/statusline-usage-fail"
|
2026-08-08 11:54:05 +01:00
|
|
|
|
needs_refresh=true
|
|
|
|
|
|
usage_data=""
|
|
|
|
|
|
|
|
|
|
|
|
if [ -f "$api_cache" ]; then
|
|
|
|
|
|
cache_mtime=$(stat -c %Y "$api_cache" 2>/dev/null || stat -f %m "$api_cache" 2>/dev/null)
|
|
|
|
|
|
now=$(date +%s)
|
|
|
|
|
|
cache_age=$(( now - cache_mtime ))
|
2026-09-02 13:01:12 +01:00
|
|
|
|
if [ "$cache_age" -lt "$USAGE_CACHE_SECS" ]; then
|
2026-08-08 11:54:05 +01:00
|
|
|
|
cached_content=$(cat "$api_cache" 2>/dev/null)
|
|
|
|
|
|
# Only use cache if it's valid data (not an error response)
|
|
|
|
|
|
if [ -n "$cached_content" ] && ! echo "$cached_content" | jq -e '.error' >/dev/null 2>&1; then
|
|
|
|
|
|
needs_refresh=false
|
|
|
|
|
|
usage_data="$cached_content"
|
|
|
|
|
|
fi
|
|
|
|
|
|
fi
|
|
|
|
|
|
fi
|
|
|
|
|
|
|
|
|
|
|
|
if $needs_refresh; then
|
Back off after a failed usage fetch instead of retrying every redraw
A failed fetch left the cache mtime untouched, so with no network every
redraw — and a redraw happens on every keystroke — tried curl again and
could wait --max-time 10 for it. Cutting the TTL from an hour to five
minutes made that start 55 minutes sooner, so it is worth fixing now.
The marker is a separate file rather than a touch of the cached
response: a cold /tmp has no response to touch, and that is exactly the
case with no stale data to fall back on, where the timeout is paid in
full with nothing to show for it.
It goes down before the request and is cleared when that lands, not
written afterwards. A fetch is in flight for as long as curl takes to
give up, and this cache dir is shared by every session, so redraws
starting inside that window are real rather than hypothetical: they now
serve stale data instead of each launching their own doomed request.
Every attempt writes it, so a repeated failure restarts the backoff, and
only an attempt does, so the redraws it suppresses cannot keep
re-stamping it and it always lapses.
Missing credentials stays outside the backoff — it costs no timeout, and
the redraw after a login should show the bars rather than wait out a
retry window it had no part in earning.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-09-02 13:12:45 +01:00
|
|
|
|
# Hold off after a failed fetch. With no network curl can burn its
|
|
|
|
|
|
# --max-time before giving up, and a redraw happens on every keystroke,
|
|
|
|
|
|
# so retrying each time would stall the whole status line. The marker is
|
|
|
|
|
|
# separate from the cached response, rather than a touch of it: a cold
|
|
|
|
|
|
# /tmp has no response to touch, which is exactly when there is also no
|
|
|
|
|
|
# stale data to fall back on and the timeout is paid in full.
|
|
|
|
|
|
attempt=true
|
|
|
|
|
|
if [ -f "$fail_marker" ]; then
|
|
|
|
|
|
fail_mtime=$(stat -c %Y "$fail_marker" 2>/dev/null || stat -f %m "$fail_marker" 2>/dev/null)
|
|
|
|
|
|
[ $(( $(date +%s) - fail_mtime )) -lt "$USAGE_RETRY_SECS" ] && attempt=false
|
|
|
|
|
|
fi
|
|
|
|
|
|
|
|
|
|
|
|
if $attempt; then
|
|
|
|
|
|
token=$(get_oauth_token)
