docs: README — measured A/B for the savings claim + drop stale version line#27
Merged
Conversation
…d A/B Two changes: 1. Drop the hardcoded `compound-engineering-v3.8.3-lite.3` shipping version. It drifts on every converter-only bump and the marketplace already surfaces the current version. 2. Replace the "~71k tokens, conditional on Claude Code loading agent bodies into the registration slot" hedge with a measured A/B table. Interactive `/context` comparison (Opus 4.7 1M, two isolated `CLAUDE_CONFIG_DIR`s, identical except the installed plugin): Custom agents: 99.9k → 0 (−99.9k) System tools: 35.5k → 22.3k (−13.2k) Skills: 4.6k → 12.5k (+7.9k) Total: 192.5k → 87.3k (−105.2k) The hedge is gone — Claude Code does load agent bodies into the registration slot in interactive mode. The savings claim was actually conservative.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Two README changes:
Drop the hardcoded version mention on the install line — it
pointed at
compound-engineering-v3.8.3-lite.3and drifts on everyconverter-only bump. The marketplace surfaces the current version
already.
Replace the "~71k tokens, conditional on Claude Code loading
agent bodies into the registration slot" hedge with measured numbers.
Measurement
Two isolated
CLAUDE_CONFIG_DIRs, identical state except the installedplugin. Compared
/contextslot-by-slot in interactive Claude Code(Opus 4.7 1M).
Claude Code does load each registered subagent's full body into context
for proactive selection. Dropping the 49 registrations frees ~99.9k.
The +7.9k skills bump is the 88 SKILL.md descriptions vs upstream's 29.
The old estimate was actually conservative.
Followup
The +7.9k skills bump is inflated by a double-registration: each skill
appears as both
<name>andce-lite:<name>in the/contextskillslist. Generated by
converter/generate_commands.py(one commandwrapper per skill, each with its own
description:field). Will fixin a separate PR — once de-duped, savings should grow another ~6k.
Test plan
nix flake checkgreen on the worktree