diff --git a/docs/assets/readme-downloads.svg b/docs/assets/readme-downloads.svg index 39008eaa963..3ea69247a26 100644 --- a/docs/assets/readme-downloads.svg +++ b/docs/assets/readme-downloads.svg @@ -1,5 +1,5 @@ - - downloads: 45m + + downloads: 46m @@ -15,7 +15,7 @@ downloads downloads - 45m - 45m + 46m + 46m diff --git a/src/main/agent-hooks/first-work-branch-rename.test.ts b/src/main/agent-hooks/first-work-branch-rename.test.ts index 9cd4d544041..8ecd4997aa0 100644 --- a/src/main/agent-hooks/first-work-branch-rename.test.ts +++ b/src/main/agent-hooks/first-work-branch-rename.test.ts @@ -100,10 +100,14 @@ describe('maybeAutoRenameBranchOnFirstWork', () => { isPendingFirstAgentMessageRename: () => true }) const items: AgentJournalRenderItem[] = [] + // A real journal's sequence only ever advances, so the feed's projection + // cache must miss on every publish here: this test is about the rename. + let sequence = 0 const journal = { snapshot: () => ({ items }), lastActivityAt: () => 1, - isReadOnly: false + isReadOnly: false, + cursor: () => ({ epoch: 1, sequence: (sequence += 1) }) } as unknown as AgentSessionJournal const pending: Promise[] = [] const observe = vi.fn((summary, options) => { @@ -175,6 +179,7 @@ describe('maybeAutoRenameBranchOnFirstWork', () => { const journal = { isReadOnly: false, lastActivityAt: () => 1, + cursor: () => ({ epoch: 1, sequence: 1 }), snapshot: () => ({ items: [ { body: { kind: 'message', role: 'user', blocks: [{ type: 'text', text: 'Fix auth' }] } }, diff --git a/src/main/claude/claude-background-task-frames.ts b/src/main/claude/claude-background-task-frames.ts new file mode 100644 index 00000000000..d954128d91b --- /dev/null +++ b/src/main/claude/claude-background-task-frames.ts @@ -0,0 +1,106 @@ +// Field readers for the Claude SDK's background-task lifecycle frames +// (task_started / task_updated / task_notification / background_tasks_changed). +// Pure and bounded: every reader rejects absent, non-string, or oversized +// values so a malformed frame degrades to "field unknown", never to a throw. + +import type { + AgentSessionBackgroundTask, + AgentSessionBackgroundTaskRunState +} from '../../shared/agent-session-wire' + +const MAX_TASK_ID_LENGTH = 512 +const MAX_TASK_TEXT_LENGTH = 512 + +export type ClaudeBackgroundTaskKind = AgentSessionBackgroundTask['kind'] + +export function record(value: unknown): Record | null { + return typeof value === 'object' && value !== null ? (value as Record) : null +} + +/** The bound every task id shares, wherever it enters. An id the roster stores + * becomes a durable entry key, so a provisional one takes the same bound the + * announced path applies — an over-long id is rejected, never truncated. */ +export function isBoundedClaudeTaskId(value: string): boolean { + return value.length > 0 && value.length <= MAX_TASK_ID_LENGTH +} + +export function taskId(message: Record): string | null { + const value = message.task_id + return typeof value === 'string' && isBoundedClaudeTaskId(value) ? value : null +} + +function boundedTaskText(value: unknown): string | undefined { + if (typeof value !== 'string') { + return undefined + } + const trimmed = value.trim().replace(/\s+/g, ' ') + return trimmed.length > 0 ? trimmed.slice(0, MAX_TASK_TEXT_LENGTH) : undefined +} + +export function taskDescription(value: unknown): string | undefined { + return boundedTaskText(value) +} + +/** The provider-reported identity for a task. Subagent frames have carried the + * type under both `agent_type` and `subagent_type` across SDK versions. */ +export function taskName(frame: Record): string | undefined { + return ( + boundedTaskText(frame.name) ?? + boundedTaskText(frame.agent_type) ?? + boundedTaskText(frame.subagent_type) + ) +} + +export function classifyClaudeBackgroundTaskKind(taskType: unknown): ClaudeBackgroundTaskKind { + switch (taskType) { + case 'local_agent': + return 'agent' + case 'local_workflow': + return 'workflow' + case 'local_bash': + return 'command' + case 'monitor': + return 'monitor' + default: + return 'unknown' + } +} + +/** Cumulative token usage from a task_progress / task_notification frame. */ +export function taskUsageTotalTokens(frame: Record): number | undefined { + const usage = record(frame.usage) + const total = usage?.total_tokens + return typeof total === 'number' && Number.isFinite(total) && total >= 0 + ? Math.floor(total) + : undefined +} + +/** Settled state for a terminal status. Null for anything else — an unreadable + * status never settles a task by itself. */ +export function terminalClaudeTaskRunState( + status: unknown +): AgentSessionBackgroundTaskRunState | null { + switch (status) { + case 'completed': + return 'done' + case 'failed': + return 'blocked' + case 'killed': + case 'stopped': + return 'idle' + default: + return null + } +} + +/** Live state for a non-terminal status. Null leaves the tracked state alone. */ +export function liveClaudeTaskRunState(status: unknown): AgentSessionBackgroundTaskRunState | null { + switch (status) { + case 'pending': + case 'running': + case 'paused': + return 'working' + default: + return null + } +} diff --git a/src/main/claude/claude-background-task-resume.test.ts b/src/main/claude/claude-background-task-resume.test.ts new file mode 100644 index 00000000000..c043d485304 --- /dev/null +++ b/src/main/claude/claude-background-task-resume.test.ts @@ -0,0 +1,92 @@ +import { describe, expect, it } from 'vitest' +import { ClaudeBackgroundTaskTracker } from './claude-background-task-tracker' + +const agent = { + task_id: 'a962f88aa82feb1c1', + task_type: 'local_agent', + description: 'Long proof writer' +} +const shell = { task_id: 'bcl6x3ixf', task_type: 'local_bash', description: 'sleep 150' } +const sibling = { task_id: 'sibling', task_type: 'local_agent' } + +function system(subtype: string, fields: Record) { + return { type: 'system', subtype, ...fields } +} + +describe('Claude background task pause/resume ownership', () => { + it('moves a retained child back to live ownership across eviction, outcome, and auto-resume', () => { + let now = 100 + const tracker = new ClaudeBackgroundTaskTracker(() => now) + const roster = (tasks: unknown[]) => + tracker.observe(system('background_tasks_changed', { tasks })) + roster([agent, sibling, shell]) + tracker.observe( + system('task_progress', { task_id: agent.task_id, usage: { total_tokens: 18000 } }) + ) + tracker.observe({ type: 'result' }) + expect(tracker.state?.tasks).toHaveLength(3) + roster([sibling, shell]) + expect(tracker.state?.settledTasks).toBeUndefined() + tracker.observe( + system('task_updated', { task_id: agent.task_id, patch: { status: 'completed' } }) + ) + tracker.observe( + system('task_notification', { + task_id: agent.task_id, + status: 'completed', + usage: { total_tokens: 19003 } + }) + ) + expect(tracker.state?.settledTasks).toEqual([ + expect.objectContaining({ + id: agent.task_id, + state: 'done', + startedAt: 100, + totalTokens: 19003 + }) + ]) + now = 150000 + roster([agent, sibling, shell]) + expect(tracker.state?.settledTasks).toBeUndefined() + expect(tracker.state?.tasks).toEqual([ + expect.objectContaining({ + id: agent.task_id, + state: 'working', + startedAt: 100, + totalTokens: 19003 + }), + expect.objectContaining({ id: sibling.task_id }), + expect.objectContaining({ id: shell.task_id }) + ]) + expect(tracker.stoppableTaskIds).toEqual([agent.task_id, sibling.task_id, shell.task_id]) + roster([sibling, shell]) + tracker.observe( + system('task_notification', { + task_id: agent.task_id, + status: 'completed', + usage: { total_tokens: 21000 } + }) + ) + expect(tracker.state?.settledTasks).toEqual([ + expect.objectContaining({ id: agent.task_id, startedAt: 100, totalTokens: 21000 }) + ]) + roster([]) + expect(tracker.state).toBeNull() + }) + + it('reconciles an edge-only resume without keeping its earlier settled copy', () => { + const tracker = new ClaudeBackgroundTaskTracker(() => 100) + for (const task of [agent, sibling]) { + tracker.observe(system('task_started', { ...task, is_backgrounded: true })) + } + tracker.observe(system('task_notification', { task_id: agent.task_id, status: 'completed' })) + tracker.observe( + system('task_updated', { + task_id: agent.task_id, + patch: { status: 'running', is_backgrounded: true } + }) + ) + expect(tracker.state?.tasks).toHaveLength(2) + expect(tracker.state?.settledTasks).toBeUndefined() + }) +}) diff --git a/src/main/claude/claude-background-task-tracker.test.ts b/src/main/claude/claude-background-task-tracker.test.ts index d8f316d7dcd..df89a64a968 100644 --- a/src/main/claude/claude-background-task-tracker.test.ts +++ b/src/main/claude/claude-background-task-tracker.test.ts @@ -16,6 +16,11 @@ function aggregate(tasks: unknown[]): Record { return system('background_tasks_changed', { tasks }) } +function trackerAt(times: number[]): ClaudeBackgroundTaskTracker { + let index = 0 + return new ClaudeBackgroundTaskTracker(() => times[Math.min(index++, times.length - 1)]) +} + describe('ClaudeBackgroundTaskTracker', () => { it('classifies SDK task types without inferring them from descriptions', () => { expect(classifyClaudeBackgroundTaskKind('local_agent')).toBe('agent') @@ -25,27 +30,30 @@ describe('ClaudeBackgroundTaskTracker', () => { expect(classifyClaudeBackgroundTaskKind('future_task')).toBe('unknown') }) - it('waits for the foreground turn to settle before monitoring a background task', () => { - const tracker = new ClaudeBackgroundTaskTracker() + it('publishes a backgrounded task while the foreground turn is still running', () => { + const tracker = trackerAt([100]) tracker.observe({ type: 'user' }, true) - tracker.observe( - system('task_started', { - task_id: 'task-1', - task_type: 'local_agent', - is_backgrounded: true - }) - ) - expect(tracker.state).toBeNull() - - expect(tracker.observe(result())).toBe(true) + expect( + tracker.observe( + system('task_started', { + task_id: 'task-1', + task_type: 'local_agent', + is_backgrounded: true + }) + ) + ).toBe(true) expect(tracker.state).toEqual({ state: 'monitoring', - tasks: [{ id: 'task-1', kind: 'agent' }] + tasks: [{ id: 'task-1', kind: 'agent', state: 'working', startedAt: 100 }] }) + + // The turn settling changes nothing the strip renders. + expect(tracker.observe(result())).toBe(false) + expect(tracker.state?.tasks).toHaveLength(1) }) it('uses an explicit background update for a foreground task and ignores progress alone', () => { - const tracker = new ClaudeBackgroundTaskTracker() + const tracker = trackerAt([100]) tracker.observe({ type: 'user' }, true) tracker.observe( system('task_started', { @@ -63,12 +71,12 @@ describe('ClaudeBackgroundTaskTracker', () => { tracker.observe(system('task_updated', { task_id: 'task-1', patch: { is_backgrounded: true } })) expect(tracker.state).toEqual({ state: 'monitoring', - tasks: [{ id: 'task-1', kind: 'command' }] + tasks: [{ id: 'task-1', kind: 'command', state: 'working', startedAt: 100 }] }) }) it('publishes bounded display details when a running task description changes', () => { - const tracker = new ClaudeBackgroundTaskTracker() + const tracker = trackerAt([100]) expect( tracker.observe( system('task_started', { @@ -81,7 +89,15 @@ describe('ClaudeBackgroundTaskTracker', () => { ).toBe(true) expect(tracker.state).toEqual({ state: 'monitoring', - tasks: [{ id: 'task-1', kind: 'command', description: 'run the build' }] + tasks: [ + { + id: 'task-1', + kind: 'command', + description: 'run the build', + state: 'working', + startedAt: 100 + } + ] }) expect( @@ -103,8 +119,193 @@ describe('ClaudeBackgroundTaskTracker', () => { ).toBe(false) }) + it('carries provider-reported names and re-derives classification per transition', () => { + const tracker = trackerAt([100]) + tracker.observe( + system('task_started', { + task_id: 'task-1', + task_type: 'future_task', + is_backgrounded: true + }) + ) + expect(tracker.state?.tasks?.[0]).toMatchObject({ kind: 'unknown' }) + + expect( + tracker.observe( + system('task_updated', { + task_id: 'task-1', + patch: { task_type: 'local_agent', agent_type: 'deep_review' } + }) + ) + ).toBe(true) + expect(tracker.state?.tasks?.[0]).toMatchObject({ + kind: 'agent', + name: 'deep_review', + state: 'working' + }) + }) + + it('retains settled siblings beside live work and exits with the last live task', () => { + const tracker = trackerAt([100, 200]) + tracker.observe( + system('task_started', { task_id: 'task-a', task_type: 'local_agent', is_backgrounded: true }) + ) + tracker.observe( + system('task_started', { task_id: 'task-b', task_type: 'local_agent', is_backgrounded: true }) + ) + + expect( + tracker.observe(system('task_updated', { task_id: 'task-a', patch: { status: 'completed' } })) + ).toBe(true) + expect(tracker.state).toEqual({ + state: 'monitoring', + tasks: [{ id: 'task-b', kind: 'agent', state: 'working', startedAt: 200 }], + settledTasks: [{ id: 'task-a', kind: 'agent', state: 'done', startedAt: 100 }] + }) + expect(tracker.stoppableTaskIds).toEqual(['task-b']) + + expect( + tracker.observe(system('task_updated', { task_id: 'task-b', patch: { status: 'killed' } })) + ).toBe(true) + expect(tracker.state).toBeNull() + }) + + it('settles a sibling from the captured producer order: aggregate eviction, then the outcome', () => { + // Verbatim sequence from a real SDK capture (2026-09-07): the aggregate + // roster arrives FIRST, already missing the finished task, and the + // terminal edges trail in the same tick. + const tracker = trackerAt([100, 200]) + tracker.observe( + system('task_started', { + task_id: 'bh4zn8der', + tool_use_id: 'toolu_01M', + description: 'Sleep for 5 seconds', + is_backgrounded: true, + task_type: 'local_bash' + }) + ) + tracker.observe( + aggregate([ + { task_id: 'bh4zn8der', task_type: 'local_bash', description: 'Sleep for 5 seconds' }, + { task_id: 'bprosaiim', task_type: 'local_bash', description: 'Sleep for 25 seconds' } + ]) + ) + + // The settling child is evicted by the aggregate before any outcome frame. + tracker.observe( + aggregate([ + { task_id: 'bprosaiim', task_type: 'local_bash', description: 'Sleep for 25 seconds' } + ]) + ) + tracker.observe( + system('task_updated', { + task_id: 'bh4zn8der', + patch: { status: 