@miphamai/cli 0.73.1 → 0.74.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/package.json +1 -1
- package/src/core/eval-harness.ts +16 -0
- package/src/shared/package-info.ts +1 -1
- package/src/ui/input.tsx +88 -66
package/package.json
CHANGED
package/src/core/eval-harness.ts
CHANGED
|
@@ -70,6 +70,7 @@ export const ANCHOR_CONTRACT_IDS: ReadonlySet<string> = new Set([
|
|
|
70
70
|
'red-team-zero-gaps',
|
|
71
71
|
'producer-rule-shape',
|
|
72
72
|
'producer-rule-idempotent',
|
|
73
|
+
'self-report-diagnostic',
|
|
73
74
|
])
|
|
74
75
|
|
|
75
76
|
/** 细粒度防回退:返回 role==='anchor' 且已 FAIL 的契约 id。空 = 无 anchor 回退。 */
|
|
@@ -333,6 +334,21 @@ export function runEval(): EvalReport {
|
|
|
333
334
|
results.push({ ...judgeBehaviorTask(task, ruleEngine), role: 'target' })
|
|
334
335
|
}
|
|
335
336
|
|
|
337
|
+
// ── 自报分数只作诊断:评分路径无 LLM,分数来自 ground-truth 契约而非模型自报 ──
|
|
338
|
+
// anchor 锁死「评分组件不暴露 LLM 的 chat 能力」。4 个组件(ruleEngine/constitution/
|
|
339
|
+
// errorDB/preflight)都是确定性组件(runEval 同步评分)。若未来有人把 LLM 注入评分
|
|
340
|
+
// 路径(给组件挂 chat / Llm 接口),此契约立即 FAIL,anchor gate 拒绝固化——即「自报
|
|
341
|
+
// 分数不可信」的机器可判定信号。
|
|
342
|
+
const scoringComponents = [ruleEngine, constitution, errorDB, preflight]
|
|
343
|
+
const llmInjected = scoringComponents.some(
|
|
344
|
+
(c) => typeof (c as { chat?: unknown }).chat === 'function',
|
|
345
|
+
)
|
|
346
|
+
results.push({
|
|
347
|
+
id: 'self-report-diagnostic',
|
|
348
|
+
description: '评分无 LLM:机制哨兵组件不暴露 chat 能力(分数只来自 ground-truth,非模型自报)',
|
|
349
|
+
passed: !llmInjected,
|
|
350
|
+
})
|
|
351
|
+
|
|
336
352
|
// 角色标注:anchor 走集中清单(门保护面单一真源),target 已在上方循环内联。
|
|
337
353
|
for (const r of results) {
|
|
338
354
|
if (ANCHOR_CONTRACT_IDS.has(r.id)) r.role = 'anchor'
|
|
@@ -9,7 +9,7 @@
|
|
|
9
9
|
export const PACKAGE_NAME = '@miphamai/cli' as const
|
|
10
10
|
|
|
11
11
|
/** 当前发布版本 */
|
|
12
|
-
export const PACKAGE_VERSION = '0.
|
|
12
|
+
export const PACKAGE_VERSION = '0.74.0' as const
|
|
13
13
|
|
|
14
14
|
/** npm install 全局安装命令 */
|
|
15
15
|
export const NPM_INSTALL_COMMAND = `npm install -g ${PACKAGE_NAME}` as const
|
package/src/ui/input.tsx
CHANGED
|
@@ -1,6 +1,5 @@
|
|
|
1
1
|
import React, { useState, useEffect, useRef, useMemo } from 'react'
|
|
2
2
|
import { Box, Text, useInput } from 'ink'
|
|
3
|
-
import TextInput from 'ink-text-input'
|
|
4
3
|
import { getCommandList } from './commands.js'
|
|
5
4
|
import { CommandPicker } from './command-picker.js'
|
|
6
5
|
import { useI18n } from '../i18n-context'
|
|
@@ -80,11 +79,11 @@ function pick<T>(arr: T[]): T {
|
|
|
80
79
|
}
|
|
81
80
|
|
|
82
81
|
/**
|
|
83
|
-
*
|
|
84
|
-
*
|
|
82
|
+
* 归一化输入:多行粘贴折叠成单行(CR/LF/Tab → 空格)。
|
|
83
|
+
* ink-text-input 是单行组件,嵌入的 \r 会拉回行首覆盖显示、\n 会 trap 光标。
|
|
85
84
|
*/
|
|
86
|
-
export function
|
|
87
|
-
return
|
|
85
|
+
export function normalizeInput(input: string): string {
|
|
86
|
+
return input.replace(/[\r\n\t]+/g, ' ')
|
|
88
87
|
}
|
|
89
88
|
|
|
90
