rtk-assisted-shell

RTK Assisted Shell

Save tokens by routing noisy shell workflows through RTK-style filtering first.

Goal

Before running a command that could dump large output, prefer an RTK-shaped command or an equivalent compact shell pattern so the model receives a smaller artifact.

Default Rewrites

Prefer these forms first when RTK is installed:

• git status -> rtk git status
• git diff -> rtk git diff
• git log -> rtk git log
• cat file, head file, tail file -> rtk read file
• rg pattern, grep pattern -> rtk grep pattern .
• ls, large find flows -> rtk ls or compact find
• pytest, npm test, cargo test, go test, tsc, linters -> rtk <command>
• docker logs, kubectl logs -> rtk docker logs ..., rtk kubectl logs ...

If RTK is not installed or does not support the command, fall back to compact native shell patterns:

• file-scoped diffs, not whole history;
• bounded excerpts, not full files;
• failing tests first, not full suites;
• error-first or summary-first log views.

Workflow

1. Decide whether the command is likely to be noisy.
2. If yes, prefer RTK rewrite first.
3. If RTK output is still insufficient, broaden one step:
   • summary -> excerpt -> raw command.
4. If RTK is unavailable, preserve the same compactness policy manually.
5. Keep exact raw output only when it materially changes diagnosis or handoff.

When To Use Aggressively

• repo exploration;
• git state and diff inspection;
• test and lint failures;
• CI logs;
• container and cluster logs;
• broad searches over large repos.

Guardrails

• Do not use RTK if output format must remain exact for downstream parsing.
• Do not hide the only decisive stack trace or assertion.
• If RTK filtering obscures the needed detail, fall back to a narrower raw command before using a broad raw command.

Final Report

State whether RTK or manual compact output shaping was used, what decisive signal was kept, and what large output was intentionally avoided.