diff --git a/README.md b/README.md
index e78fb60..4e2e5aa 100644
--- a/README.md
+++ b/README.md
@@ -4,10 +4,11 @@
 > The core concept — separating deterministic operations from agent reasoning —
 > is sound, but my approach has since evolved. I now use
 > [Pi.dev](https://pi.dev) with a more conversational prompting style, favouring
-> **extensions** (in-process deterministic plugins that are ~10× faster than
+> **extensions** (in-process deterministic plugins that are much faster than
 > shell scripts and easier to debug) over compiled Go binaries generated by this
-> skill. The lines between stored prompts, plugins, and ad-hoc instructions have
-> blurred; this repo documents an earlier point in that journey.
+> skill. At the moment I work in a more improvisational, conversational way with stored prompts, plugins, and ad-hoc instructions
+> ; this repo documents an earlier point in that journey.
+> All text in this repository, apart from this line, have been generated conversationally with Claude Code (November 2025, the week before Opus 4.5) and [Pi.dev](https://pi.dev) with Deepseek v4 flash. The idea of a meta-skill some people found interesting in conversations, and I had a need to respond to a comment on LinkedIn where this would be a good explanation of something I tried, and liked. I used this to make a set of skills that let long running agents implement most of QWAN Tracker while I kept an eye on the user experience. Enjoy, Willem van den Ende.
 ---
 
 # Meta Skill Generator - Deliverables