rtk
shorter shell output before the agent reads it, so fewer input tokens.
- status
- optional ships with colonizer, off until you switch it on.
- version
- v0.49.0
- pinned
- tag v0.49.0 · archive sha256 74b226ab
- licence
- apache-2.0
- source
- rtk-ai/rtk ↗
- docs
- token savings
what it does for you
rtk shortens shell output before the agent reads it: commands such as git status or a test run go through rtk rewrite, and commands rtk has no rule for run unchanged.
what it needs
- the claude code settler
how to turn it on
settings → modules on the claude code settler: the rtk switch. off by default.
how far it has been tested
it has run in a real microvm, and its rewrites through the real agent sdk. it has not run in a colony yet.
how colonizer runs it
- pinned as source in vendor.lock (tag v0.49.0) and built by scripts/build-rtk.sh into a static musl binary, because upstream's aarch64 release needs glibc 2.39 and the colony image has 2.36
- the runner registers an in-process PreToolUse hook on Bash: exit 0 or 3 with output replaces the command, anything else (no equivalent, a deny rule, rtk missing, or no answer within 2 seconds) runs it unchanged
- read, grep and glob do not go through the shell and are never rewritten; rtk's telemetry is opt-in and never switched on in a colony
what it costs
not measured as a separate figure: the binary ships with the install at <COLONIZER_HOME>/bin/rtk and mounts at /opt/colonizer/bin/rtk.