docs(workbook): add evergreen RTK deep cuts to token savings field guide - #10
docs(workbook): add evergreen RTK deep cuts to token savings field guide#10mikeangstadt wants to merge 1 commit into
Conversation
Deepen the Token Savings Field Guide with three principle-first additions inspired by field write-ups on RTK token savings: - Move 1: explain RTK is content-aware (compresses noisy output, leaves source code untouched) so the reasoning context stays intact. - Move 1: add the evergreen predictor for compression payoff (the more predictable a command's output, the more compressible noise it carries) instead of citing version-specific benchmark numbers. - New Move 9 "Keep the front of the conversation stable": teach prefix caching as a durable principle (pin durable context at the top, let churn happen at the bottom); renumber checklist to Move 10 and update the intro count. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
mikeangstadt
left a comment
mikeangstadt
left a comment
There was a problem hiding this comment.
Reviewed as a docs/workbook change. Overall this is clean, well-scoped, and technically sound: the numbering update (nine to ten) is consistent across the intro paragraph and the renumbered checklist, the new Move 9 reinforces (does not contradict) the existing guided-compaction guidance in the Compaction checkpoint, both links resolve (200), and no em-dashes were introduced. The evergreen framing is a good call. Two small precision notes below, neither blocking.
| <p>Token savings are not about making Claude think less. They are about keeping the context window focused on the material that changes the outcome. Read the ten moves below as a rough progression, not a checklist of rules: the early ones are reflexes you build in your first week, the later ones are habits you grow into as your sessions get more ambitious. You will not do all ten on every task, but the more advanced your work gets, the more of them you will reach for without thinking.</p> | ||
| <h2>1. Enable RTK for noisy command output</h2> | ||
| <p><strong><a href="https://github.com/rtk-ai/rtk" target="_blank" rel="noopener noreferrer">RTK (Rust Token Killer)</a></strong> is a CLI proxy that compresses noisy command output before it reaches Claude's context, so a run that would have streamed hundreds of lines into the session lands as a compact, structured summary instead. It is useful when shell output is large, repetitive, or noisy. The impact compounds because command output does not just cost tokens once; it remains in the session and can be re-read on later turns.</p> | ||
| <p><strong>The deep cut:</strong> RTK is content-aware, and that distinction is the whole point. It compresses the low-signal output (test runs, build logs, status noise) and leaves your <strong>source code untouched</strong>, because compressing the code Claude is reasoning over would degrade the reasoning itself. This is the same principle as the rest of this guide: cut the noise, never the signal. It is why you can leave RTK on by default instead of toggling it per command.</p> |
There was a problem hiding this comment.
Minor precision nit (non-blocking): "leaves your source code untouched" is a slightly stronger claim than RTK's actual scope as described in this same move ("a CLI proxy that compresses noisy command output"). RTK operates on the output stream of shell commands; it does not sit on top of Claude's file-read/Read-tool results generally. The point you're making (it compresses low-signal output, not the reasoning material) is correct and worth keeping, but consider phrasing it as "leaves high-signal command output such as file contents and search results untouched" so a reader doesn't infer RTK inspects/guards source files directly. Purely a clarity refinement.
| /compact focusing on decisions made, evidence pointers, files touched, | ||
| unresolved questions, and residual risk.</code></pre> | ||
| <h2>9. End-of-task token checklist</h2> | ||
| <h2>9. Keep the front of the conversation stable</h2> |
There was a problem hiding this comment.
Optional consistency check: this new Move 9 ("Keep the front of the conversation stable" / prefix caching) and the existing Compaction checkpoint both advocate compacting only at phase boundaries — good, they agree. No change needed; flagging only to confirm the reinforcement is intentional and reads well in sequence.
1 participant
What
Deepens the Token Savings Field Guide in
workbook/index.htmlwith three principle-first additions, inspired by field write-ups on RTK token savings (e.g. token-savings-rtk-headroom). Each is framed around the durable, evergreen principle rather than version-specific install steps or benchmark numbers that age.Changes
/compactat a phase boundary beats constant mid-task edits. Checklist renumbered to Move 10; intro "nine moves" updated to "ten".Out of scope (deliberately)
Install/PATH/jq/Headroom setup (owned by the pre-work checklist), the
permissionDecisionwrapper internals, and competing proxies. Kept to evergreen operating habits.Notes
No em-dashes; existing markup and coaching voice preserved.
🤖 Generated with Claude Code