'completed', end_time: 1788804376515 } + }) + ) + expect( + tracker.observe( + system('task_notification', { + task_id: 'bh4zn8der', + tool_use_id: 'toolu_01M', + status: 'completed', + summary: 'Background command "Sleep for 5 seconds" completed (exit code 0)', + usage: { total_tokens: 18130, tool_uses: 1, duration_ms: 10772 } + }) + ) + ).toBe(true) + expect(tracker.state).toEqual({ + state: 'monitoring', + tasks: [ + { + id: 'bprosaiim', + kind: 'command', + description: 'Sleep for 25 seconds', + state: 'working', + startedAt: 200 + } + ], + settledTasks: [ + { + id: 'bh4zn8der', + kind: 'command', + description: 'Sleep for 5 seconds', + state: 'done', + startedAt: 100, + totalTokens: 18130 + } + ] + }) + + // Last task killed, same captured order: the strip exits. + tracker.observe(aggregate([])) + tracker.observe(system('task_updated', { task_id: 'bprosaiim', patch: { status: 'killed' } })) + tracker.observe(system('task_notification', { task_id: 'bprosaiim', status: 'stopped' })) + expect(tracker.state).toBeNull() + }) + + it('carries task_progress usage into a live row without clobbering its name', () => { + const tracker = trackerAt([100]) + tracker.observe( + system('task_started', { + task_id: 'agent-1', + task_type: 'local_agent', + subagent_type: 'general-purpose', + description: 'Sleep 6 seconds test', + is_backgrounded: true + }) + ) + expect( + tracker.observe( + system('task_progress', { + task_id: 'agent-1', + description: 'Running Sleep for 6 seconds', + subagent_type: 'general-purpose', + usage: { total_tokens: 14866, tool_uses: 1, duration_ms: 2818 }, + last_tool_name: 'Bash' + }) + ) + ).toBe(true) + expect(tracker.state?.tasks?.[0]).toEqual({ + id: 'agent-1', + kind: 'agent', + // Progress descriptions are transient activity, never the task's name. + description: 'Sleep 6 seconds test', + name: 'general-purpose', + state: 'working', + startedAt: 100, + totalTokens: 14866 + }) + }) + + it('maps terminal statuses onto settled states', () => { + const tracker = trackerAt([100, 200]) + tracker.observe( + system('task_started', { task_id: 'live', task_type: 'local_agent', is_backgrounded: true }) + ) + tracker.observe( + system('task_started', { task_id: 'failed', task_type: 'local_agent', is_backgrounded: true }) + ) + tracker.observe(system('task_notification', { task_id: 'failed', status: 'failed' })) + expect(tracker.state?.settledTasks).toEqual([ + { id: 'failed', kind: 'agent', state: 'blocked', startedAt: 200 } + ]) + }) + + it('leaves a task open when a patch cannot be read', () => { + const tracker = trackerAt([100]) + tracker.observe( + system('task_started', { task_id: 'task-1', task_type: 'local_agent', is_backgrounded: true }) + ) + expect(tracker.observe(system('task_updated', { task_id: 'task-1', patch: 'garbage' }))).toBe( + false + ) + expect(tracker.state?.tasks).toHaveLength(1) + expect(tracker.state?.settledTasks).toBeUndefined() + }) + it('replaces its roster from aggregate lifecycle frames and preserves stoppable provider ids', () => { - const tracker = new ClaudeBackgroundTaskTracker() + const tracker = trackerAt([100]) expect( tracker.observe( aggregate([ @@ -117,8 +318,8 @@ describe('ClaudeBackgroundTaskTracker', () => { expect(tracker.state).toEqual({ state: 'monitoring', tasks: [ - { id: 'task-agent', kind: 'agent', description: 'agent' }, - { id: 'task-bash', kind: 'command', description: 'bash' } + { id: 'task-agent', kind: 'agent', description: 'agent', state: 'working', startedAt: 100 }, + { id: 'task-bash', kind: 'command', description: 'bash', state: 'working', startedAt: 100 } ] }) @@ -128,18 +329,31 @@ describe('ClaudeBackgroundTaskTracker', () => { ) ).toBe(true) expect(tracker.stoppableTaskIds).toEqual(['task-next']) - expect(tracker.state).toEqual({ - state: 'monitoring', - tasks: [{ id: 'task-next', kind: 'workflow', description: 'workflow' }] - }) expect(tracker.observe(aggregate([]))).toBe(true) expect(tracker.stoppableTaskIds).toEqual([]) expect(tracker.state).toBeNull() }) + it('preserves first-seen timestamps across aggregate roster replacement', () => { + const tracker = trackerAt([100, 200]) + tracker.observe( + system('task_started', { task_id: 'task-1', task_type: 'local_agent', is_backgrounded: true }) + ) + tracker.observe( + aggregate([ + { task_id: 'task-1', task_type: 'local_agent' }, + { task_id: 'task-2', task_type: 'local_bash' } + ]) + ) + expect(tracker.state?.tasks).toEqual([ + { id: 'task-1', kind: 'agent', state: 'working', startedAt: 100 }, + { id: 'task-2', kind: 'command', state: 'working', startedAt: 200 } + ]) + }) + it('excludes ambient aggregate tasks', () => { - const tracker = new ClaudeBackgroundTaskTracker() + const tracker = trackerAt([100]) tracker.observe( aggregate([ { task_id: 'ambient', task_type: 'monitor', description: 'watcher', ambient: true }, @@ -151,7 +365,7 @@ describe('ClaudeBackgroundTaskTracker', () => { }) it('does not let late edge frames revive tasks cleared by an aggregate roster', () => { - const tracker = new ClaudeBackgroundTaskTracker() + const tracker = trackerAt([100]) tracker.observe( aggregate([{ task_id: 'task-late', task_type: 'local_agent', description: 'agent' }]) ) @@ -173,7 +387,7 @@ describe('ClaudeBackgroundTaskTracker', () => { }) it('lets an authoritative aggregate roster replace earlier terminal-edge evidence', () => { - const tracker = new ClaudeBackgroundTaskTracker() + const tracker = trackerAt([100]) tracker.observe(system('task_notification', { task_id: 'task-live', status: 'completed' })) tracker.observe( @@ -183,12 +397,28 @@ describe('ClaudeBackgroundTaskTracker', () => { expect(tracker.stoppableTaskIds).toEqual(['task-live']) expect(tracker.state).toEqual({ state: 'monitoring', - tasks: [{ id: 'task-live', kind: 'agent', description: 'agent' }] + tasks: [ + { id: 'task-live', kind: 'agent', description: 'agent', state: 'working', startedAt: 100 } + ] }) }) + it('retracts a settled copy when an authoritative roster reports the task live again', () => { + const tracker = trackerAt([100, 200, 300]) + const tasks = [ + { task_id: 'agent', task_type: 'local_agent', description: 'Review sample' }, + { task_id: 'shell', task_type: 'local_bash' } + ] + tracker.observe(aggregate(tasks)) + tracker.observe(system('task_notification', { task_id: 'agent', status: 'completed' })) + expect(tracker.state?.settledTasks).toHaveLength(1) + tracker.observe(aggregate(tasks)) + expect(tracker.state?.tasks?.map((task) => task.id)).toEqual(['agent', 'shell']) + expect(tracker.state?.settledTasks).toBeUndefined() + }) + it('keeps terminal edges authoritative on either side of aggregate replacement', () => { - const terminalFirst = new ClaudeBackgroundTaskTracker() + const terminalFirst = trackerAt([100]) terminalFirst.observe( system('task_notification', { task_id: 'task-first', status: 'completed' }) ) @@ -202,7 +432,7 @@ describe('ClaudeBackgroundTaskTracker', () => { ) expect(terminalFirst.state).toBeNull() - const terminalLast = new ClaudeBackgroundTaskTracker() + const terminalLast = trackerAt([100]) terminalLast.observe( aggregate([{ task_id: 'task-last', task_type: 