89
|
/** True when typing a leading `/` should auto-open the slash-command picker. */
|
|
@@ -92,6 +91,81 @@ export function shouldAutoOpenPicker(value: string, prevValue: string, enabled:
|
|
|
92
91
|
return enabled && value.startsWith('/') && !prevValue.startsWith('/')
|
|
93
92
|
}
|
|
94
93
|
|
|
94
|
+
/**
|
|
95
|
+
* Mipham 自有的单行文本输入,替代 ink-text-input。
|
|
96
|
+
*
|
|
97
|
+
* 为什么不用 ink-text-input:它把粘贴按「光标偏移切片插入」逐块处理,而
|
|
98
|
+
* cursorOffset 与受控 value 都来自渲染闭包——Ink 会把长粘贴按 stdin read()
|
|
99
|
+
* 边界拆成多块,同一轮 synchronous flush 里后续块读到的仍是旧值,于是「覆盖
|
|
100
|
+
* 前块 / 插到中段」,表现为粘贴内容乱序、丢失、冻住。
|
|
101
|
+
*
|
|
102
|
+
* 这里用 ref 做同步真值:每块按当前 ref 原子追加(光标恒在末尾),不依赖
|
|
103
|
+
* React 渲染时序,分块粘贴自然累积成完整文本。
|
|
104
|
+
*/
|
|
105
|
+
function MiphamTextInput({
|
|
106
|
+
value,
|
|
107
|
+
onChange,
|
|
108
|
+
onSubmit,
|
|
109
|
+
placeholder = '',
|
|
110
|
+
focus = true,
|
|
111
|
+
}: {
|
|
112
|
+
value: string
|
|
113
|
+
onChange: (next: string) => void
|
|
114
|
+
onSubmit: (value: string) => void
|
|
115
|
+
placeholder?: string
|
|
116
|
+
focus?: boolean
|
|
117
|
+
}) {
|
|
118
|
+
// 同步真值:valueRef 永远是最新文本;受控 value 仅在渲染时落后于 ref。
|
|
119
|
+
const valueRef = useRef(value)
|
|
120
|
+
|
|
121
|
+
// 外部改动(箭头历史回填、提交清空)时,把真值对齐回受控 prop。
|
|
122
|
+
useEffect(() => {
|
|
123
|
+
valueRef.current = value
|
|
124
|
+
}, [value])
|
|
125
|
+
|
|
126
|
+
useInput(
|
|
127
|
+
(input, key) => {
|
|
128
|
+
if (!focus) return
|
|
129
|
+
// 上行/下行/制表/Ctrl 由 InputBar 自己的 useInput 处理(历史导航、热键)。
|
|
130
|
+
if (key.upArrow || key.downArrow || key.tab || (key.shift && key.tab) || key.ctrl) return
|
|
131
|
+
|
|
132
|
+
if (key.return) {
|
|
133
|
+
onSubmit(valueRef.current)
|
|
134
|
+
return
|
|
135
|
+
}
|
|
136
|
+
if (key.backspace || key.delete) {
|
|
137
|
+
if (valueRef.current.length > 0) {
|
|
138
|
+
const next = valueRef.current.slice(0, -1)
|
|
139
|
+
valueRef.current = next
|
|
140
|
+
onChange(next)
|
|
141
|
+
}
|
|
142
|
+
return
|
|
143
|
+
}
|
|
144
|
+
|
|
145
|
+
// 打字 / 粘贴:归一化控制符后原子追加(光标恒在末尾)。
|
|
146
|
+
const cleaned = normalizeInput(input)
|
|
147
|
+
if (!cleaned) return
|
|
148
|
+
const next = valueRef.current + cleaned
|
|
149
|
+
valueRef.current = next
|
|
150
|
+
onChange(next)
|
|
151
|
+
},
|
|
152
|
+
{ isActive: focus },
|
|
153
|
+
)
|
|
154
|
+
|
|
155
|
+
return (
|
|
156
|
+
<Text>
|
|
157
|
+
{value.length === 0 && placeholder ? (
|
|
158
|
+
<Text dimColor>{placeholder}</Text>
|
|
159
|
+
) : (
|
|
160
|
+
<>
|
|
161
|
+
{value}
|
|
162
|
+
<Text inverse> </Text>
|
|
163
|
+
</>
|
|
164
|
+
)}
|
|
165
|
+
</Text>
|
|
166
|
+
)
|
|
167
|
+
}
|
|
168
|
+
|
|
95
169
|
export function InputBar({
|
|
96
170
|
onSubmit,
|
|
97
171
|
isLoading,
|
|
@@ -110,10 +184,7 @@ export function InputBar({
|
|
|
110
184
|
const { t } = useI18n()
|
|
111
185
|
const [value, setValue] = useState('')
|
|
112
186
|
// Ref mirror of value — used by useInput to read latest without stale closure.
