Repository navigation
Show the reply on models that hide text before a tool call - #24
Merged
Merged
Conversation
…t before a tool call Claude Sonnet 5.5, Opus 5.5 and Fable 5 return a note longer than a sentence or two written before a tool call as a progress update inside a thinking block, so a reply written before the chips was never shown and the chips_alone hold kept the model looping. The text_before_tool_calls capability (false for those models) gives the chips tool a leading reply field, adds a short prompt section, shows the reply as text while it streams (eager_input_streaming via ToolSpec::$streamsInput) and keeps it in the message text and the memory. Other models see no change. Also: prices and capabilities keep model names as written (dashes were turned into underscores).
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Claude Sonnet 5.5, Opus 5.5 and Fable 5 return a note longer than a sentence or two, written before a tool call, inside a
thinkingblock instead of as text. A turn without a card ends withpresent_suggestions, so its reply was never shown, and thechips_alonehold made the model loop (4–6 rounds, ending without chips).Changes
text_before_tool_calls. It is false for those models in the Anthropic adapter; with it off:present_suggestionsgets a leadingreplyfield;reply.ToolSpec::$streamsInput. The adapter sendseager_input_streamingfor that tool, so the reply arrives as it is written.