'local_agent', description: 'agent' }]) ) @@ -218,7 +448,7 @@ describe('ClaudeBackgroundTaskTracker', () => { }) it('keeps terminal evidence authoritative across duplicates and out-of-order starts', () => { - const tracker = new ClaudeBackgroundTaskTracker() + const tracker = trackerAt([100]) const terminal = system('task_notification', { task_id: 'task-late', status: 'completed' }) tracker.observe(terminal) tracker.observe(terminal) @@ -239,7 +469,7 @@ describe('ClaudeBackgroundTaskTracker', () => { ) expect(tracker.state).toEqual({ state: 'monitoring', - tasks: [{ id: 'task-live', kind: 'monitor' }] + tasks: [{ id: 'task-live', kind: 'monitor', state: 'monitoring', startedAt: 100 }] }) expect( tracker.observe(system('task_updated', { task_id: 'task-live', patch: { status: 'killed' } })) @@ -249,17 +479,24 @@ describe('ClaudeBackgroundTaskTracker', () => { it('recognizes task types that are registered only as background work', () => { for (const taskType of ['local_workflow', 'monitor']) { - const tracker = new ClaudeBackgroundTaskTracker() + const tracker = trackerAt([100]) tracker.observe(system('task_started', { task_id: taskType, task_type: taskType })) expect(tracker.state).toEqual({ state: 'monitoring', - tasks: [{ id: taskType, kind: taskType === 'local_workflow' ? 'workflow' : 'monitor' }] + tasks: [ + { + id: taskType, + kind: taskType === 'local_workflow' ? 'workflow' : 'monitor', + state: taskType === 'local_workflow' ? 'working' : 'monitoring', + startedAt: 100 + } + ] }) } }) it('admits unknown background updates conservatively and bounds edge-only fallback ids', () => { - const tracker = new ClaudeBackgroundTaskTracker() + const tracker = trackerAt([100]) tracker.observe( system('task_updated', { task_id: 'unknown', patch: { is_backgrounded: true } }) ) @@ -278,7 +515,7 @@ describe('ClaudeBackgroundTaskTracker', () => { }) it('bounds aggregate rosters and resets to the edge-only fallback on clear', () => { - const tracker = new ClaudeBackgroundTaskTracker() + const tracker = trackerAt([100]) tracker.observe( aggregate( Array.from({ length: 400 }, (_, index) => ({ @@ -301,23 +538,30 @@ describe('ClaudeBackgroundTaskTracker', () => { expect(tracker.stoppableTaskIds).toEqual(['edge-after-reset']) }) - it('gates aggregate monitoring behind foreground turn completion', () => { - const tracker = new ClaudeBackgroundTaskTracker() + it('publishes an aggregate roster observed mid-turn', () => { + const tracker = trackerAt([100]) tracker.observe({ type: 'user' }, true) - tracker.observe( - aggregate([{ task_id: 'task-live', task_type: 'local_bash', description: 'command' }]) - ) - expect(tracker.state).toBeNull() - - expect(tracker.observe(result())).toBe(true) + expect( + tracker.observe( + aggregate([{ task_id: 'task-live', task_type: 'local_bash', description: 'command' }]) + ) + ).toBe(true) expect(tracker.state).toEqual({ state: 'monitoring', - tasks: [{ id: 'task-live', kind: 'command', description: 'command' }] + tasks: [ + { + id: 'task-live', + kind: 'command', + description: 'command', + state: 'working', + startedAt: 100 + } + ] }) }) it('ignores ambient SDK tasks and clears all liveness when the session ends', () => { - const tracker = new ClaudeBackgroundTaskTracker() + const tracker = trackerAt([100]) tracker.observe( system('task_started', { task_id: 'ambient', diff --git a/src/main/claude/claude-background-task-tracker.ts b/src/main/claude/claude-background-task-tracker.ts index de24b9a4fba..68c7f4f1185 100644 --- a/src/main/claude/claude-background-task-tracker.ts +++ b/src/main/claude/claude-background-task-tracker.ts @@ -1,78 +1,54 @@ import type { AgentSessionBackgroundTask, + AgentSessionBackgroundTaskRunState, AgentSessionBackgroundTaskState } from '../../shared/agent-session-wire' +import { + classifyClaudeBackgroundTaskKind, + liveClaudeTaskRunState, + record, + taskDescription, + taskId, + taskName, + taskUsageTotalTokens, + terminalClaudeTaskRunState +} from './claude-background-task-frames' +import { + ClaudeSettledBackgroundTasks, + claudeBackgroundTaskDetail, + type TrackedClaudeBackgroundTask +} from './claude-settled-background-tasks' + +// `claude-subagent-*` reads this channel through these names; the readers themselves +// live in the frames module so both consumers share one definition. +export { + classifyClaudeBackgroundTaskKind, + isBoundedClaudeTaskId, + taskDescription as claudeTaskDescription, + taskId as claudeTaskId +} from './claude-background-task-frames' +export type { ClaudeBackgroundTaskKind } from './claude-background-task-frames' const MAX_TRACKED_TASKS = 256 -const MAX_TASK_ID_LENGTH = 512 -const MAX_TASK_DESCRIPTION_LENGTH = 512 -const TERMINAL_TASK_STATES = new Set(['completed', 'failed', 'killed', 'stopped']) - -export type ClaudeBackgroundTaskKind = AgentSessionBackgroundTask['kind'] - -type TrackedTask = { - backgrounded: boolean - kind: ClaudeBackgroundTaskKind - description?: string -} - -function record(value: unknown): Record | null { - return typeof value === 'object' && value !== null ? (value as Record) : null -} - -/** The bound every task id shares, wherever it enters. An id the roster stores - * becomes a durable entry key, so a provisional one takes the same bound the - * announced path applies — an over-long id is rejected, never truncated. */ -export function isBoundedClaudeTaskId(value: string): boolean { - return value.length > 0 && value.length <= MAX_TASK_ID_LENGTH -} - -/** The task's canonical, resume-stable id. Shared with the subagent roster so - * both readers of this channel agree on what identifies a task. */ -export function claudeTaskId(message: Record): string | null { - const value = message.task_id - return typeof value === 'string' && isBoundedClaudeTaskId(value) ? value : null -} - -/** A task's human label, collapsed and bounded. */ -export function claudeTaskDescription(value: unknown): string | undefined { - if (typeof value !== 'string') { - return undefined - } - const trimmed = value.trim().replace(/\s+/g, ' ') - return trimmed.length > 0 ? trimmed.slice(0, MAX_TASK_DESCRIPTION_LENGTH) : undefined -} - -export function classifyClaudeBackgroundTaskKind(taskType: unknown): ClaudeBackgroundTaskKind { - switch (taskType) { - case 'local_agent': - return 'agent' - case 'local_workflow': - return 'workflow' - case 'local_bash': - return 'command' - case 'monitor': - return 'monitor' - default: - return 'unknown' - } -} export class ClaudeBackgroundTaskTracker { - private readonly tasks = new Map() + private readonly tasks = new Map() + private readonly retention = new ClaudeSettledBackgroundTasks() private readonly terminalTaskIds = new Set() private aggregateRosterObserved = false - private foregroundTurnActive = false private monitoring = false private publishedTasksFingerprint = '' + constructor(private readonly now: () => number = () => Date.now()) {} + get state(): AgentSessionBackgroundTaskState | null { if (!this.monitoring) { return null } return { state: 'monitoring', - tasks: this.backgroundTaskDetails() + tasks: this.backgroundTaskDetails(), + ...