Consider it done. The open-source AI agent that works out of the box · 想到,就能做到。开源、开箱即用的 AI Agent。
TypeScript agentai-agentai-assistantandroidclaude-codecodexelectroniosllmmacosreact-nativetypescriptwindows
import { Button } from '@/components/ui/button';
/**
* LearnReviewPanel —— /learn 蒸馏提案的 diff 审查面板。
*
* 数据流:open 时 learn:get-proposal-diff 拉 FileChange[](全新 skill = 全 added;
* 覆盖已装 skill = 与现有版本的差异)。底部动作:「应用到本地技能」(learn:apply,
* 用户确认后才落盘 —— 提案永远不自动写)、「放弃」(learn:discard,丢弃 staging)。
*
* 复用:DiffPanelShell(滑入容器)+ FileChangeGroup(单文件折叠 diff,与
* SkillhubDiffPanel 共用)。
*/
import { useEffect, useMemo, useState } from 'react';
import { useTranslation } from 'react-i18next';
import { FileDiff as FileDiffIcon, ShieldAlert, Replace } from 'lucide-react';
import { toast } from '@/lib/toast';
import { DiffPanelShell } from '@/components/diff-panel/DiffPanelShell';
import { FileChangeGroup, type FileChange } from '@/components/diff-panel/FileChangeGroup';
import { computeDiffStats } from '@/lib/agent-actions/diffStats';
import { mapIpcErrorToI18nKey } from '@/utils/ipcError';
import type { LearnRunPublic } from '../../../shared/learnTypes';
import { learnApiFor } from './learnTransport';
interface LearnReviewPanelProps {
open: boolean;
onClose: () => void;
run: LearnRunPublic;
/** 所在会话视图的 sessionId —— 决定 diff/apply/discard 路由到哪台设备的 learn-host
* (device-link 远程会话经隧道到被控端);本机会话可缺省。 */
contextSessionId?: string | null;
}
type LoadState =
| { kind: 'idle' }
| { kind: 'loading' }
| { kind: 'error'; message: string }
| { kind: 'ready'; targetExists: boolean; changes: FileChange[] };
export function LearnReviewPanel({ open, onClose, run, contextSessionId }: LearnReviewPanelProps) {
const { t } = useTranslation();
const [state, setState] = useState<LoadState>({ kind: 'idle' });
const [acting, setActing] = useState<'apply' | 'discard' | null>(null);
// 打开时拉 diff;关闭清空(重开重新拉,保证最新)
useEffect(() => {
if (!open) {
setState({ kind: 'idle' });
return;
}
let cancelled = false;
setState({ kind: 'loading' });
learnApiFor(contextSessionId)
.getProposalDiff({ runId: run.runId })
.then((res) => {
if (cancelled) return;
setState({ kind: 'ready', targetExists: res.targetExists, changes: res.changes });
})
.catch((err: unknown) => {
if (cancelled) return;
setState({ kind: 'error', message: err instanceof Error ? err.message : String(err) });
});
return () => {
cancelled = true;
};
// run.proposalFingerprint:面板开着时用户在蒸馏会话继续对话、提案被 watcher
// 刷新(learn:event 推动 run 更新)→ 重新拉 diff,不展示过期提案(Codex review)。
// 不能依赖 run.updatedAt:getProposalDiff 本身会登记 reviewed 指纹并 bump
// updatedAt + 广播,依赖它会"读→更新→再读"无限循环(Codex review #548 P1);
// 指纹只在提案内容真变时才变。contextSessionId:device-link 路由上下文变化时重拉。
}, [open, run.runId, run.proposalFingerprint, contextSessionId]);
const totals = useMemo(() => {
if (state.kind !== 'ready') return null;
let add = 0;
let del = 0;
for (const c of state.changes) {
if (c.isBinary) continue;
const s = computeDiffStats(c.oldContent, c.newContent);
add += s.add;
del += s.del;
}
return { files: state.changes.length, add, del };
}, [state]);
const handleApply = async (): Promise<void> => {
setActing('apply');
try {
const result = await learnApiFor(contextSessionId).apply({ runId: run.runId });
toast.success(t('learn.review.appliedToast', { name: result.name }));
onClose();
} catch (err) {
toast.error(t(mapIpcErrorToI18nKey(err, { namespace: 'learn.ipcError', fallback: 'learn.review.applyFailed' })));
} finally {
setActing(null);
}
};
const handleDiscard = async (): Promise<void> => {