|
|
113
|
-
// Also used by throttled onChange to hold the latest pending value.
|
|
114
187
|
const valueRef = useRef(value)
|
|
115
|
-
// Throttle timer for onChange — prevents React render floods during paste.
|
|
116
|
-
const onChangeTimerRef = useRef<ReturnType<typeof setTimeout> | null>(null)
|
|
117
188
|
const [verb, setVerb] = useState(() => t(pick(LOADING_KEYS)))
|
|
118
189
|
const [completionVerb, setCompletionVerb] = useState<string | null>(null)
|
|
119
190
|
const prevLoading = useRef(isLoading)
|
|
@@ -190,15 +261,6 @@ export function InputBar({
|
|
|
190
261
|
prevLoading.current = isLoading
|
|
191
262
|
}, [isLoading])
|
|
192
263
|
|
|
193
|
-
// Track pre-shortcut value so we can revert ink-text-input's Ctrl-key insertions.
|
|
194
|
-
// ink-text-input inserts 'p'/'f'/'o' for Ctrl+P/F/O (Ink normalizes control chars
|
|
195
|
-
// to their letter names before passing to useInput).
|
|
196
|
-
const valueBeforeShortcut = useRef(value)
|
|
197
|
-
// Keep the ref in sync with state on every change that isn't a Ctrl shortcut revert.
|
|
198
|
-
useEffect(() => {
|
|
199
|
-
valueBeforeShortcut.current = value
|
|
200
|
-
}, [value])
|
|
201
|
-
|
|
202
264
|
useInput((input, key) => {
|
|
203
265
|
// ── Escape: cancel loading → clear draft ──
|
|
204
266
|
if (key.escape) {
|
|
@@ -220,9 +282,9 @@ export function InputBar({
|
|
|
220
282
|
onCyclePermission?.()
|
|
221
283
|
return
|
|
222
284
|
}
|
|
223
|
-
// Tab → 接受 ghost-text
|
|
285
|
+
// Tab → 接受 ghost-text 建议
|
|
224
286
|
if (key.tab && !key.shift && suggestion) {
|
|
225
|
-
const next =
|
|
287
|
+
const next = valueRef.current + suggestion
|
|
226
288
|
setValue(next)
|
|
227
289
|
valueRef.current = next
|
|
228
290
|
setSuggestion(null)
|
|
@@ -230,27 +292,22 @@ export function InputBar({
|
|
|
230
292
|
}
|
|
231
293
|
// Ctrl+P → toggle model picker
|
|
232
294
|
// NOTE: Ink passes input=keypress.name (just 'p') when ctrl is true, not raw \x10.
|
|
233
|
-
// ink-text-input inserts 'p' as literal text — revert it.
|
|
234
295
|
if (key.ctrl && input === 'p') {
|
|
235
|
-
setValue(valueBeforeShortcut.current)
|
|
236
296
|
onTogglePicker?.()
|
|
237
297
|
return
|
|
238
298
|
}
|
|
239
299
|
// Ctrl+F → toggle focus mode
|
|
240
300
|
if (key.ctrl && input === 'f') {
|
|
241
|
-
setValue(valueBeforeShortcut.current)
|
|
242
301
|
onToggleFocus?.()
|
|
243
302
|
return
|
|
244
303
|
}
|
|
245
304
|
// Ctrl+O → expand/collapse last tool call
|
|
246
305
|
if (key.ctrl && input === 'o') {
|
|
247
|
-
setValue(valueBeforeShortcut.current)
|
|
248
306
|
onToggleExpand?.()
|
|
249
307
|
return
|
|
250
308
|
}
|
|
251
309
|
// Ctrl+G → toggle agent view dashboard
|
|
252
310
|
if (key.ctrl && input === 'g') {
|
|
253
|
-
setValue(valueBeforeShortcut.current)
|
|
254
311
|
onToggleAgentView?.()
|
|
255
312
|
return
|
|
256
313
|
}
|
|
@@ -267,13 +324,7 @@ export function InputBar({
|
|
|
267
324
|
if (submittedHistory.length === 0) return
|
|
268
325
|
// Save current draft the first time we enter history browsing
|
|
269
326
|
if (historyIndexRef.current === -1) {
|
|
270
|
-
//
|
|
271
|
-
// history navigation (paste → immediate ↑ race). Save the latest value
|
|
272
|
-
// from the ref — state `value` is stale inside the throttle window.
|
|
273
|
-
if (onChangeTimerRef.current) {
|
|
274
|
-
clearTimeout(onChangeTimerRef.current)
|
|
275
|
-
onChangeTimerRef.current = null
|
|
276
|
-
}
|
|
327
|
+
// Save the latest value from the ref — state `value` may lag a render.