(this.retention.hasSettled ? { settledTasks: this.retention.settledDetails() } : {}) } } @@ -87,16 +63,14 @@ export class ClaudeBackgroundTaskTracker { } observe(message: Record, startsTurn = false): boolean { - if (startsTurn) { - this.foregroundTurnActive = true - } - if (message.type === 'result') { - this.foregroundTurnActive = false - } else if (message.type === 'system') { + // Background work publishes through a foreground turn: the strip stays + // honest mid-fan-out and the client alone decides when the idle-only + // monitoring label may speak. + if (message.type === 'system') { if (!this.observeSystemFrame(message) && !startsTurn) { return false } - } else if (!startsTurn) { + } else if (!startsTurn && message.type !== 'result') { return false } return this.refreshMonitoring() @@ -104,47 +78,51 @@ export class ClaudeBackgroundTaskTracker { clear(): boolean { this.tasks.clear() + this.retention.clear() this.terminalTaskIds.clear() this.aggregateRosterObserved = false - this.foregroundTurnActive = false return this.refreshMonitoring() } + private settle( + id: string, + state: AgentSessionBackgroundTaskRunState, + outcome: { totalTokens?: number } = {} + ): void { + this.retention.settle(id, state, outcome, this.tasks.get(id)) + this.finish(id) + } + private observeSystemFrame(message: Record): boolean { if (message.subtype === 'background_tasks_changed') { this.replaceAggregateRoster(message.tasks) return true } - const id = claudeTaskId(message) + const id = taskId(message) if (!id) { return false } if (message.subtype === 'task_notification') { - this.finish(id) + // The notification is affirmative terminal evidence even when its status + // field is unreadable — matching the liveness semantics this edge always had. + this.settle(id, terminalClaudeTaskRunState(message.status) ?? 'done', { + totalTokens: taskUsageTotalTokens(message) + }) + return true + } + if (message.subtype === 'task_progress') { + // Progress `description` is the current activity ("Running "), not + // the task's name — only usage (and a missing identity) may update. + const existing = this.tasks.get(id) + const totalTokens = taskUsageTotalTokens(message) + if (!existing?.backgrounded || totalTokens === undefined) { + return false + } + this.tasks.set(id, { ...existing, totalTokens, name: existing.name ?? taskName(message) }) return true } if (message.subtype === 'task_updated') { - const patch = record(message.patch) - if (!patch) { - return false - } - if (TERMINAL_TASK_STATES.has(String(patch.status))) { - this.finish(id) - return true - } - const existing = this.tasks.get(id) - if ( - (patch.is_backgrounded === true || claudeTaskDescription(patch.description)) && - (!this.aggregateRosterObserved || existing) - ) { - this.upsert(id, { - backgrounded: patch.is_backgrounded === true || existing?.backgrounded === true, - kind: existing?.kind ?? 'unknown', - description: claudeTaskDescription(patch.description) ?? existing?.description - }) - return true - } - return false + return this.observeTaskUpdated(id, message) } if (message.subtype !== 'task_started' || this.terminalTaskIds.has(id)) { return false @@ -160,15 +138,55 @@ export class ClaudeBackgroundTaskTracker { this.upsert(id, { backgrounded: message.is_backgrounded === true || kind === 'workflow' || kind === 'monitor', kind, - description: claudeTaskDescription(message.description) + description: taskDescription(message.description), + name: taskName(message), + state: liveClaudeTaskRunState(message.status) ?? undefined, + startedAt: this.now() }) return true } + private observeTaskUpdated(id: string, message: Record): boolean { + const patch = record(message.patch) + if (!patch) { + return false + } + const settledState = terminalClaudeTaskRunState(patch.status) + if (settledState) { + this.settle(id, settledState) + return true + } + const existing = this.tasks.get(id) + // Classification is re-derived per transition: a later frame that reveals a + // real type moves the task between buckets instead of pinning first-seen. + const patchKind = + 'task_type' in patch ? classifyClaudeBackgroundTaskKind(patch.task_type) : undefined + const liveState = liveClaudeTaskRunState(patch.status) + const hasContent = + patch.is_backgrounded === true || + taskDescription(patch.description) !== undefined || + taskName(patch) !== undefined || + liveState !== null || + (patchKind !== undefined && patchKind !== 'unknown') + if (hasContent && (!this.aggregateRosterObserved || existing)) { + this.upsert(id, { + backgrounded: patch.is_backgrounded === true || existing?.backgrounded === true, + kind: patchKind ?? existing?.kind ?? 'unknown', + description: taskDescription(patch.description), + name: taskName(patch), + state: liveState ?? undefined, + startedAt: this.now() + }) + return true + } + return false + } + private replaceAggregateRoster(value: unknown): void { if (!Array.isArray(value)) { return } + const prior = new Map(this.tasks) this.aggregateRosterObserved = true this.tasks.clear() this.terminalTaskIds.clear() @@ -180,29 +198,33 @@ export class ClaudeBackgroundTaskTracker { if (!task || task.ambient === true) { continue } - const id = claudeTaskId(task) + const id = taskId(task) if (!id) { continue } + // An authoritative live roster supersedes an earlier terminal edge. + const retained = this.retention.resume(id) + const existing = prior.get(id) ?? retained + const kind = classifyClaudeBackgroundTaskKind(task.task_type) this.tasks.set(id, { backgrounded: true, - kind: classifyClaudeBackgroundTaskKind(task.task_type), - description: claudeTaskDescription(task.description) + kind: kind !== 'unknown' ? kind : (existing?.kind ?? 'unknown'), + description: taskDescription(task.description) ?? existing?.description, + name: taskName(task) ?? existing?.name, + state: liveClaudeTaskRunState(task.status) ?? existing?.state, + startedAt: existing?.startedAt ?? this.now(), + totalTokens: existing?.totalTokens }) } + for (const [id, task] of prior) { + if (task.backgrounded && !this.tasks.has(id)) { + this.retention.rememberRemoved(id, task) + } + } } - private upsert(id: string, task: TrackedTask): void { - const existing = this.tasks.get(id) - if (existing) { - this.tasks.set(id, { - backgrounded: existing.backgrounded || task.backgrounded, - kind: existing.kind === 'unknown' ? task.kind : existing.kind, - description: task.description ?? existing.description - }) - return - } - if (this.tasks.size >= MAX_TRACKED_TASKS) { + private upsert(id: string, task: TrackedClaudeBackgroundTask): void { + if (!this.tasks.has(id) && this.tasks.size >= MAX_TRACKED_TASKS) { let foregroundId: string | undefined for (const [candidateId, candidate] of this.tasks) { if (!candidate.backgrounded) { @@ -215,6 +237,20 @@ export class ClaudeBackgroundTaskTracker { } this.tasks.delete(foregroundId) } + const existing = this.tasks.get(id) ?? this.retention.resume(id) + this.terminalTaskIds.delete(id) + if (existing) { + this.tasks.set(id, { + backgrounded: existing.backgrounded || task.backgrounded, + kind: task.kind !