|
|
|
|
|
|
if [ -n "$token" ] && [ "$token" != "null" ]; then
|
|
|
|
|
|
# Marked before the call, cleared when it lands. A fetch is in
|
|
|
|
|
|
# flight for as long as curl takes to time out, and this cache
|
|
|
|
|
|
# dir is shared by every session, so redraws that start inside
|
|
|
|
|
|
# that window are real: they now serve stale data instead of
|
|
|
|
|
|
# each launching their own doomed request. Written on every
|
|
|
|
|
|
# attempt, so a repeated failure restarts the backoff, and only
|
|
|
|
|
|
# by an attempt, so the redraws it suppresses cannot keep
|
|
|
|
|
|
# re-stamping it and it always lapses.
|
|
|
|
|
|
#
|
|
|
|
|
|
# No credentials is not a failure worth suppressing: it costs no
|
|
|
|
|
|
# timeout, and the next redraw after a login should show the
|
|
|
|
|
|
# bars rather than wait out a retry window. Hence inside the
|
|
|
|
|
|
# token check.
|
|
|
|
|
|
: > "$fail_marker"
|
|
|
|
|
|
response=$(curl -s --max-time 10 \
|
|
|
|
|
|
-H "Accept: application/json" \
|
|
|
|
|
|
-H "Content-Type: application/json" \
|
|
|
|
|
|
-H "Authorization: Bearer $token" \
|
|
|
|
|
|
-H "anthropic-beta: oauth-2025-04-20" \
|
|
|
|
|
|
-H "User-Agent: claude-code/2.1.34" \
|
|
|
|
|
|
"https://api.anthropic.com/api/oauth/usage" 2>/dev/null)
|
|
|
|
|
|
if [ -n "$response" ] && echo "$response" | jq . >/dev/null 2>&1; then
|
|
|
|
|
|
# Only cache successful (non-error) responses
|
|
|
|
|
|
if ! echo "$response" | jq -e '.error' >/dev/null 2>&1; then
|
|
|
|
|
|
usage_data="$response"
|
|
|
|
|
|
echo "$response" > "$api_cache"
|
|
|
|
|
|
rm -f "$fail_marker"
|
|
|
|
|
|
fi
|
2026-08-08 11:54:05 +01:00
|
|
|
|
fi
|
|
|
|
|
|
fi
|
|
|
|
|
|
fi
|
Back off after a failed usage fetch instead of retrying every redraw
A failed fetch left the cache mtime untouched, so with no network every
redraw — and a redraw happens on every keystroke — tried curl again and
could wait --max-time 10 for it. Cutting the TTL from an hour to five
minutes made that start 55 minutes sooner, so it is worth fixing now.
The marker is a separate file rather than a touch of the cached
response: a cold /tmp has no response to touch, and that is exactly the
case with no stale data to fall back on, where the timeout is paid in
full with nothing to show for it.
It goes down before the request and is cleared when that lands, not
written afterwards. A fetch is in flight for as long as curl takes to
give up, and this cache dir is shared by every session, so redraws
starting inside that window are real rather than hypothetical: they now
serve stale data instead of each launching their own doomed request.
Every attempt writes it, so a repeated failure restarts the backoff, and
only an attempt does, so the redraws it suppresses cannot keep
re-stamping it and it always lapses.
Missing credentials stays outside the backoff — it costs no timeout, and
the redraw after a login should show the bars rather than wait out a
retry window it had no part in earning.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-09-02 13:12:45 +01:00
|
|
|
|
|
2026-08-08 11:54:05 +01:00
|
|
|
|
# Fall back to stale cache if refresh failed — skip if it's an error response
|
|
|
|
|
|
if [ -z "$usage_data" ] && [ -f "$api_cache" ]; then
|
|
|
|
|
|
stale=$(cat "$api_cache" 2>/dev/null)
|
|
|
|
|
|
! echo "$stale" | jq -e '.error' >/dev/null 2>&1 && usage_data="$stale"
|
|
|
|
|
|
fi
|
|
|
|
|
|
fi
|
|
|
|
|
|
|
|
|
|
|
|
if [ -n "$usage_data" ] && echo "$usage_data" | jq -e . >/dev/null 2>&1; then
|
|
|
|
|
|
bar_width=6
|
|
|
|
|
|
[ "$width_tier" = "split" ] && bar_width=4
|
|
|
|
|
|
|
Show the per-model usage limit, and size the status line by measurement
/usage reports a weekly limit scoped to a single model (Fable) that the
status line did not surface. It is absent from both the status line's stdin
JSON and the legacy seven_day_opus/seven_day_sonnet fields, which are always
null now; it lives in the usage API's .limits[] under kind "weekly_scoped".
Render it labelled by scope.model.display_name so it follows whichever model
the limit applies to, inside the 7d segment since both are weekly limits
sharing one reset time.
Fitting it exposed a problem with sizing by width tier. Tiers only know the
terminal width, so content that varies with session state — branch name, cwd,
model display name — could push a line past the edge and be clipped by the
renderer. Measure the assembled line instead (vis_len) and emit the richest of
four rate-limit variants that fits, on one line or two: with reset times, with
extra-usage credits, bars only, or bars without the per-model segment. Every
branch is fit-checked, including with wrapping disabled. Two lines render
correctly in the status line.