setActing('discard');
try {
await learnApiFor(contextSessionId).discard({ runId: run.runId });
toast.success(t('learn.review.discardedToast'));
onClose();
} catch (err) {
toast.error(t(mapIpcErrorToI18nKey(err, { namespace: 'learn.ipcError', fallback: 'learn.review.discardFailed' })));
} finally {
setActing(null);
}
};
return (
<DiffPanelShell
open={open}
onClose={onClose}
variant="floating"
ariaLabel={t('learn.review.ariaLabel')}
title={t('learn.review.title', { name: run.skillName ?? '' })}
defaultWidth={560}
storageKey="diff-panel-shell:learn-width"
rightHeader={
totals && totals.files > 0 ? (
<span className="text-xs text-muted-foreground tabular-nums">
{t('skillhub.diffPanel.filesCount', { count: totals.files })}
<span className="text-[var(--diff-add-fg)]">+{totals.add}</span>{' '}
<span className="text-[var(--diff-del-fg)]">-{totals.del}</span>
</span>
) : null
}
>
<div className="flex h-full flex-col">
<div className="min-h-0 flex-1 overflow-y-auto">
{/* 提示条:敏感内容 warning(redaction 第二层防线)+ 覆盖已装版本提示。
蒸馏说明与迭代都在会话本身 —— 面板只做 diff 核对 + 落盘闸门。 */}
{run.redactionWarnings && run.redactionWarnings.length > 0 && (
<div className="mx-3 mt-3 flex items-start gap-2 rounded-md bg-[var(--warning-bg-soft)] px-3 py-2">
<ShieldAlert size={14} className="mt-0.5 shrink-0 text-muted-foreground" />
<p className="text-xs text-muted-foreground">
{t('learn.review.redactionWarning', {
categories: run.redactionWarnings.join(', '),
})}
</p>
</div>
)}
{state.kind === 'ready' && state.targetExists && (
<div className="mx-3 mt-3 flex items-start gap-2 rounded-md bg-[var(--warning-bg-soft)] px-3 py-2">
<Replace size={14} className="mt-0.5 shrink-0 text-muted-foreground" />
<p className="text-xs text-muted-foreground">{t('learn.review.overwriteWarning')}</p>
</div>
)}
{state.kind === 'loading' && (
<div className="flex h-full items-center justify-center">
<p className="text-sm text-muted-foreground">{t('skillhub.diffPanel.loading')}</p>
</div>
)}
{state.kind === 'error' && (
<div className="flex h-full flex-col items-center justify-center gap-3 px-6 text-center">
<FileDiffIcon size={28} className="text-muted-foreground" />
<p className="text-sm text-muted-foreground">{t('learn.review.errorTitle')}</p>
<p className="max-w-xs break-all text-xs text-muted-foreground">{state.message}</p>
</div>
)}
{state.kind === 'ready' && (
<ul className="flex flex-col gap-2 p-3">
{state.changes.map((c) => (
// 审查场景内容即主体:SKILL.md 一律展开;文件少(≤3)时全展开,
// 避免"全新 skill 单文件折叠 = 打开一片空白"的反体验。
<FileChangeGroup
key={c.path}
change={c}
defaultExpanded={c.path === 'SKILL.md' || state.changes.length <= 3}
/>
))}
</ul>
)}
</div>
{/* 底部动作条:apply 是唯一落盘入口(提案永不自动写)。想改?直接在
蒸馏会话里说 —— 对话即迭代,面板不承载对话。 */}
<div className="flex shrink-0 items-center justify-between gap-2 border-t border-border px-3 py-2.5">
<span className="min-w-0 truncate text-xs text-muted-foreground">
{t('learn.review.reviseHint')}
</span>
<div className="flex shrink-0 items-center gap-2">
<Button
variant="secondary"
size="md"
compact
tone="quiet"
loading={acting === 'discard'}
type="button"
disabled={acting !== null}
onClick={() => void handleDiscard()}
>
{acting === 'discard' ? t('learn.review.discarding') : t('learn.review.discard')}
</Button>
<Button
variant="cta"
size="md"
compact
loading={acting === 'apply'}
type="button"
disabled={acting !== null || state.kind !== 'ready'}
onClick={() => void handleApply()}
>
{acting === 'apply' ? t('learn.review.applying') : t('learn.review.apply')}
</Button>
</div>
</div>
</div>
</DiffPanelShell>
);
}