== 'unknown' ? task.kind : existing.kind, + description: task.description ?? existing.description, + name: task.name ?? existing.name, + state: task.state ?? existing.state, + startedAt: existing.startedAt, + totalTokens: existing.totalTokens + }) + return + } this.tasks.set(id, task) } @@ -231,9 +267,12 @@ export class ClaudeBackgroundTaskTracker { } private refreshMonitoring(): boolean { - const details = this.foregroundTurnActive ? [] : this.backgroundTaskDetails() + const details = this.backgroundTaskDetails() + if (details.length === 0 && this.retention.hasSettled) { + this.retention.flushSettled() + } const next = details.length > 0 - const fingerprint = next ? JSON.stringify(details) : '' + const fingerprint = next ? JSON.stringify([details, this.retention.settledDetails()]) : '' if (next === this.monitoring && fingerprint === this.publishedTasksFingerprint) { return false } @@ -248,11 +287,7 @@ export class ClaudeBackgroundTaskTracker { if (!task.backgrounded) { continue } - details.push({ - id, - kind: task.kind, - ...(task.description ? { description: task.description } : {}) - }) + details.push(claudeBackgroundTaskDetail(id, task)) } return details } diff --git a/src/main/claude/claude-settled-background-tasks.ts b/src/main/claude/claude-settled-background-tasks.ts new file mode 100644 index 00000000000..91e975d07bb --- /dev/null +++ b/src/main/claude/claude-settled-background-tasks.ts @@ -0,0 +1,127 @@ +// Retention state for background tasks that have reached a terminal edge. +// +// The real producer settles a task in two steps inside one tick: +// `background_tasks_changed` arrives FIRST with the task already absent, then +// `task_updated` / `task_notification` carry the outcome. So the terminal edge +// must be able to settle a task the live roster no longer holds — that is what +// `rememberRemoved` preserves. A removal whose outcome frame never arrives +// simply vanishes: removed tasks are never rendered and never guessed into a +// finished state. + +import type { + AgentSessionBackgroundTask, + AgentSessionBackgroundTaskRunState +} from '../../shared/agent-session-wire' + +const MAX_RETAINED_TASKS = 256 + +export type TrackedClaudeBackgroundTask = { + backgrounded: boolean + kind: AgentSessionBackgroundTask['kind'] + description?: string + name?: string + state?: AgentSessionBackgroundTaskRunState + /** First-observed epoch ms; preserved across updates and roster replacement + * so clients can render elapsed and keep a stable first-seen sort. */ + startedAt: number + totalTokens?: number +} + +export function claudeBackgroundTaskDetail( + id: string, + task: TrackedClaudeBackgroundTask +): AgentSessionBackgroundTask { + return { + id, + kind: task.kind, + ...(task.description ? { description: task.description } : {}), + ...(task.name ? { name: task.name } : {}), + state: task.state ?? (task.kind === 'monitor' ? 'monitoring' : 'working'), + startedAt: task.startedAt, + ...(task.totalTokens !== undefined ? { totalTokens: task.totalTokens } : {}) + } +} + +function setBounded(map: Map, key: K, value: V): void { + map.delete(key) + map.set(key, value) + if (map.size > MAX_RETAINED_TASKS) { + const oldest = map.keys().next() + if (!oldest.done) { + map.delete(oldest.value) + } + } +} + +export class ClaudeSettledBackgroundTasks { + private readonly settled = new Map() + private readonly recentlyRemoved = new Map() + + /** An aggregate roster evicted a still-live backgrounded task; hold its + * details so the outcome frame trailing in the same tick can settle it. */ + rememberRemoved(id: string, task: TrackedClaudeBackgroundTask): void { + setBounded(this.recentlyRemoved, id, task) + } + + /** Terminal edge for `id`. `liveSource` is the live roster's entry when it + * still has one; otherwise the recently-removed copy is consumed. A second + * edge (updated, then notification) re-derives the settled state and can + * add the final usage the first edge lacked. */ + settle( + id: string, + state: AgentSessionBackgroundTaskRunState, + outcome: { totalTokens?: number }, + liveSource: TrackedClaudeBackgroundTask | undefined + ): void { + const source = liveSource ?? this.recentlyRemoved.get(id) + const already = this.settled.get(id) + if (source?.backgrounded) { + setBounded(this.settled, id, { + ...claudeBackgroundTaskDetail(id, { + ...source, + totalTokens: outcome.totalTokens ?? source.totalTokens + }), + state + }) + } else if (already) { + this.settled.set(id, { + ...already, + state, + ...(outcome.totalTokens !== undefined ? { totalTokens: outcome.totalTokens } : {}) + }) + } + this.recentlyRemoved.delete(id) + } + + /** Positive live evidence transfers identity back to the tracker, never the old outcome. */ + resume(id: string): TrackedClaudeBackgroundTask | undefined { + const settled = this.settled.get(id) + const removed = this.recentlyRemoved.get(id) + this.settled.delete(id) + this.recentlyRemoved.delete(id) + const source = settled ?? removed + if (!source || source.startedAt === undefined) { + return undefined + } + return { ...source, backgrounded: true, state: undefined, startedAt: source.startedAt } + } + + get hasSettled(): boolean { + return this.settled.size > 0 + } + + settledDetails(): AgentSessionBackgroundTask[] { + return [...this.settled.values()] + } + + /** Settled context only makes sense beside live work; the strip exits at the + * same instant it always has — when the last live task ends. */ + flushSettled(): void { + this.settled.clear() + } + + clear(): void { + this.settled.clear() + this.recentlyRemoved.clear() + } +} diff --git a/src/main/claude/claude-structured-session-close.test.ts b/src/main/claude/claude-structured-session-close.test.ts index 670c0daaf8c..0de5049a71c 100644 --- a/src/main/claude/claude-structured-session-close.test.ts +++ b/src/main/claude/claude-structured-session-close.test.ts @@ -55,7 +55,9 @@ describe('Claude published session close lifecycle', () => { expect(backgroundStates).toEqual([ { state: 'monitoring', - tasks: [{ id: 'background-1', kind: 'agent' }], + tasks: [ + { id: 'background-1', kind: 'agent', state: 'working', startedAt: expect.any(Number) } + ], supportsTaskStop: true } ]) @@ -74,7 +76,9 @@ describe('Claude published session close lifecycle', () => { expect(backgroundStates).toEqual([ { state: 'monitoring', - tasks: [{ id: 'background-1', kind: 'agent' }], + tasks: [ + { id: 'background-1', kind: 'agent', state: 'working', startedAt: expect.any(Number) } + ], supportsTaskStop: true }, null diff --git a/src/main/native-chat/agent-session-wire/structured-agent-session-background-task-channel.ts b/src/main/native-chat/agent-session-wire/structured-agent-session-background-task-channel.ts index 4592dea26e9..6dd9db80e1d 100644 --- a/src/main/native-chat/agent-session-wire/structured-agent-session-background-task-channel.ts +++ b/src/main/native-chat/agent-session-wire/structured-agent-session-background-task-channel.ts @@ -21,7 +21,10 @@ export class StructuredAgentSessionBackgroundTaskChannel { private readonly requireSession: (sessionId: string) => StructuredAgentSessionHostSession, private readonly handoffStatus: ( sessionId: string - ) => Parameters[0]['handoff'] + ) => Parameters[0]['handoff'], + /** Task edges change the status summary too; the feed's equality check + * keeps a no-op re-projection from reaching subscribers. */ + private readonly onPublished: (sessionId: string) => void ) {} history(request: AgentSessionHistoryRequest): AgentSessionHistoryResult { @@ -53,6 +56,7 @@ export class StructuredAgentSessionBackgroundTaskChannel { const state = publishedState !