Cap the cwd basename as well: nothing else shortened line one, so a long
project directory could overflow it on its own.
Correct the usable width. Claude Code exports COLUMNS but applies the `padding`
setting on top of it rather than deducting it first, so a line sized to COLUMNS
is clipped; deduct padding on both sides plus a column of margin.
Sanitise payload data before rendering. printf %b interprets backslash escapes,
which is how the colour variables work, so a cwd or model name containing a
literal \n — or a real newline, which jq decodes from the JSON — would split
the line and break both the width measurement and the two-line guarantee.
Parse the usage payload in one jq pass rather than ten. The status line runs on
every redraw and per-field parsing had become the dominant cost; this brings a
render back to roughly what it was before (~155ms vs ~115ms parent), the
remainder being the reset times the tiered version did not show at this width.
Fields are read one per line rather than via @tsv: tab is IFS whitespace, so
`read` collapses runs of it and an empty field — no scoped limit, which is the
common case — silently shifted every later field along by one.
Tests cover the API payload shapes including malformed and hostile input, the
width invariants (never exceed the usable width, never more than two lines,
never wrap unnecessarily), and a sweep over widths, branch lengths, model
names, cwd lengths and extra-usage. They restore the live usage cache on exit,
and remove the fixture outright when there was no cache to restore — otherwise
a test run would leave fabricated usage figures live for an hour.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-27 17:55:35 +01:00
|
|
|
|
# One jq pass for the whole payload. Parsing it field by field meant
|
|
|
|
|
|
# ~10 jq processes per render on the same JSON; the status line runs on
|
|
|
|
|
|
# every redraw, so that was the dominant cost. Rounding happens in jq
|
|
|
|
|
|
# too, which removes the per-field awk calls.
|
|
|
|
|
|
#
|
|
|
|
|
|
# The per-model weekly limit (e.g. Fable) lives in .limits[] under kind
|
|
|
|
|
|
# "weekly_scoped" -- the legacy seven_day_opus / seven_day_sonnet fields
|
|
|
|
|
|
# are always null now. Prefer an is_active entry, else take the first.
|
|
|
|
|
|
# `.limits[]?` tolerates limits being absent or not an array, and
|
|
|
|
|
|
# `tonumber? // 0` tolerates a percentage arriving as a string.
|
|
|
|
|
|
# One field per line, not @tsv: tab is IFS whitespace, so `read` collapses
|
|
|
|
|
|
# runs of it and an empty field (no scoped limit -> empty display name)
|
|
|
|
|
|
# would silently shift every later field along by one.
|
|
|
|
|
|
{
|
|
|
|
|
|
IFS= read -r five_hour_pct
|
|
|
|
|
|
IFS= read -r seven_day_pct
|
|
|
|
|
|
IFS= read -r five_hour_reset_iso
|
|
|
|
|
|
IFS= read -r seven_day_reset_iso
|
|
|
|
|
|
IFS= read -r scoped_name
|
|
|
|
|
|
IFS= read -r scoped_pct
|
|
|
|
|
|
IFS= read -r extra_enabled
|
|
|
|
|
|
IFS= read -r extra_pct
|
|
|
|
|
|
IFS= read -r extra_used
|
|
|
|
|
|
IFS= read -r extra_limit
|
|
|
|
|
|
} <<EOF
|
|
|
|
|
|
$(echo "$usage_data" | jq -r '
|
|
|
|
|
|
def num: (tonumber? // 0);
|
|
|
|
|
|
([.limits[]? | select(.kind == "weekly_scoped")]
|
|
|
|
|
|
| (map(select(.is_active)) + .) | first) as $scoped
|
|
|
|
|
|
| [ (.five_hour.utilization | num | round),
|
|
|
|
|
|
(.seven_day.utilization | num | round),
|
|
|
|
|
|
(.five_hour.resets_at // ""),
|
|
|
|
|
|
(.seven_day.resets_at // ""),
|
|
|
|
|
|
(($scoped.scope.model.display_name // "") | gsub("[\\n\\r\\t]"; " ")),
|
|
|
|
|
|
($scoped.percent | num | round),
|
|
|
|
|
|
(.extra_usage.is_enabled // false),
|
|
|
|
|
|
(.extra_usage.utilization | num | round),
|
|
|
|
|
|
((.extra_usage.used_credits | num) / 100 * 100 | round / 100),
|
|
|
|
|
|
((.extra_usage.monthly_limit | num) / 100 * 100 | round / 100)
|
|
|
|
|
|
] | .[] | tostring' 2>/dev/null)
