diff --git a/src/engine/piece-runner.ts b/src/engine/piece-runner.ts index 3eb4b16..6afd777 100644 --- a/src/engine/piece-runner.ts +++ b/src/engine/piece-runner.ts @@ -607,7 +607,11 @@ export async function runPiece( const cfg = loadConfig(); flushAndStageRecording( options.taskId, - options.userId, + // No-auth mode passes userId=null; fall back to the same 'local' + // namespace as the ToolContext below, otherwise + // flushAndStageRecording's `if (!ownerId) return` silently discards + // every browser recording and self-healing patch in no-auth. + options.userId ?? 'local', cfg.userFolderRoot ?? './data/users', ); } diff --git a/src/scheduler.test.ts b/src/scheduler.test.ts index 38d27e6..21092c1 100644 --- a/src/scheduler.test.ts +++ b/src/scheduler.test.ts @@ -262,6 +262,81 @@ describe('Scheduler.executeScheduledTask: ownership inheritance', () => { }); }); +describe('Scheduler.tick: skip-on-in-progress', () => { + // Regression: a scheduled task's previous job ending in `waiting_human` (ASK) + // used to be treated as "in progress" and block all future scheduled runs + // forever — waiting_human has no auto-recovery, and an unattended schedule has + // nobody to answer the ASK, so the schedule silently died. Now waiting_human is + // excluded from the skip set: each scheduled run is an independent task, so we + // leave the stale waiting_human task as-is and start a fresh run. + let tempDir: string; + let repo: Repository; + let scheduler: Scheduler; + + beforeEach(() => { + tempDir = mkdtempSync(join(tmpdir(), 'agent-sched-skip-')); + repo = new Repository(join(tempDir, 'db.sqlite')); + scheduler = new Scheduler(repo, join(tempDir, 'workspaces')); + }); + + afterEach(() => { + repo.close(); + rmSync(tempDir, { recursive: true, force: true }); + }); + + // Run a first tick to produce a job, force that job into `status`, then re-arm + // the schedule (past next_run_at) and return its id + the first job id. + async function armWithLastJobStatus(status: string): Promise<{ schedId: number; firstJobId: string }> { + const sched = await repo.createScheduledTask({ + body: 'recurring thing', + pieceName: 'chat', + profile: 'auto', + outputFormat: 'markdown', + cronExpression: '0 9 * * *', + nextRunAt: '2000-01-01 09:00:00', + }); + const first = await scheduler.tick(); + expect(first).toBe(1); + const after = await repo.getScheduledTask(sched.id); + const firstJobId = after!.lastJobId!; + await repo.updateJob(firstJobId, { status: status as any }); + // Re-arm: getScheduledTasksDue advanced next_run_at to the future. + await repo.updateScheduledTask(sched.id, { nextRunAt: '2000-01-01 09:00:00' }); + return { schedId: sched.id, firstJobId }; + } + + it('starts a NEW run when the previous job is waiting_human (no longer blocks)', async () => { + const { schedId, firstJobId } = await armWithLastJobStatus('waiting_human'); + + const executed = await scheduler.tick(); + expect(executed).toBe(1); + + const after = await repo.getScheduledTask(schedId); + expect(after?.lastJobId).toBeTruthy(); + expect(after?.lastJobId).not.toBe(firstJobId); // a fresh job was created + }); + + it('still skips when the previous job is running (genuinely in progress)', async () => { + const { schedId, firstJobId } = await armWithLastJobStatus('running'); + + const executed = await scheduler.tick(); + expect(executed).toBe(0); + + const after = await repo.getScheduledTask(schedId); + expect(after?.lastJobId).toBe(firstJobId); // unchanged — skipped + }); + + it('still skips when the previous job is waiting_subtasks (auto-recovering transient)', async () => { + const { schedId, firstJobId } = await armWithLastJobStatus('waiting_subtasks'); + + const executed = await scheduler.tick(); + expect(executed).toBe(0); + + const after = await