From e7320b748a00f0ff2657117873815d665436bf1f Mon Sep 17 00:00:00 2001 From: ak2k <19240940+ak2k@users.noreply.github.com> Date: Wed, 20 May 2026 19:09:38 -0400 Subject: [PATCH] =?UTF-8?q?docs:=20README=20=E2=80=94=20drop=20stale=20ver?= =?UTF-8?q?sion=20line,=20replace=20estimate=20with=20measured=20A/B?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit 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. --- README.md | 19 +++++++++++++++---- 1 file changed, 15 insertions(+), 4 deletions(-) diff --git a/README.md b/README.md index b6e0a29..36230a8 100644 --- a/README.md +++ b/README.md @@ -2,7 +2,7 @@ Lightweight-delegation variant of [compound-engineering-plugin](https://github.com/EveryInc/compound-engineering-plugin). -Drops the persistent agent registrations (~71k tokens of idle context savings, conditional on Claude Code loading agent bodies into the registration slot — see [the docstring](converter/generate_wrappers.py) for the measurement caveat) and dispatches every specialist via a single resolver shim instead. Specialist prompts move to `references/agent-prompts/`; the resolver (`bin/ce-lite-persona`) loads them on demand. Same coverage as upstream CE, near-zero idle cost, and a `UserPromptSubmit` hook restores proactive specialist invocation across all 49 personas. +Drops the persistent agent registrations (**measured: ~99.9k tokens of idle context, gone**) and dispatches every specialist via a single resolver shim instead. Specialist prompts move to `references/agent-prompts/`; the resolver (`bin/ce-lite-persona`) loads them on demand. Same coverage as upstream CE, near-zero idle cost, and a `UserPromptSubmit` hook restores proactive specialist invocation across all 49 personas. ## Install @@ -11,13 +11,24 @@ Drops the persistent agent registrations (~71k tokens of idle context savings, c /plugins install ce-lite@ce-lite ``` -During trial, slash commands are namespaced under `/ce-lite:*` so the original `compound-engineering` plugin can be installed alongside for direct comparison. The current shipping version is `compound-engineering-v3.8.3-lite.3`; daily `upstream-watch.yml` auto-bumps as EveryInc releases. +During trial, slash commands are namespaced under `/ce-lite:*` so the original `compound-engineering` plugin can be installed alongside for direct comparison. Daily `upstream-watch.yml` auto-bumps the plugin as EveryInc tags new releases. ## Trade-offs vs. upstream CE -| | Upstream CE | ce-lite | +Measured against an interactive Claude Code session (Opus 4.7 1M-context, two isolated `CLAUDE_CONFIG_DIR`s, identical state except the installed plugin): + +| Context slot | Upstream CE | ce-lite | Δ | +|---|---:|---:|---:| +| **Total** | **192.5k** | **87.3k** | **−105.2k** | +| Custom agents | 99.9k (49 subagent registrations) | 0 (no registrations) | −99.9k | +| System tools | 35.5k | 22.3k | −13.2k | +| Skills | 4.6k | 12.5k | +7.9k | +| System prompt / MCP / memory / messages | 52.5k | 52.5k | 0 | + +The ~99.9k of custom-agents context — Claude Code loads each registered subagent's full body for proactive selection — is the dominant savings. ce-lite trades a +7.9k skills bump (88 SKILL.md descriptions vs upstream's 29) for that. + +| Behavior | Upstream CE | ce-lite | |---|---|---| -| Idle context cost | ~77k tokens of agent bodies (49 registrations, if Claude Code loads bodies into the registration slot) | ~5,250 tokens (88 SKILL.md `description:` fields only); persona bodies lazy-loaded on dispatch | | Specialist coverage | 49 personas | 49 personas (prompts moved to `references/agent-prompts/`, dispatched via `bin/ce-lite-persona`) | | Proactive specialist invocation | ✅ Any conversation can pull `security-sentinel` via registered subagent | ✅ `UserPromptSubmit` hook keyword-matches every prompt against all 49 personas and injects suggestions (~70-84% activation per cited research, vs ~20% baseline for description-only routing) | | Tool-restriction enforcement | Enforced by harness on registered subagent | Advisory only — dispatch via `general-purpose` subagent. See [caveat below](#caveat-tool-restrictions-are-advisory). |