Tell the M1 agent how to browse: m1-browse, not the Chrome extension

The Claude browser extension cannot work on M1 — it links to a single
persistent interactive session, and the relay runs one-shot processes. It
drives a real visible Chrome over CDP instead, so manual intervention over
screen sharing still works and logins persist.
This commit is contained in:
2026-09-03 19:44:19 +02:00
parent d26b1f7b29
commit 72916d3c79
+22
View File
@@ -19,3 +19,25 @@ and do not run builds or tests here.
rsync, never an overwrite. Then `git log` / `git status` show what changed. rsync, never an overwrite. Then `git log` / `git status` show what changed.
- M1 is the git remote `m1`. If something truly must be committed from here, - M1 is the git remote `m1`. If something truly must be committed from here,
push it with `git push m1 main` — but prefer telling M1 to do it. push it with `git push m1 main` — but prefer telling M1 to do it.
### Browsing and testing on M1
M1 has **no working Claude browser extension** — that extension links to one
persistent interactive session and cannot attach to the relay's one-shot runs.
Use `m1-browse` instead. It drives a **real, visible Chrome** on M1 over the
DevTools protocol, so Vincent can screen-share in and click things himself when
something needs doing by hand; that profile keeps its own logins.
```
m1-browse open <url> # navigate, prints title + final URL
m1-browse text [chars] # visible text of the current page
m1-browse html [chars] # page HTML
m1-browse screenshot [path] # PNG (default /tmp/m1-shot.png); add "full" for full page
m1-browse click <selector>
m1-browse type <selector> <text>
m1-browse eval <js> # run JS in the page, prints the result
m1-browse tabs # list open tabs
```
It starts Chrome itself if it is not running. To show Vincent a screenshot, take
one and say where it is — the M5 side fetches and displays it.