|
|
|
|
|
|
EOF
|
|
|
|
|
|
|
|
|
|
|
|
# Three variants of the rate-limit group, richest first:
|
|
|
|
|
|
# rl_rich bars + reset times + extra usage
|
|
|
|
|
|
# rl_mid bars + extra usage
|
|
|
|
|
|
# rl_lean bars only
|
|
|
|
|
|
# The ladder at the end emits the richest one that fits the space it
|
|
|
|
|
|
# has. Keeping a lean variant matters: extra-usage credits add ~30
|
|
|
|
|
|
# columns, which can overflow line two on its own.
|
|
|
|
|
|
seg_5h="${dim}5h${reset} $(build_bar "${five_hour_pct:-0}" "$bar_width") ${cyan}${five_hour_pct:-0}%${reset}"
|
|
|
|
|
|
seg_7d="${sep}${dim}7d${reset} $(build_bar "${seven_day_pct:-0}" "$bar_width") ${cyan}${seven_day_pct:-0}%${reset}"
|
|
|
|
|
|
|
|
|
|
|
|
# Rendered inside the 7d segment rather than as its own: both are weekly
|
|
|
|
|
|
# limits resetting at the same time, so one shared reset label covers
|
|
|
|
|
|
# the pair and saves a separator plus a second timestamp.
|
|
|
|
|
|
seg_scoped=""
|
|
|
|
|
|
if $SHOW_MODEL_LIMIT && [ -n "$scoped_name" ]; then
|
|
|
|
|
|
seg_scoped=" ${dim}$(esc_data "$(truncate_str "$scoped_name" 8)")${reset} $(build_bar "${scoped_pct:-0}" "$bar_width") ${cyan}${scoped_pct:-0}%${reset}"
|
|
|
|
|
|
fi
|
|
|
|
|
|
|
|
|
|
|
|
# Extra usage, when the account has it enabled.
|
|
|
|
|
|
seg_extra=""
|
|
|
|
|
|
if [ "$extra_enabled" = "true" ]; then
|
|
|
|
|
|
seg_extra="${sep}${dim}extra${reset} $(build_bar "${extra_pct:-0}" "$bar_width") ${cyan}\$$(printf '%.2f' "${extra_used:-0}")${dim}/\$$(printf '%.2f' "${extra_limit:-0}")${reset}"
|
|
|
|
|
|
fi
|
|
|
|
|
|
|
|
|
|
|
|
# rl_bare drops the per-model bar too: the last thing worth giving up,
|
|
|
|
|
|
# and the only way to fit at all when wrapping is disabled and the
|
|
|
|
|
|
# terminal is narrow.
|
|
|
|
|
|
rl_bare="${seg_5h}${seg_7d}"
|
|
|
|
|
|
rl_lean="${rl_bare}${seg_scoped}"
|
|
|
|
|
|
rl_mid="${rl_lean}${seg_extra}"
|
|
|
|
|
|
rl_rich="$rl_mid"
|
|
|
|
|
|
|
|
|
|
|
|
# Formatting the two reset timestamps costs ~10 subprocesses (date has
|
|
|
|
|
|
# no portable one-shot form here), so only pay for it when there is a
|
|
|
|
|
|
# chance they will be shown: appended to line one, or alone on line two.
|
|
|
|
|
|
# RESET_COST is the combined width of " reset 4:40p.m." and
|
|
|
|
|
|
# " reset aug 28, 3:00a.m.".
|
|
|
|
|
|
line1_len=$(vis_len "$out")
|
|
|
|
|
|
mid_len=$(vis_len "$rl_mid")
|
|
|
|
|
|
RESET_COST=30
|
|
|
|
|
|
if [ $(( line1_len + 3 + mid_len + RESET_COST )) -le "$USABLE_WIDTH" ] \
|
|
|
|
|
|
|| { $WRAP_NARROW && [ $(( mid_len + RESET_COST )) -le "$USABLE_WIDTH" ]; }; then
|
2026-08-08 11:54:05 +01:00
|
|
|
|
five_hour_reset=$(format_reset_time "$five_hour_reset_iso" "time")
|
|
|
|
|
|
seven_day_reset=$(format_reset_time "$seven_day_reset_iso" "datetime")
|
Show the per-model usage limit, and size the status line by measurement
/usage reports a weekly limit scoped to a single model (Fable) that the
status line did not surface. It is absent from both the status line's stdin
JSON and the legacy seven_day_opus/seven_day_sonnet fields, which are always
null now; it lives in the usage API's .limits[] under kind "weekly_scoped".