repo.getScheduledTask(schedId); + expect(after?.lastJobId).toBe(firstJobId); // unchanged — skipped + }); +}); + describe('Scheduler.executeScheduledTask: task_kind="script"', () => { // Scheduler runs a user-authored script directly (no LLM agent loop) when // task_kind='script'. The job row gets created in a pre-completed state. @@ -393,6 +468,46 @@ module.exports = main; await expect(scheduler.executeById(sched.id)).rejects.toThrow(/script_name is null/); }); + it('runs as the "local" user when the schedule has no owner (no-auth mode)', async () => { + // Regression: no-auth deployments store scheduled tasks with ownerId=null + // (scheduled-tasks-api has no authenticated user). executeScriptScheduledTask + // used to throw "requires an owner_id", so no-auth scheduled scripts could + // never run. Now it falls back to the 'local' namespace (same as the + // RunUserScript tool), resolving the script from data/users/local/scripts/. + writeScript('local', 'noauth.js', `async function main() { + return { ran: 'no-auth-local' }; +} +module.exports = main; +`); + + const sched = await repo.createScheduledTask({ + body: '', + cronExpression: '0 9 * * *', + nextRunAt: '2099-01-01 09:00:00', + // ownerId intentionally omitted → null, as in no-auth mode + taskKind: 'script', + scriptName: 'noauth', + }); + + await scheduler.executeById(sched.id); + + const after = await repo.getScheduledTask(sched.id); + expect(after?.lastJobId).toBeTruthy(); + const job = await repo.getJob(after!.lastJobId!); + expect(job?.status).toBe('succeeded'); + + const task = await repo.getLocalTask(job!.issueNumber); + const output = readFileSync(join(task!.workspacePath!, 'output', 'script-output.txt'), 'utf-8'); + expect(output).toContain('no-auth-local'); // script's return value was saved + + // The audit row records the resolved 'local' owner, not null. + const audits = repo.getDb() + .prepare("SELECT detail FROM audit_log WHERE job_id = ? AND action = 'user_script_run'") + .all(after!.lastJobId!) as Array<{ detail: string }>; + expect(audits).toHaveLength(1); + expect((JSON.parse(audits[0].detail) as { userId: string }).userId).toBe('local'); + }); + it('refuses to run when the user-script gate is disabled', async () => { const gatedScheduler = new Scheduler(repo, join(tempDir, 'workspaces'), { userFolderRoot, diff --git a/src/scheduler.ts b/src/scheduler.ts index 773dcd7..ad46b55 100644 --- a/src/scheduler.ts +++ b/src/scheduler.ts @@ -7,9 +7,16 @@ import { logger } from './logger.js'; import { loadConfig } from './config.js'; import { resolveAndRunUserScript } from './user-folder/script-orchestrator.js'; -// 進行中とみなすステータス（これらの場合はスキップ） +// 進行中とみなすステータス（これらの場合は次回スケジュール実行をスキップ）。 +// これらはいずれも「真に実行中」か「自動回復する一時状態」のみ: +// queued/dispatching/running … 実行中 +// waiting_subtasks … 並列サブタスク完了待ち。requeueWaitingSubtasks が自動回復させる +// waiting_human (ASK) は意図的に含めない。自動回復もタイムアウトも無いため、 +// 無人実行のスケジュールタスクが一度 ASK を出すとその後の実行が永久にスキップされ、 +// スケジュールが事実上死ぬ。各スケジュール実行は独立した新規タスクなので、 +// 前回の waiting_human タスクは残置し（ユーザーが後から回答 or 放置できる）、新規実行を走らせる。 const IN_PROGRESS_STATUSES = new Set([ - 'queued', 'dispatching', 'running', 'waiting_human', 'waiting_subtasks', + 'queued', 'dispatching', 'running', 'waiting_subtasks', ]); export interface ScheduleInput { @@ -295,9 +302,12 @@ export class Scheduler { if (!this.userFolderRoot) { throw new Error(`scheduled_task=${item.id}: task_kind='script' but Scheduler.userFolderRoot was not configured`); } - if (!item.ownerId) { - throw new Error(`scheduled_task=${item.id}: task_kind='script' requires an owner_id (scripts are per-user)`); - } + // No-auth mode stores scheduled tasks with ownerId=null (scheduled-tasks-api + // has no authenticated user). Scripts are per-user (data/users/{id}/scripts/), + // so resolve to the same 'local' namespace the RunUserScript tool uses in + // no-auth (ctx.userId='local'). In auth mode item.ownerId is always set, so + // scriptOwner === item.ownerId and behaviour is unchanged. + const scriptOwner = item.ownerId ?? 