patch-settings.conf

# Settings owned by the patches in patches/, read at patch time.
# Kimi's own settings live in ~/.kimi-code/config.toml, the launcher's in
# env-profile.conf. Defaults are registered in lib/patch_settings.py.

# Height of the slash-command popup: default | half | full
suggestion_height=full

# Add a /wd command for switching the working directory: on | off
wd_command = on

# Click into the composer to place the cursor (fullscreen only): on | off
click_cursor = on

# Extra filenames Kimi accepts as project instructions: off | claude | all
# Kimi's own AGENTS.md keeps priority; at most one plain file per directory is
# ever read, so this can never double the prompt.
agents_md_names=claude

# Prefix each line the Read tool returns with its number: on | off
# Off saves tokens on every file read, and costs the model the ability to cite
# a line number from a Read alone.
read_line_numbers = on

# Reasoning effort per turn, derived from the prompt: off | pin | free
effort_router = pin

# The working indicator: default | mirror | custom, or one of the presets
# kimi-code-mods | braille | dots | moon | blocks | wave | glow | arc | star
spinner_style = kimi-code-mods
# Milliseconds per spinner frame: default, or 20..2000
spinner_interval_ms = default
# The frames `custom` uses. Separated by spaces where there are any, taken one
# character at a time where there are not — so "a b c" and "abc" are the same
# three frames, and a frame made of several characters needs the spaced form.
spinner_frames = default
# Rotate the word beside the spinner: on | off
thinking_verbs = off
# The words to rotate through: default, or your own, comma separated
thinking_verbs_list = default
# Where the word goes; {} is the word
thinking_verbs_format = {}

# Your own messages in the transcript
user_message_marker = none
user_message_border = topbottom
user_message_style = plain

# The frame around the composer: default | off | single | double | bold
input_box_border = topbottom
spinner_mirror=on
expanded_by_default=on
auto_accept_plan=on
fullscreen=on

# Fire JSON files dropped into <sessionDir>/cron/ as native one-shot cron
# turns (external bus daemon): on | off
cron_drop_dir = on

# A standing list of subagents under the composer, instead of a tree that
# scrolls away with the transcript: off | running | all
# `all` also keeps the ones that just finished, briefly — which means it
# overrides Kimi's own habit of dropping a foreground subagent's record the
# moment the turn ends.
agent_dock = all

# Subagenten von der Turn-Ausfuehrung loesen: default | always | immediate
# `always` setzt run_in_background auf jedem Agent-Aufruf, auch wenn das
# Modell nicht daran gedacht hat — die Eingabezeile bleibt damit bedienbar,
# waehrend Agenten arbeiten. `immediate` beendet den Turn zusaetzlich sofort
# nach dem Dispatch (stopTurn), ohne dem Modell eine Fortsetzung zu erlauben.
agent_background = always
read_limits=1000
user_message_border_color=#ec4242
copy_on_mark=on
input_box_prompt=none
input_box_text_color=#ffffff
at_file_suggestions=full
thinking_display=full
context_position=top
steer_mid_turn=on
agent_dock_rows=5
welcome_banner=on
working_style=custom
working_mirror=on

# Toolcall-Header: Bash-Kommando statt "Ran a command" (on), Used/Using
# aus allen Tool-Headern entfernen (off): on | off
bash_one_liner=on
tool_call_used=off
status_hints=off
input_box_style=plain
working_frames=🟦 🟩 🟨 🟧 🟥