Restoring a checkpoint and an earlier point in the conversation #239
Replies: 4 comments
-
|
Agree with this - "double escape" has become a useful (and necessary) convention for working with Agents - would be keen to support. |
Beta Was this translation helpful? Give feedback.
-
|
We would also love to see this feature! (branch out conversations, edit previous messages, etc.) |
Beta Was this translation helpful? Give feedback.
-
|
Seeing how Zed has some approve/revert (at least on Editor changes), i tried prototyping some Review/Revert behaviors for Eclipse. I seem to be able to intercept fs/write_text_file calls and either capture a reference to the SCM/IDE History snapshot of affected editor/file before the first change or create a new backup if no one exists... Looking at the terminal commands though for Move (mv) or Delete (rm) for Gemini, it doesn't look like the File locations values are populated fo rthe tool calls though. Maybe this can be fixed in gemini, maybe i can parse the title, or maybe if i enabled the terminal capabilities of the client i might be able to test each argument if it matches an existing file path... Some of the challenges are:
|
Beta Was this translation helpful? Give feedback.
-
|
Can an Agent's checkpointing commands be exposed as Slash Commands Also of note for Gemini, checkpointing cant be enabled using a startup flag: The Checkpointing feature is disabled by default. To enable it, you need to edit your settings.json file. Note: The --checkpointing command-line flag was removed in version 0.11.0. Checkpointing can now only be enabled through the settings.json configuration file. |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
-
It seems like there are 2 missing capabilities in the protocol right now but are available in the underlying CLIs:
“Restore to before this change/tool call”
“Restore to an earlier point in the conversation”
Right now, different agents expose this in ad-hoc ways (e.g., gemini has checkpointing) but it seems like the current protocol doesn't support this.
Beta Was this translation helpful? Give feedback.
All reactions