|
|
277
328
|
savedDraftRef.current = valueRef.current
|
|
278
329
|
}
|
|
279
330
|
const newIndex = Math.min(historyIndexRef.current + 1, submittedHistory.length - 1)
|
|
@@ -319,13 +370,7 @@ export function InputBar({
|
|
|
319
370
|
}, [value])
|
|
320
371
|
|
|
321
372
|
const handleSubmit = (val: string) => {
|
|
322
|
-
//
|
|
323
|
-
if (onChangeTimerRef.current) {
|
|
324
|
-
clearTimeout(onChangeTimerRef.current)
|
|
325
|
-
onChangeTimerRef.current = null
|
|
326
|
-
}
|
|
327
|
-
// Use the latest value from the ref — state lags behind during throttle, and
|
|
328
|
-
// the onSubmit `val` is the stale controlled prop in that window.
|
|
373
|
+
// Use the latest value from the ref; `val` is the value passed by MiphamTextInput.
|
|
329
374
|
const finalValue = valueRef.current || val
|
|
330
375
|
if (!finalValue.trim()) return
|
|
331
376
|
// Submitting while a response streams interrupts it (Claude Code parity)
|
|
@@ -371,7 +416,7 @@ export function InputBar({
|
|
|
371
416
|
<Box marginRight={1}>
|
|
372
417
|
<Text color={isLoading ? 'yellow' : 'cyan'}>{'>'}</Text>
|
|
373
418
|
</Box>
|
|
374
|
-
<
|
|
419
|
+
<MiphamTextInput
|
|
375
420
|
value={value}
|
|
376
421
|
onChange={(val) => {
|
|
377
422
|
// Reset history browsing when user starts typing
|
|
@@ -379,9 +424,6 @@ export function InputBar({
|
|
|
379
424
|
historyIndexRef.current = -1
|
|
380
425
|
savedDraftRef.current = ''
|
|
381
426
|
}
|
|
382
|
-
// Normalize newlines → spaces. ink-text-input is single-line; multi-line
|
|
383
|
-
// paste would trap arrow-key navigation on the first line.
|
|
384
|
-
const normalized = val.replace(/\n/g, ' ')
|
|
385
427
|
// ── Ghost-text 自动补全:每次输入清 suggestion + 重排防抖 ──
|
|
386
428
|
setSuggestion(null)
|
|
387
429
|
const suggestionReqId = ++suggestionReqIdRef.current
|
|
@@ -389,16 +431,12 @@ export function InputBar({
|
|
|
389
431
|
clearTimeout(suggestionTimerRef.current)
|
|
390
432
|
suggestionTimerRef.current = null
|
|
391
433
|
}
|
|
392
|
-
if (
|
|
393
|
-
llm &&
|
|
394
|
-
autocompleteEnabled &&
|
|
395
|
-
shouldAutocomplete(normalized, isLoading, pickerActive)
|
|
396
|
-
) {
|
|
434
|
+
if (llm && autocompleteEnabled && shouldAutocomplete(val, isLoading, pickerActive)) {
|
|
397
435
|
suggestionTimerRef.current = setTimeout(() => {
|
|
398
436
|
requestSuggestion(
|
|
399
437
|
llm,
|
|
400
438
|
recentMessages ?? [],
|
|
401
|
-
|
|
439
|
+
val,
|
|
402
440
|
() => suggestionReqId !== suggestionReqIdRef.current,
|
|
403
441
|
)
|
|
404
442
|
.then((completion) => {
|
|
@@ -409,24 +447,8 @@ export function InputBar({
|
|
|
409
447
|
})
|
|
410
448
|
}, autocompleteDebounceMs)
|
|
411
449
|
}
|
|
412
|
-
|
|
413
|
-
|
|
414
|
-
const bulk = isBulkInput(valueRef.current, normalized)
|
|
415
|
-
valueRef.current = normalized
|
|
416
|
-
if (!bulk) {
|
|
417
|
-
if (onChangeTimerRef.current) {
|
|
418
|
-
clearTimeout(onChangeTimerRef.current)
|
|
419
|
-
onChangeTimerRef.current = null
|
|
420
|
-
}
|
|
421
|
-
setValue(normalized)
|
|
422
|
-
return
|
|
423
|
-
}
|
|
424
|
-
if (onChangeTimerRef.current) return // timer pending, latest value in ref
|
|
425
|
-
setValue(normalized)
|
|
426
|
-
onChangeTimerRef.current = setTimeout(() => {
|
|
427
|
-
onChangeTimerRef.current = null
|
|
428
|
-
setValue(valueRef.current)
|
|
429
|
-
}, 33) // ~30fps
|
|
450
|
+
valueRef.current = val
|
|
451
|
+
setValue(val)
|
|
430
452
|
}}
|
|
431
453
|
onSubmit={handleSubmit}
|
|
432
454
|
placeholder={
|