Render it labelled by scope.model.display_name so it follows whichever model
the limit applies to, inside the 7d segment since both are weekly limits
sharing one reset time.
Fitting it exposed a problem with sizing by width tier. Tiers only know the
terminal width, so content that varies with session state — branch name, cwd,
model display name — could push a line past the edge and be clipped by the
renderer. Measure the assembled line instead (vis_len) and emit the richest of
four rate-limit variants that fits, on one line or two: with reset times, with
extra-usage credits, bars only, or bars without the per-model segment. Every
branch is fit-checked, including with wrapping disabled. Two lines render
correctly in the status line.
Cap the cwd basename as well: nothing else shortened line one, so a long
project directory could overflow it on its own.
Correct the usable width. Claude Code exports COLUMNS but applies the `padding`
setting on top of it rather than deducting it first, so a line sized to COLUMNS
is clipped; deduct padding on both sides plus a column of margin.
Sanitise payload data before rendering. printf %b interprets backslash escapes,
which is how the colour variables work, so a cwd or model name containing a
literal \n — or a real newline, which jq decodes from the JSON — would split
the line and break both the width measurement and the two-line guarantee.
Parse the usage payload in one jq pass rather than ten. The status line runs on
every redraw and per-field parsing had become the dominant cost; this brings a
render back to roughly what it was before (~155ms vs ~115ms parent), the
remainder being the reset times the tiered version did not show at this width.
Fields are read one per line rather than via @tsv: tab is IFS whitespace, so
`read` collapses runs of it and an empty field — no scoped limit, which is the
common case — silently shifted every later field along by one.
Tests cover the API payload shapes including malformed and hostile input, the
width invariants (never exceed the usable width, never more than two lines,
never wrap unnecessarily), and a sweep over widths, branch lengths, model
names, cwd lengths and extra-usage. They restore the live usage cache on exit,
and remove the fixture outright when there was no cache to restore — otherwise
a test run would leave fabricated usage figures live for an hour.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-27 17:55:35 +01:00
|
|
|
|
r5=""; [ -n "$five_hour_reset" ] && r5=" ${dim}↺ ${five_hour_reset}${reset}"
|
|
|
|
|
|
r7=""; [ -n "$seven_day_reset" ] && r7=" ${dim}↺ ${seven_day_reset}${reset}"
|
|
|
|
|
|
rl_rich="${seg_5h}${r5}${seg_7d}${seg_scoped}${r7}${seg_extra}"
|
2026-08-08 11:54:05 +01:00
|
|
|
|
fi
|
Show the per-model usage limit, and size the status line by measurement
/usage reports a weekly limit scoped to a single model (Fable) that the
status line did not surface. It is absent from both the status line's stdin
JSON and the legacy seven_day_opus/seven_day_sonnet fields, which are always
null now; it lives in the usage API's .limits[] under kind "weekly_scoped".
Render it labelled by scope.model.display_name so it follows whichever model
the limit applies to, inside the 7d segment since both are weekly limits
sharing one reset time.
Fitting it exposed a problem with sizing by width tier. Tiers only know the
terminal width, so content that varies with session state — branch name, cwd,
model display name — could push a line past the edge and be clipped by the
renderer. Measure the assembled line instead (vis_len) and emit the richest of
four rate-limit variants that fits, on one line or two: with reset times, with
extra-usage credits, bars only, or bars without the per-model segment. Every
branch is fit-checked, including with wrapping disabled. Two lines render
correctly in the status line.
Cap the cwd basename as well: nothing else shortened line one, so a long
project directory could overflow it on its own.
Correct the usable width. Claude Code exports COLUMNS but applies the `padding`
setting on top of it rather than deducting it first, so a line sized to COLUMNS
is clipped; deduct padding on both sides plus a column of margin.
Sanitise payload data before rendering. printf %b interprets backslash escapes,
which is how the colour variables work, so a cwd or model name containing a
literal \n — or a real newline, which jq decodes from the JSON — would split
the line and break both the width measurement and the two-line guarantee.
Parse the usage payload in one jq pass rather than ten. The status line runs on
every redraw and per-field parsing had become the dominant cost; this brings a
render back to roughly what it was before (~155ms vs ~115ms parent), the
remainder being the reset times the tiered version did not show at this width.