'local'; // Same security gates as the LLM-facing RunUserScript tool: global config // toggle + optional per-user allowlist. A scheduled run is automated, so a @@ -308,9 +318,9 @@ export class Scheduler { `scheduled_task=${item.id}: user scripts are disabled (tools.user_scripts_enabled=false)`, ); } - if (Array.isArray(gate.allowUserids) && gate.allowUserids.length > 0 && !gate.allowUserids.includes(item.ownerId)) { + if (Array.isArray(gate.allowUserids) && gate.allowUserids.length > 0 && !gate.allowUserids.includes(scriptOwner)) { throw new Error( - `scheduled_task=${item.id}: owner "${item.ownerId}" is not in tools.user_scripts_allow_userids`, + `scheduled_task=${item.id}: owner "${scriptOwner}" is not in tools.user_scripts_allow_userids`, ); } @@ -375,7 +385,7 @@ export class Scheduler { try { const runResult = await resolveAndRunUserScript({ rootDir: this.userFolderRoot, - userId: item.ownerId, + userId: scriptOwner, name: item.scriptName, params, sessRepo: this.sessRepo, @@ -417,7 +427,7 @@ export class Scheduler { await this.repo.addAuditLog(job.id, 'user_script_run', 'scheduler', { scheduledTaskId: item.id, - userId: item.ownerId, + userId: scriptOwner, scriptName: item.scriptName, ok: !runFailed, ...(runFailed && errorMessage ? { error: errorMessage.slice(0, 500) } : {}), diff --git a/src/worker.reflection-enqueue.test.ts b/src/worker.reflection-enqueue.test.ts index b5f1d3b..126ded9 100644 --- a/src/worker.reflection-enqueue.test.ts +++ b/src/worker.reflection-enqueue.test.ts @@ -67,15 +67,44 @@ describe('maybeEnqueueReflection', () => { expect(await repo.getJobsByStatus('queued')).toHaveLength(0); }); - it('does nothing when ownerId is null (no user to learn for)', async () => { + it('falls back to the "local" namespace when ownerId is null (no-auth mode)', async () => { + // No-auth mode runs every job with ownerId=null. Reflection must still run, + // keyed to the same 'local' namespace the rest of the no-auth path uses. const job = await repo.createJob({ repo: 'local/task-1', issueNumber: 1, instruction: 'x', pieceName: 'chat', - } as any); // no ownerId + } as any); // no ownerId → null in no-auth await repo.updateJob(job.id, { status: 'succeeded' }); await maybeEnqueueReflection(repo, job, 'succeeded', { enabled: true, workerRequired: false, perUserDailyBudgetTokens: 0, }); + const queued = await repo.getJobsByStatus('queued'); + expect(queued).toHaveLength(1); + expect(queued[0].taskKind).toBe('reflection'); + expect(queued[0].ownerId).toBe('local'); + expect(JSON.parse(queued[0].payload!).userId).toBe('local'); + }); + + it('no-auth budget check is keyed to "local" (not null)', async () => { + // A prior null-owner reflection that spent the whole budget must block the + // next null-owner enqueue — proving the budget query uses 'local', not null. + const job = await repo.createJob({ + repo: 'local/task-8', issueNumber: 8, + instruction: 'x', pieceName: 'chat', + } as any); // no ownerId + await repo.updateJob(job.id, { status: 'succeeded' }); + + const cap = 1_000_000; + const todayStartMs = Date.UTC( + new Date().getUTCFullYear(), + new Date().getUTCMonth(), + new Date().getUTCDate(), + ); + seedMetric('local', 600_000, 500_000, todayStartMs + 1_000); // total = 1 100 000 > cap + + await maybeEnqueueReflection(repo, job, 'succeeded', { + enabled: true, workerRequired: false, perUserDailyBudgetTokens: cap, + }); expect(await repo.getJobsByStatus('queued')).toHaveLength(0); }); diff --git a/src/worker.ts b/src/worker.ts index 4d4b197..0b63d60 100644 --- a/src/worker.ts +++ b/src/worker.ts @@ -271,10 +271,12 @@ export async function maybeEnqueueReflection( ): Promise { if (!cfg.enabled) return; if (job.taskKind === 'reflection') return; - if (!job.ownerId) { - logger.warn(`[reflection] skip enqueue job=${job.id} reason=no_owner`); - return; - } + // No-auth mode runs every job with ownerId=null. Reflection is per-user + // (memory/pieces live under data/users/{userId}/), so fall back to the same + // 'local' namespace the rest of the no-auth path uses (ToolContext, pieces, + // user-folder). Without this the enqueue gate skipped forever and reflection + // silently never ran in no-auth deployments. + const reflectionOwner = job.ownerId ?? 'local'; // worker_required enforcement: when true, at least one worker must have 'reflection' in its roles if (cfg.workerRequired) { @@ -282,7 +284,7 @@ export async function maybeEnqueueReflection( (w) => Array.isArray(w.roles) && w.roles.includes('reflection'), ); if (!hasReflectionWorker) { - logger.warn(`[reflection] enqueue skipped reason=no_reflection_worker user=${job.ownerId}`); + logger.warn(`[reflection] enqueue skipped reason=no_reflection_worker user=${reflectionOwner}`); return; } } @@ -294,19 +296,19 @@ export async function maybeEnqueueReflection( // Compute today's start in UTC (00:00:00.000 UTC). const now = new Date(); const todayStartMs = Date.UTC(now.getUTCFullYear(), now.getUTCMonth(), now.getUTCDate()); - const metrics = repo.aggregateReflectionMetrics(job.ownerId, todayStartMs); + const metrics = repo.aggregateReflectionMetrics(reflectionOwner, todayStartMs); const spent = metrics.tokensIn + metrics.tokensOut; if (spent >= cap) { const spentM = (spent / 1_000_000).toFixed(1); const capM = (cap / 1_000_000).toFixed(1); - logger.info(`[reflection] enqueue skipped reason=budget user=${job.ownerId} spent=${spentM}M cap=${capM}M`); + logger.info(`[reflection] enqueue skipped reason=budget user=${reflectionOwner} spent=${spentM}M cap=${capM}M`); return; } } const payload = JSON.stringify({ originalJobId: job.id, - userId: job.ownerId, + userId: reflectionOwner, pieceName: job.pieceName, outcome, }); @@ -316,12 +318,12 @@ export async function maybeEnqueueReflection( instruction: '', pieceName: 'reflection', role: 'reflection', - ownerId: job.ownerId, + ownerId: reflectionOwner, visibility: 'private', taskKind: 'reflection', payload, } as any); - logger.info(`[reflection] enqueued original=${job.id} owner=${job.ownerId} piece=${job.pieceName} outcome=${outcome}`); + logger.info(`[reflection] enqueued original=${job.id} owner=${reflectionOwner} piece=${job.pieceName} outcome=${outcome}`); } export class Worker { diff --git a/ui/index.html b/ui/index.html index a73135a..7521107 100644 --- a/ui/index.html +++ b/ui/index.html @@ -39,7 +39,7 @@ } html { background: #ffffff; } - html[data-theme="dark"] { background: #0a0a0c; } + html[data-theme="dark"] { background: #18181c; } body { margin: 0; diff --git a/ui/src/components/chat/ChatPane.tsx b/ui/src/components/chat/ChatPane.tsx index 7e1773b..ba167e2 100644 --- a/ui/src/components/chat/ChatPane.tsx +++ b/ui/src/components/chat/ChatPane.tsx @@ -453,17 +453,24 @@ export function ChatPane({ task, comments, onSubmit, onCancel, onOpenDetail }: C )} @@ -471,8 +478,12 @@ export function ChatPane({ task, comments, onSubmit, onCancel, onOpenDetail }: C )} diff --git a/ui/src/components/settings/BrowserSettingsForm.tsx b/ui/src/components/settings/BrowserSettingsForm.tsx index 504fd1e..11b3721 100644 --- a/ui/src/components/settings/BrowserSettingsForm.tsx +++ b/ui/src/components/settings/BrowserSettingsForm.tsx @@ -48,6 +48,28 @@ export function BrowserSettingsForm({ config, onChange }: SectionFormProps) {

Sessions (CDP)

+
+ CAPTCHA Solve Mode + + + novnc にすると WebSearch 等が CAPTCHA を踏んだとき共有 noVNC セッション（CAPTCHA Pool）を立て、 + CAPTCHA タブで手動解決できる。ホストに Xvfb / x11vnc / websockify が必要。 + 未インストールなら自動的に headless にフォールバックする。skip（既定）では CAPTCHA は手動解決されない。 + +
