Re-verify all 13 chapters against current upstream (2.1.x)

[anneheartrecord]

The analysis was performed against the sourcemap snapshot dated 2026-03-31, whose build self-reported 2.1.888. Upstream has moved since. This issue tracks a chapter-by-chapter re-verification so the version coverage table in the README can be updated with an honest "last re-verified" date.

Until a chapter is checked off here, treat its details as accurate to the March snapshot rather than to whatever version you are running.

Chapters

• 01 Architecture Overview — tech stack versions, file line counts, core file paths
• 02 Value Debate — mostly analysis, low drift risk
• 03 Agent Loop — six-phase loop, AsyncGenerator design
• 04 Context Engineering — system prompt composition, CLAUDE.md five-layer loading
• 05 Compaction System — three tiers, the context-window − 13,000 token threshold
• 06 Permission System — three modes, the 42 hardcoded rules, classifier behavior
• 07 Memory System — five-layer loading, @include, MEMORY.md
• 08 Tools and Skills — the 40+ tool registry, pipeline, hooks
• 09 MCP Integration — transports, OAuth, config scopes
• 10 Future Features — which of the 82 flags have since shipped, and what they became
• 11 AI Code Review — analysis, low drift risk
• 12 Agent Security Design — analysis built on chapter 06, recheck after 06
• 13 Source Code Findings — analysis, low drift risk

Priority

Chapter 10 first: it predicted unreleased features from feature flags, so it is the chapter most likely to be either vindicated or wrong by now, and the most interesting either way. Then 06 and 05, which state exact numbers.

How to help

Pick one chapter, check its concrete claims against what you can observe, and open a version drift issue per finding — or comment here to claim a chapter so two people do not do the same one.