Fields are read one per line rather than via @tsv: tab is IFS whitespace, so
`read` collapses runs of it and an empty field — no scoped limit, which is the
common case — silently shifted every later field along by one.
Tests cover the API payload shapes including malformed and hostile input, the
width invariants (never exceed the usable width, never more than two lines,
never wrap unnecessarily), and a sweep over widths, branch lengths, model
names, cwd lengths and extra-usage. They restore the live usage cache on exit,
and remove the fixture outright when there was no cache to restore — otherwise
a test run would leave fabricated usage figures live for an hour.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-27 17:55:35 +01:00
|
|
|
|
fi
|
|
|
|
|
|
fi
|
2026-08-08 11:54:05 +01:00
|
|
|
|
|
Wrap the status line rather than drop the per-model limit
The measured ladder tried every single-line variant before considering a
second line, so whenever line one plus the full group did not fit — from
around 92 usable columns upwards, depending on how long the branch, cwd and
model names are — it emitted rl_bare (5h and 7d only) and silently dropped
the per-model (Fable) bar that is the main reason the group is worth
rendering. On a 110-column laptop the limit was invisible.
Reorder so a second line beats losing that bar: one line rich/mid/lean, then
wrap, and rl_bare only when WRAP_NARROW is false. Reset times and
extra-usage credits are still given up rather than wrapped for, which makes
the rendered content non-monotone in width; the comment on the ladder spells
that out.
Also stop the tests writing fixtures to the caches the live status line
reads. ~/.claude/statusline.sh is a symlink to the script, so a test run was
visibly rendering fabricated usage — a Fable bar at 10%, extra-usage credits
of $1234.56/$2000.00 — in whatever session happened to be open, and a run
killed before its trap fired would have left that in place for an hour. The
cache directory is now $STATUSLINE_CACHE_DIR (default /tmp/claude) and each
suite points it at a temporary directory, which also removes the
backup/restore dance and the deletion of the live git-status cache.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-31 17:28:55 +01:00
|
|
|
|
# Attach the rate-limit group, emitting the richest layout that fits.
|
|
|
|
|
|
#
|
|
|
|
|
|
# A second line is worth it for the per-model bar and nothing else: rl_bare is
|
|
|
|
|
|
# the only variant that drops that bar, and it is the whole reason the group is
|
|
|
|
|
|
# worth showing on an account that has one. Reset times and extra-usage credits
|
|
|
|
|
|
# are given up instead of wrapped for, so content is NOT monotone in width -- at
|
|
|
|
|
|
# one column narrower, rung 3 stops fitting and the wrap that follows has room
|
|
|
|
|
|
# for the timestamps as well. Only wrapping being switched off makes rl_bare the
|
|
|
|
|
|
# answer.
|
|
|
|
|
|
# 1-3. one line: with resets / with extra usage / bars + per-model
|
Show the per-model usage limit, and size the status line by measurement
/usage reports a weekly limit scoped to a single model (Fable) that the
status line did not surface. It is absent from both the status line's stdin
JSON and the legacy seven_day_opus/seven_day_sonnet fields, which are always
null now; it lives in the usage API's .limits[] under kind "weekly_scoped".
Render it labelled by scope.model.display_name so it follows whichever model
the limit applies to, inside the 7d segment since both are weekly limits
sharing one reset time.
Fitting it exposed a problem with sizing by width tier. Tiers only know the
terminal width, so content that varies with session state — branch name, cwd,
model display name — could push a line past the edge and be clipped by the
renderer. Measure the assembled line instead (vis_len) and emit the richest of
four rate-limit variants that fits, on one line or two: with reset times, with
extra-usage credits, bars only, or bars without the per-model segment. Every
branch is fit-checked, including with wrapping disabled. Two lines render
correctly in the status line.
Cap the cwd basename as well: nothing else shortened line one, so a long
project directory could overflow it on its own.
Correct the usable width. Claude Code exports COLUMNS but applies the `padding`
setting on top of it rather than deducting it first, so a line sized to COLUMNS
is clipped; deduct padding on both sides plus a column of margin.
Sanitise payload data before rendering. printf %b interprets backslash escapes,
which is how the colour variables work, so a cwd or model name containing a
literal \n — or a real newline, which jq decodes from the JSON — would split
the line and break both the width measurement and the two-line guarantee.