== undefined ? publishedState : this.state(sessionId) if (session && state !== undefined) { this.subscribers.backgroundTasks(sessionId, state, session.fence) + this.onPublished(sessionId) } } diff --git a/src/main/native-chat/agent-session-wire/structured-agent-session-host.ts b/src/main/native-chat/agent-session-wire/structured-agent-session-host.ts index 2df2fdf5412..358ca952b57 100644 --- a/src/main/native-chat/agent-session-wire/structured-agent-session-host.ts +++ b/src/main/native-chat/agent-session-wire/structured-agent-session-host.ts @@ -53,7 +53,8 @@ import type { StructuredAgentSessionHostSession, StructuredAgentSessionReveal } from './structured-agent-session-host-types' -import { StructuredAgentSessionStatusFeed } from './structured-agent-session-status-feed' +import { createStructuredAgentSessionHostStatusFeed } from './structured-agent-session-status-feed' +import type { StructuredAgentSessionStatusSubscriber } from './structured-agent-session-status-feed' import { StructuredAgentSessionEventRecovery } from './structured-agent-session-event-recovery' import { StructuredAgentSessionBackgroundTaskChannel } from './structured-agent-session-background-task-channel' export type { StructuredAgentSessionHostDeps } from './structured-agent-session-host-types' @@ -64,11 +65,10 @@ export class StructuredAgentSessionHost { this ) private readonly sessions = new Map() - private readonly statusFeed = new StructuredAgentSessionStatusFeed({ + private readonly statusFeed = createStructuredAgentSessionHostStatusFeed({ sessions: this.sessions, - getRecord: (sessionId) => this.deps.store.getRecord(sessionId), now: () => this.now(), - onStatusChanged: (summary, options) => this.deps.onSessionStatusChanged?.(summary, options) + deps: () => this.deps }) private readonly subscribers = new AgentSessionSubscribers({ readCommands: (sessionId) => this.deps.adapter.readCommands?.(sessionId), @@ -91,7 +91,8 @@ export class StructuredAgentSessionHost { this.sessions, this.subscribers, (sessionId) => this.requireSession(sessionId), - (sessionId) => this.handoffs.status(sessionId) + (sessionId) => this.handoffs.status(sessionId), + (sessionId) => this.statusFeed.publish(sessionId) ) this.runtimeState = new StructuredAgentSessionHostRuntimeState( deps, @@ -345,7 +346,7 @@ export class StructuredAgentSessionHost { unsubscribe = (sessionId: string, id: string): void => this.subscribers.close(sessionId, id) /** Every session's projected status for session lists; unlike `subscribe`, retains nothing. */ - subscribeStatus: StructuredAgentSessionStatusFeed['subscribe'] = (subscriber) => + subscribeStatus = (subscriber: StructuredAgentSessionStatusSubscriber): (() => void) => this.statusFeed.subscribe(subscriber) private requireSession(sessionId: string): StructuredAgentSessionHostSession { diff --git a/src/main/native-chat/agent-session-wire/structured-agent-session-status-feed.test.ts b/src/main/native-chat/agent-session-wire/structured-agent-session-status-feed.test.ts index 7bd53742430..5fbd58016d0 100644 --- a/src/main/native-chat/agent-session-wire/structured-agent-session-status-feed.test.ts +++ b/src/main/native-chat/agent-session-wire/structured-agent-session-status-feed.test.ts @@ -1,9 +1,12 @@ import { mkdtemp, rm } from 'node:fs/promises' import { tmpdir } from 'node:os' import { join } from 'node:path' -import { afterEach, beforeEach, describe, expect, it } from 'vitest' +import { afterEach, beforeEach, describe, expect, it, vi } from 'vitest' import type { AgentSessionRecord } from '../../../shared/agent-session-record' -import type { AgentSessionStatusEvent } from '../../../shared/agent-session-wire' +import type { + AgentSessionBackgroundTask, + AgentSessionStatusEvent +} from '../../../shared/agent-session-wire' import { createClaudeJournalTranslator } from '../../claude/claude-structured-journal-translation' import { publishCodexTurnLifecycle } from '../../codex/codex-structured-journal-translation-turns' import { createDeferredStructuredAgentSessionEventSink } from './structured-agent-session-event-sink' @@ -74,11 +77,13 @@ function feedFor( { journal: Awaited>; hasProviderChild?: boolean; fence?: number } >, record: Partial | null = null, - onStatusChanged?: StructuredAgentSessionStatusFeedDeps['onStatusChanged'] + onStatusChanged?: StructuredAgentSessionStatusFeedDeps['onStatusChanged'], + readBackgroundTasks?: StructuredAgentSessionStatusFeedDeps['readBackgroundTasks'] ) { let now = 1_000 const feed = new StructuredAgentSessionStatusFeed({ ...(onStatusChanged ? { onStatusChanged } : {}), + ...(readBackgroundTasks ? { readBackgroundTasks } : {}), sessions: { get: (sessionId: string) => { const session = sessions.get(sessionId) @@ -621,6 +626,147 @@ describe('StructuredAgentSessionStatusFeed', () => { session: expect.objectContaining({ status: 'idle', latestPrompt: 'hello' }) }) }) + + it('reuses the journal projection across task progress and invalidates on journal changes', async () => { + const journal = await openJournal() + await journal.appendItem( + USER_IDENTITY, + { kind: 'message', role: 'user', blocks: [{ type: 'text', text: 'fan out' }] }, + { fence: 1 } + ) + await journal.appendItem( + TURN_IDENTITY, + { kind: 'status', text: 'Working', turnLifecycle: { turnId: 'turn-1', state: 'running' } }, + { fence: 1 } + ) + const snapshot = vi.spyOn(journal, 'snapshot') + let taskState: 'working' | 'waiting' = 'working' + const { feed, events } = feedFor(new Map([[SESSION, { journal }]]), null, undefined, () => ({ + state: 'monitoring', + tasks: [{ id: 'child', kind: 'agent', state: taskState }] + })) + for (let tick = 1; tick <= 100; tick++) { + taskState = tick % 2 === 1 ? 