goose powertools
#10356
Replies: 2 comments
|
While this is not a core feature request, I would love to see an integrated terminal in the desktop app. Something that opens on the side or at the bottom, similar to how other desktop agent apps do it. |
0 replies
|
Please don't abandon the desktop app. Its currently one of the only "agent" apps available that can compete with claude cowork / chatgpt work, etc. |
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Powertools
The primary future of Goose is as a developer kit: the GDK. We will continue supporting the current desktop app as a reference client for the GDK. We expect the primary users of that reference client to be developers, so we want to add features that turn the client into a workbench full of power tools for agentic development.
Below are ideas that have come up in various discussions.
Show conversation
You can already do something similar by going to all sessions and exporting the conversation, but this would be inline and quickly allow you to check what is in the session object, including the full conversation (agent & user visible) and any of the data stored in the session for token usage or under extension specific data like todo.
pr: #9678
Show model interactions
Downloading the diagnostics bundle and running the diagnostics viewer gives you something similar already today, so this means building this interaction into the goose client. For each recent interaction, show the request as a json document and the response as a series of json documents, one for each streaming packet, plus whatever we want to expose at the model level. Adding an edit & replay function here could be interesting.
pr: #9806
Tool calls & responses
Similar to the previous, but now focussed on the interaction with a particular tool. Probably best to directly attach to the tool call anchor in the conversation. Opens a viewer showing the request as a json document and the response as a series of json documents, one for each streaming package received. Edit and replay for testing is useful here too.
Context viewer
Viewer showing the elements that go into the current context and where they come from. A combination of the actual system prompt, the various AGENT.md’s we’ve picked up, recipe/agent additions, tool/extension descriptions and executed skills (though technically skills are currently not rendered in the system prompt). In the process we should also clean up what we pick up and remove some random things.
Message Stats
Show token message for each request/reply pair, including time to first token, total time, cost and caching plus the derived tokens per second.
Hooks
if hooks were called, what they were and what the result was
Import/Export
Add the ability to import/export sessions and agent traces from all the major harnesses and standards.
Context management
Expand the current viewer for the context manager to show what the context is spent on instead of just how far we are in the current window. Add settings to control the tool request/reply compaction. Pick the model for compaction instead of auto picking fast model per provider.
All reactions