Parse the usage payload in one jq pass rather than ten. The status line runs on
every redraw and per-field parsing had become the dominant cost; this brings a
render back to roughly what it was before (~155ms vs ~115ms parent), the
remainder being the reset times the tiered version did not show at this width.
Fields are read one per line rather than via @tsv: tab is IFS whitespace, so
`read` collapses runs of it and an empty field — no scoped limit, which is the
common case — silently shifted every later field along by one.
Tests cover the API payload shapes including malformed and hostile input, the
width invariants (never exceed the usable width, never more than two lines,
never wrap unnecessarily), and a sweep over widths, branch lengths, model
names, cwd lengths and extra-usage. They restore the live usage cache on exit,
and remove the fixture outright when there was no cache to restore — otherwise
a test run would leave fabricated usage figures live for an hour.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-27 17:55:35 +01:00
|
|
|
|
# 4-6. two lines: same order, line two having room the single line lacked
|
Wrap the status line rather than drop the per-model limit
The measured ladder tried every single-line variant before considering a
second line, so whenever line one plus the full group did not fit — from
around 92 usable columns upwards, depending on how long the branch, cwd and
model names are — it emitted rl_bare (5h and 7d only) and silently dropped
the per-model (Fable) bar that is the main reason the group is worth
rendering. On a 110-column laptop the limit was invisible.
Reorder so a second line beats losing that bar: one line rich/mid/lean, then
wrap, and rl_bare only when WRAP_NARROW is false. Reset times and
extra-usage credits are still given up rather than wrapped for, which makes
the rendered content non-monotone in width; the comment on the ladder spells
that out.
Also stop the tests writing fixtures to the caches the live status line
reads. ~/.claude/statusline.sh is a symlink to the script, so a test run was
visibly rendering fabricated usage — a Fable bar at 10%, extra-usage credits
of $1234.56/$2000.00 — in whatever session happened to be open, and a run
killed before its trap fired would have left that in place for an hour. The
cache directory is now $STATUSLINE_CACHE_DIR (default /tmp/claude) and each
suite points it at a temporary directory, which also removes the
backup/restore dance and the deletion of the live git-status cache.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-31 17:28:55 +01:00
|
|
|
|
# 7. one line, bars only -- WRAP_NARROW=false, the least-bad single line
|
Show the per-model usage limit, and size the status line by measurement
/usage reports a weekly limit scoped to a single model (Fable) that the
status line did not surface. It is absent from both the status line's stdin
JSON and the legacy seven_day_opus/seven_day_sonnet fields, which are always
null now; it lives in the usage API's .limits[] under kind "weekly_scoped".
Render it labelled by scope.model.display_name so it follows whichever model
the limit applies to, inside the 7d segment since both are weekly limits
sharing one reset time.
Fitting it exposed a problem with sizing by width tier. Tiers only know the
terminal width, so content that varies with session state — branch name, cwd,
model display name — could push a line past the edge and be clipped by the
renderer. Measure the assembled line instead (vis_len) and emit the richest of
four rate-limit variants that fits, on one line or two: with reset times, with
extra-usage credits, bars only, or bars without the per-model segment. Every
branch is fit-checked, including with wrapping disabled. Two lines render
correctly in the status line.
Cap the cwd basename as well: nothing else shortened line one, so a long
project directory could overflow it on its own.
Correct the usable width. Claude Code exports COLUMNS but applies the `padding`
setting on top of it rather than deducting it first, so a line sized to COLUMNS
is clipped; deduct padding on both sides plus a column of margin.
Sanitise payload data before rendering. printf %b interprets backslash escapes,
which is how the colour variables work, so a cwd or model name containing a
literal \n — or a real newline, which jq decodes from the JSON — would split
the line and break both the width measurement and the two-line guarantee.
Parse the usage payload in one jq pass rather than ten. The status line runs on
every redraw and per-field parsing had become the dominant cost; this brings a
render back to roughly what it was before (~155ms vs ~115ms parent), the
remainder being the reset times the tiered version did not show at this width.
Fields are read one per line rather than via @tsv: tab is IFS whitespace, so
`read` collapses runs of it and an empty field — no scoped limit, which is the
common case — silently shifted every later field along by one.