+ +
+ Max CAPTCHA Pages + onChange('browser.maxCaptchaPages', Number(v))} /> + CAPTCHA Pool が同時に開けるページ数の上限（novnc モード時のみ有効）。デフォルト: 5 +
+
VNC Base Port = the light-mode + // pairing (~3.9:1 for the default blue — that pattern is inherently sub-AA in + // BOTH themes since a mid-bright brand color can't reach 4.5:1 on any visible + // tint; this just ensures dark is not a regression). Near-white row titles ~15:1. + const isDark = document.documentElement.dataset.theme === 'dark'; + const softL = isDark ? 0.1 : Math.min(0.96, Math.max(0.92, 1 - l * 0.05)); + const softS = isDark ? Math.min(s, 0.5) : Math.min(s, 0.55); // Luminance for contrast decision (BT.601). const luminance = (0.299 * rgb.r + 0.587 * rgb.g + 0.114 * rgb.b) / 255; const fg = luminance > 0.65 ? '#0f172a' : '#ffffff'; @@ -145,8 +155,17 @@ export function useBranding(): Branding { useEffect(() => { document.title = branding.appName; - applyBrandColors(branding.primaryColor); applyFavicon(branding.faviconUrl); + const applyColors = () => applyBrandColors(branding.primaryColor); + applyColors(); + // The soft tint is theme-dependent → re-derive when the theme changes. + window.addEventListener(THEME_CHANGE_EVENT, applyColors); + const mq = window.matchMedia('(prefers-color-scheme: dark)'); + mq.addEventListener('change', applyColors); + return () => { + window.removeEventListener(THEME_CHANGE_EVENT, applyColors); + mq.removeEventListener('change', applyColors); + }; }, [branding.appName, branding.primaryColor, branding.faviconUrl]); return branding; diff --git a/ui/src/index.css b/ui/src/index.css index 3d8a8b2..8077af8 100644 --- a/ui/src/index.css +++ b/ui/src/index.css @@ -25,6 +25,9 @@ /* Dark accent fallback (used only when /api/branding hasn't set --brand-primary*). Full runtime-branding + WCAG pairing is Phase 2. */ --accent-on-dark: #fafafa; --accent-on-dark-fg: #18181b; + /* Selected/active soft tint default (no-branding case). useBranding sets + --brand-primary-soft inline when configured (theme-aware, see hook). */ + --brand-primary-soft: #f4f4f5; } [data-theme="dark"] { @@ -32,18 +35,21 @@ /* Inverted neutral ramp (zinc-tuned). TUNE ON A REAL DISPLAY — dense UI borders rely on hairline; verify surface/surface-2/hairline separation and WCAG AA text contrast before shipping (spec §4.1 #6). */ - --slate-50: #0a0a0c; --slate-100: #131316; --slate-200: #202024; - /* slate-400/500 lifted to clear WCAG AA on canvas (#0a0a0c): secondary - text (timestamps, version, meta) was ~3:1 at #52525b — too dim. */ - --slate-300: #383840; --slate-400: #7a7a85; --slate-500: #9b9ba5; - --slate-600: #a1a1aa; --slate-700: #c4c4cc; --slate-800: #dedee2; - --slate-900: #f1f1f3; --slate-950: #fafafa; + /* Dark backgrounds lifted a notch (was #0a0a0c/#16161a/#202024) for a + softer, less pitch-black feel per user feedback. */ + --slate-50: #18181c; --slate-100: #202024; --slate-200: #2a2a30; + /* slate-400/500 lifted to clear WCAG AA on the (now brighter) canvas. */ + --slate-300: #44444d; --slate-400: #8a8a95; --slate-500: #a6a6b0; + --slate-600: #b4b4bd; --slate-700: #c8c8d0; --slate-800: #e0e0e4; + --slate-900: #f2f2f4; --slate-950: #fafafa; --gray-400: #6b7280; --gray-500: #9ca3af; --gray-700: #d1d5db; - --canvas: #0a0a0c; --surface: #16161a; --surface-2: #202024; - --hairline: #2e2e34; --hairline-soft: #202024; + --canvas: #18181c; --surface: #202024; --surface-2: #2a2a30; + --hairline: #3a3a42; --hairline-soft: #2a2a30; --ink: #e7e7ea; --muted: #a1a1aa; --scrollbar-thumb: #3f3f46; --scrollbar-thumb-hover: #52525b; --accent-on-dark: #fafafa; --accent-on-dark-fg: #18181b; + /* Dark default for the selected/active soft tint (no-branding case). */ + --brand-primary-soft: #2c2c34; } /* When stored pref is 'system' the inline script sets data-theme already,