Merge branch 'main' into bb/gui

This commit is contained in:
emozilla
2026-05-20 16:01:41 -04:00
72 changed files with 2726 additions and 742 deletions
@@ -379,11 +379,11 @@ describe('createGatewayEventHandler', () => {
const handler = createGatewayEventHandler(ctx)
handler({
payload: { message: 'Chrome launched and listening on port 9222' },
payload: { message: 'Chromium-family browser launched and listening on port 9222' },
type: 'browser.progress'
} as any)
expect(ctx.system.sys).toHaveBeenCalledWith('Chrome launched and listening on port 9222')
expect(ctx.system.sys).toHaveBeenCalledWith('Chromium-family browser launched and listening on port 9222')
})
it('annotates gateway.start_timeout with stderr tail lines so users can diagnose without /logs', () => {
@@ -387,8 +387,8 @@ describe('createSlashHandler', () => {
Promise.resolve({
connected: false,
messages: [
"Chrome isn't running with remote debugging — attempting to launch...",
'Browser not connected — start Chrome with remote debugging and retry /browser connect'
"Chromium-family browser isn't running with remote debugging — attempting to launch...",
'Browser not connected — start a Chromium-family browser with remote debugging and retry /browser connect'
],
url: 'http://127.0.0.1:9222'
})
@@ -397,14 +397,14 @@ describe('createSlashHandler', () => {
const ctx = buildCtx({ gateway: { ...buildGateway(), rpc } })
expect(createSlashHandler(ctx)('/browser connect')).toBe(true)
expect(ctx.transcript.sys).toHaveBeenCalledWith('checking Chrome remote debugging at http://127.0.0.1:9222...')
expect(ctx.transcript.sys).toHaveBeenCalledWith('checking Chromium-family browser remote debugging at http://127.0.0.1:9222...')
await vi.waitFor(() => {
expect(ctx.transcript.sys).toHaveBeenCalledWith(
"Chrome isn't running with remote debugging — attempting to launch..."
"Chromium-family browser isn't running with remote debugging — attempting to launch..."
)
expect(ctx.transcript.sys).toHaveBeenCalledWith(
'Browser not connected — start Chrome with remote debugging and retry /browser connect'
'Browser not connected — start a Chromium-family browser with remote debugging and retry /browser connect'
)
expect(ctx.transcript.sys).not.toHaveBeenCalledWith('browser connect failed')
})
+18 -3
View File
@@ -21,11 +21,26 @@ describe('splitReasoning', () => {
expect(text).toBe('body')
})
it('treats unclosed trailing <think>… as reasoning', () => {
const { reasoning, text } = splitReasoning('answer start <think>still deciding')
it('treats unclosed leading <think>… as reasoning (real reasoning-model stream)', () => {
const { reasoning, text } = splitReasoning('<think>still deciding')
expect(reasoning).toBe('still deciding')
expect(text).toBe('answer start')
expect(text).toBe('')
})
it('does not strip trailing prose after a stray mid-text <think> mention', () => {
// Regression for "TUI eats last paragraph of output": when the model
// emits a literal `<think>` somewhere in prose (quoted explanation, code
// example, partial stream-mid-tag), the trailing greedy unclosed-tag
// regex used to consume every paragraph after it. Real unclosed
// reasoning blocks always lead the message — anchor to ^ so prose
// mentions are preserved.
const { reasoning, text } = splitReasoning(
'final answer paragraph one.\n\n<think>internal note never closed\n\nfinal answer paragraph two.'
)
expect(reasoning).toBe('')
expect(text).toBe('final answer paragraph one.\n\n<think>internal note never closed\n\nfinal answer paragraph two.')
})
it('returns empty reasoning and untouched text when no tags present', () => {
+2 -2
View File
@@ -155,7 +155,7 @@ export const opsCommands: SlashCommand[] = [
const url = action === 'connect' ? rest.join(' ').trim() || 'http://127.0.0.1:9222' : undefined
if (url) {
ctx.transcript.sys(`checking Chrome remote debugging at ${url}...`)
ctx.transcript.sys(`checking Chromium-family browser remote debugging at ${url}...`)
}
ctx.gateway
@@ -181,7 +181,7 @@ export const opsCommands: SlashCommand[] = [
}
if (r.connected) {
ctx.transcript.sys('Browser connected to live Chrome via CDP')
ctx.transcript.sys('Browser connected to live Chromium-family browser via CDP')
ctx.transcript.sys(`Endpoint: ${r.url || '(url unavailable)'}`)
ctx.transcript.sys('next browser tool call will use this CDP endpoint')
}
+6 -1
View File
@@ -21,7 +21,12 @@ export function splitReasoning(input: string): SplitReasoning {
return ''
})
const unclosed = new RegExp(`<${tag}>([\\s\\S]*)$`, 'i')
// Anchor to start-of-input so a literal `<think>` mid-prose (model quoting
// the word, code blocks containing the tag, etc.) doesn't eat every
// paragraph after it. Real unclosed reasoning blocks always lead the
// message — that's how reasoning models stream. See test
// "does not strip trailing prose after a stray mid-text <think> mention".
const unclosed = new RegExp(`^\\s*<${tag}>([\\s\\S]*)$`, 'i')
text = text.replace(unclosed, (_m, inner: string) => {
const trimmed = inner.trim()