Tests cover the API payload shapes including malformed and hostile input, the
width invariants (never exceed the usable width, never more than two lines,
never wrap unnecessarily), and a sweep over widths, branch lengths, model
names, cwd lengths and extra-usage. They restore the live usage cache on exit,
and remove the fixture outright when there was no cache to restore — otherwise
a test run would leave fabricated usage figures live for an hour.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-27 17:55:35 +01:00
|
|
|
|
# Every branch is fit-checked, so no layout is chosen that would be clipped by
|
|
|
|
|
|
# the renderer, whatever the branch name, cwd, model name or extra-usage width.
|
|
|
|
|
|
if [ -n "$rl_bare" ]; then
|
|
|
|
|
|
rich_len=$(vis_len "$rl_rich")
|
|
|
|
|
|
lean_len=$(vis_len "$rl_lean")
|
|
|
|
|
|
if [ $(( line1_len + 3 + rich_len )) -le "$USABLE_WIDTH" ]; then out+="${sep}${rl_rich}"
|
|
|
|
|
|
elif [ $(( line1_len + 3 + mid_len )) -le "$USABLE_WIDTH" ]; then out+="${sep}${rl_mid}"
|
|
|
|
|
|
elif [ $(( line1_len + 3 + lean_len )) -le "$USABLE_WIDTH" ]; then out+="${sep}${rl_lean}"
|
|
|
|
|
|
elif $WRAP_NARROW; then
|
|
|
|
|
|
if [ "$rich_len" -le "$USABLE_WIDTH" ]; then out+=$'\n'"$rl_rich"
|
|
|
|
|
|
elif [ "$mid_len" -le "$USABLE_WIDTH" ]; then out+=$'\n'"$rl_mid"
|
|
|
|
|
|
elif [ "$lean_len" -le "$USABLE_WIDTH" ]; then out+=$'\n'"$rl_lean"
|
|
|
|
|
|
else out+=$'\n'"$rl_bare"
|
2026-08-08 11:54:05 +01:00
|
|
|
|
fi
|
Show the per-model usage limit, and size the status line by measurement
/usage reports a weekly limit scoped to a single model (Fable) that the
status line did not surface. It is absent from both the status line's stdin
JSON and the legacy seven_day_opus/seven_day_sonnet fields, which are always
null now; it lives in the usage API's .limits[] under kind "weekly_scoped".
Render it labelled by scope.model.display_name so it follows whichever model
the limit applies to, inside the 7d segment since both are weekly limits
sharing one reset time.
Fitting it exposed a problem with sizing by width tier. Tiers only know the
terminal width, so content that varies with session state — branch name, cwd,
model display name — could push a line past the edge and be clipped by the
renderer. Measure the assembled line instead (vis_len) and emit the richest of
four rate-limit variants that fits, on one line or two: with reset times, with
extra-usage credits, bars only, or bars without the per-model segment. Every
branch is fit-checked, including with wrapping disabled. Two lines render
correctly in the status line.
Cap the cwd basename as well: nothing else shortened line one, so a long
project directory could overflow it on its own.
Correct the usable width. Claude Code exports COLUMNS but applies the `padding`
setting on top of it rather than deducting it first, so a line sized to COLUMNS
is clipped; deduct padding on both sides plus a column of margin.
Sanitise payload data before rendering. printf %b interprets backslash escapes,
which is how the colour variables work, so a cwd or model name containing a
literal \n — or a real newline, which jq decodes from the JSON — would split
the line and break both the width measurement and the two-line guarantee.
Parse the usage payload in one jq pass rather than ten. The status line runs on
every redraw and per-field parsing had become the dominant cost; this brings a
render back to roughly what it was before (~155ms vs ~115ms parent), the
remainder being the reset times the tiered version did not show at this width.
Fields are read one per line rather than via @tsv: tab is IFS whitespace, so
`read` collapses runs of it and an empty field — no scoped limit, which is the
common case — silently shifted every later field along by one.
Tests cover the API payload shapes including malformed and hostile input, the
width invariants (never exceed the usable width, never more than two lines,
never wrap unnecessarily), and a sweep over widths, branch lengths, model
names, cwd lengths and extra-usage. They restore the live usage cache on exit,
and remove the fixture outright when there was no cache to restore — otherwise
a test run would leave fabricated usage figures live for an hour.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-27 17:55:35 +01:00
|
|
|
|
else
|
|
|
|
|
|
out+="${sep}${rl_bare}"
|
2026-08-08 11:54:05 +01:00
|
|
|
|
fi
|
|
|
|
|
|
fi
|
|
|
|
|
|
|
|
|
|
|
|
printf "%b" "$out"
|
|
|
|
|
|
exit 0
|