v1.5.169

Bug Fixes

• close --all no longer closes other people's sessions by accident. While another session is live, it closes nothing, lists the other sessions, and says to run chrome-use close for your own or add --force for all. In an 8-agent test, two agents ran close --all when stuck and wiped the other seven sessions mid-task. (#429)
• A session closed from outside says so. Its next command used to run silently in a new blank tab. Agents saw about:blank, assumed the form was lost, and submitted twice. Now the next command warns that the tabs were closed by close --all (or session stop/prune) from a named session, and says to check whether a submission already went through before redoing it. (#431)
• tab select --activate won't hide another session's tab. While another live session's tab is in front of the window, it is refused; add --force to override. The background-tab note now explains that a click there usually did reach the page: wait for the result and re-read, and don't resubmit. (#431)
• pick handles autocomplete fields. pick @e6 --option "Kyoto" types the text, waits for the suggestions, clicks the best match ("Kyoto" over "Kyoto Station") and checks the field took it. pick <ref> "<text>" works too, and select on a non-native combobox points to pick. Almost every agent in the test fell back to typing and clicking by hand. (#430)

Improvements

• The core skill again lists the commands to use instead of eval, in 5 rows: text, click, counting, waiting, and setting values. (#428)

Contributors

• @leeguooooo