Merged
Conversation
…ppearance - `stopSpinner(false)` now always outputs a blank line before the `›` response prefix, creating clear visual separation between tool calls and model responses. Also fixes a bug where `›` was missing after extended thinking blocks. - `printToolCall` removes the leading `\n` so tool calls render directly on the cleared spinner line, eliminating wasted blank lines before the first tool. - `closeAssistantBox` reduced from `\n\n` to `\n` to avoid excess trailing space now that indentation adds an implicit blank line at response end. - Streaming text in the REPL gets 2-space indentation (`\n` → `\n `) so multi- line responses are visually distinct from tool calls and the input prompt. - Remove the pre-spinner `\n` from `processInput` since the blank separator is now provided by `stopSpinner(false)` at the start of each response. https://claude.ai/code/session_01TpMwMFqNRqfoB7FfkA8j7Y
Model responses now start with bold bright-cyan 'ra' followed by a dim '›', making it immediately clear who is speaking. The user's readline input prompt remains the plain bright-cyan '›', so the two are visually distinct at a glance. https://claude.ai/code/session_01TpMwMFqNRqfoB7FfkA8j7Y
User input keeps the bright-cyan › from readline. Model response now flows without any prefix — clean and unindented, differentiated by the blank line separator and the presence/absence of the › prompt. https://claude.ai/code/session_01TpMwMFqNRqfoB7FfkA8j7Y
Model text is indented 2 spaces (initial indent from stopSpinner, newlines replaced with newline+2spaces), keeping responses visually distinct from the user's flush-left › prompt without any prefix character. https://claude.ai/code/session_01TpMwMFqNRqfoB7FfkA8j7Y
stopSpinner was writing \n before the 2-space indent, but readline already advances a line on Enter — resulting in a blank line between user and model. Drop the \n so the response starts on the very next line. https://claude.ai/code/session_01TpMwMFqNRqfoB7FfkA8j7Y
- closeAssistantBox now writes \n\n instead of \n - fallback path in repl also writes \n\n - stream handler buffers trailing newlines and discards them at response end, so the model's trailing whitespace doesn't add extra blank lines before the separator https://claude.ai/code/session_01TpMwMFqNRqfoB7FfkA8j7Y
LLM text is now wrapped at terminalWidth-2 columns so wrapped lines stay aligned with the 2-space indent. A lineCol cursor tracks position across streaming chunks so wrapping is accurate even when chunks split mid-word. lineCol resets to 0 around tool calls. https://claude.ai/code/session_01TpMwMFqNRqfoB7FfkA8j7Y
The previous approach wrote the space immediately on seeing it, then wrapped before the next word — leaving a trailing space at the end of the line. Replace _flushWord with _emitWord(withLeadingSpace): the space is deferred as pendingSpace until we know the next word's length. If the word wraps, the newline+indent replaces the space; if it fits, the space is written normally. This correctly handles words split across streaming chunks too. https://claude.ai/code/session_01TpMwMFqNRqfoB7FfkA8j7Y
Previously, words longer than the available line width (URLs, file paths, code snippets) were written as-is, causing the terminal to wrap them at column 0 — breaking out of the indented bounding box. Now _emitWord has three paths: 1. Word fits on current line → write normally 2. Word fits on a fresh line → soft-wrap before it 3. Word too long for any line → hard-break char by char at column limit Every continuation line starts with the 2-space indent so text never escapes the bounding box. https://claude.ai/code/session_01TpMwMFqNRqfoB7FfkA8j7Y
The hand-rolled LineWrapper had two classes of bugs: 1. Post-tool-call indent: text resuming after a tool call wrote at col 0 (wrapper.col=0 but no prefix was emitted before the content) 2. Column tracking: char-level counting was fragile — off for wide chars, ANSI codes, and edge cases across streaming chunks New approach: - StreamBuffer buffers text until '\n', then wraps each complete logical line with wrap-ansi (handles ANSI codes, wide chars, hard-breaks long words like URLs). Zero manual column tracking. - boxOpened is reset to false after each tool call so text resuming after tools goes through the normal stopSpinner() path, which writes the prefix and creates a fresh StreamBuffer. No special-case needed. - startSpinner() is called unconditionally in afterToolExecution so the model computing phase always shows a spinner. Visual improvement: response lines are prefixed with " │ " (dim left border) making the response area visually distinct from tool call output. https://claude.ai/code/session_01TpMwMFqNRqfoB7FfkA8j7Y
The │ box-drawing character only fills the centre of the character cell, creating a floating-bar artefact that looks wrong next to normal text. Plain whitespace indent is cleaner and avoids any font rendering issues. https://claude.ai/code/session_01TpMwMFqNRqfoB7FfkA8j7Y
TypeScript's control-flow analysis does not track assignments to `let` variables made inside async closures. It sees `streamBuf` initialized to `null` and keeps that narrow type at the post-loop read sites, making `if (streamBuf)` narrow to `never`. Cast `(streamBuf as tui.StreamBuffer | null)?.end()` to explicitly widen back to the declared union type and bypass the spurious narrowing. https://claude.ai/code/session_01TpMwMFqNRqfoB7FfkA8j7Y
printToolCall now accepts the raw JSON arguments string, collapses it to a single line (stripping outer braces for readability), and truncates with '…' when it would exceed the terminal width. https://claude.ai/code/session_01TpMwMFqNRqfoB7FfkA8j7Y
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
2 participants
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.
No description provided.