'waiting' : 'working' + feed.publish(SESSION) + } + expect(events).toHaveLength(101) + expect(snapshot).toHaveBeenCalledTimes(1) + expect(events.at(-1)).toMatchObject({ + type: 'status', + session: { status: 'working', backgroundTasks: [{ state: 'working' }] } + }) + await journal.appendTombstone(TURN_IDENTITY, { fence: 1 }) + feed.publish(SESSION) + expect(snapshot).toHaveBeenCalledTimes(2) + expect(events.at(-1)).toMatchObject({ type: 'status', session: { status: 'idle' } }) + }) + + it('invalidates cached status on unreadability and keeps record metadata live', async () => { + const journal = await openJournal() + await journal.appendItem( + USER_IDENTITY, + { kind: 'message', role: 'user', blocks: [{ type: 'text', text: 'hello' }] }, + { fence: 1 } + ) + const record = { options: { model: 'first-model' }, providerHandleChain: [] } + const { feed, events } = feedFor(new Map([[SESSION, { journal }]]), record) + record.options.model = 'second-model' + feed.publish(SESSION) + expect(events.at(-1)).toMatchObject({ + type: 'status', + session: { status: 'idle', model: 'second-model' } + }) + const readOnly = vi.spyOn(journal, 'isReadOnly', 'get').mockReturnValue(true) + feed.publish(SESSION) + expect(events.at(-1)).toMatchObject({ type: 'status', session: { status: null } }) + readOnly.mockRestore() + feed.publish(SESSION) + expect(events.at(-1)).toMatchObject({ type: 'status', session: { status: 'idle' } }) + }) + + it('projects live background tasks and republishes a task-only state change', async () => { + const journal = await openJournal() + let tasks = [ + { id: 'task-1', kind: 'agent' as const, name: 'deep_review', state: 'working' as const } + ] + const { feed, events } = feedFor(new Map([[SESSION, { journal }]]), null, undefined, () => ({ + state: 'monitoring', + tasks + })) + await journal.appendItem( + USER_IDENTITY, + { kind: 'message', role: 'user', blocks: [{ type: 'text', text: 'fan out' }] }, + { fence: 1 } + ) + feed.publish(SESSION, journal) + expect(events.at(-1)).toEqual({ + type: 'status', + session: expect.objectContaining({ + backgroundTasks: [{ id: 'task-1', kind: 'agent', name: 'deep_review', state: 'working' }] + }) + }) + + // No journal change: only the task state moved. + tasks = [{ id: 'task-1', kind: 'agent', name: 'deep_review', state: 'waiting' as never }] + const before = events.length + feed.publish(SESSION, journal) + expect(events).toHaveLength(before + 1) + expect(events.at(-1)).toEqual({ + type: 'status', + session: expect.objectContaining({ + backgroundTasks: [expect.objectContaining({ state: 'waiting' })] + }) + }) + + // An identical projection is suppressed. + feed.publish(SESSION, journal) + expect(events).toHaveLength(before + 1) + }) + + it('omits task usage so a progress tick never re-broadcasts the summary', async () => { + const journal = await openJournal() + let tasks: AgentSessionBackgroundTask[] = [ + { id: 'task-1', kind: 'agent', name: 'deep_review', state: 'working', totalTokens: 10 } + ] + const { feed, events } = feedFor(new Map([[SESSION, { journal }]]), null, undefined, () => ({ + state: 'monitoring', + tasks + })) + await journal.appendItem( + USER_IDENTITY, + { kind: 'message', role: 'user', blocks: [{ type: 'text', text: 'fan out' }] }, + { fence: 1 } + ) + feed.publish(SESSION, journal) + const before = events.length + + // A `task_progress` frame moves only usage, which no status-summary reader renders; + // re-broadcasting the whole summary per frame would cost every remote subscriber. + tasks = [ + { id: 'task-1', kind: 'agent', name: 'deep_review', state: 'working', totalTokens: 4_200 } + ] + feed.publish(SESSION, journal) + expect(events).toHaveLength(before) + expect(events.at(-1)).toEqual({ + type: 'status', + session: expect.objectContaining({ + backgroundTasks: [{ id: 'task-1', kind: 'agent', name: 'deep_review', state: 'working' }] + }) + }) + + // A state change on the same task still reaches subscribers. + tasks = [ + { id: 'task-1', kind: 'agent', name: 'deep_review', state: 'waiting', totalTokens: 4_200 } + ] + feed.publish(SESSION, journal) + expect(events).toHaveLength(before + 1) + expect(events.at(-1)).toEqual({ + type: 'status', + session: expect.objectContaining({ + backgroundTasks: [expect.objectContaining({ state: 'waiting' })] + }) + }) + }) }) /** diff --git a/src/main/native-chat/agent-session-wire/structured-agent-session-status-feed.ts b/src/main/native-chat/agent-session-wire/structured-agent-session-status-feed.ts index 3c4ce4b97a6..bddb9ae93ce 100644 --- a/src/main/native-chat/agent-session-wire/structured-agent-session-status-feed.ts +++ b/src/main/native-chat/agent-session-wire/structured-agent-session-status-feed.ts @@ -14,9 +14,11 @@ import { agentProviderSessionsEqual } from '../../../shared/agent-session-resume import type { AgentSessionRecord } from '../../../shared/agent-session-record' import { normalizeOptionalField } from '../../../shared/agent-status-field-normalization' import { AGENT_MODEL_MAX_LENGTH } from '../../../shared/agent-status-types' -import type { - AgentSessionStatusEvent, - AgentSessionStatusSummary +import { + agentSessionBackgroundTasksEqual, + type AgentSessionBackgroundTaskState, + type AgentSessionStatusEvent, + type AgentSessionStatusSummary } from '../../../shared/agent-session-wire' import { projectStructuredAgentSessionStatusSummary } from '../../../shared/structured-agent-session-projection' import type { AgentSessionJournal } from '../agent-session-journal/journal-store' @@ -41,6 +43,9 @@ export type StructuredAgentSessionStatusFeedDeps = { /** Every projection change, whether or not anyone is subscribed. `replay` marks a re-projection * of state the host already knew (restore, an arriving subscriber) rather than a journal edge. */ onStatusChanged?: (summary: AgentSessionStatusSummary, options: { replay: boolean }) => void + /** Live provider-owned background tasks for the summary, so session lists can + * render subagent children. Optional: a provider without the hook projects none. */ + readBackgroundTasks?: (sessionId: string) => AgentSessionBackgroundTaskState | null | undefined } function summariesEqual(a: AgentSessionStatusSummary, b: AgentSessionStatusSummary): boolean { @@ -57,13 +62,50 @@ function summariesEqual(a: AgentSessionStatusSummary, b: AgentSessionStatusSumma a.toolName === b.toolName && a.toolInput === b.toolInput && a.lastAssistantMessage === b.lastAssistantMessage && + agentSessionBackgroundTasksEqual(a.backgroundTasks, b.backgroundTasks) && agentProviderSessionsEqual(undefined, a.providerSession, b.providerSession) ) } +/** Wire the host's own deps into a feed; keeps the host at one call site. + * `deps` is a thunk because the host builds the feed in a field initializer, + * before its constructor parameters are assigned. */ +export function createStructuredAgentSessionHostStatusFeed(args: { + sessions: StructuredAgentSessionStatusFeedDeps['sessions'] + now: () => number + deps: () => { + store: { getRecord: (sessionId: string) => AgentSessionRecord | null } + adapter: { + backgroundTaskState?: ( + sessionId: string + ) => AgentSessionBackgroundTaskState | null | undefined + } + onSessionStatusChanged?: StructuredAgentSessionStatusFeedDeps['onStatusChanged'] + } +}): StructuredAgentSessionStatusFeed { + return new StructuredAgentSessionStatusFeed({ + sessions: args.sessions, + getRecord: (sessionId) => args.deps().store.getRecord(sessionId), + now: args.now, + onStatusChanged: (summary, options) => args.deps().onSessionStatusChanged?.(summary, options), + readBackgroundTasks: (sessionId) => args.deps().adapter.backgroundTaskState?.(sessionId) + }) +} + export class StructuredAgentSessionStatusFeed { private readonly subscribers = new Map() private readonly published = new Map() + // Task progress must not sort and scan an unchanged conversation. Journal identity owns cleanup. + private readonly journalProjections = new WeakMap< + AgentSessionJournal, + { + epoch: string + sequence: number + readOnly: boolean + fence: number | undefined + summary: ReturnType + } + >() constructor(private readonly deps: StructuredAgentSessionStatusFeedDeps) {} @@ -154,24 +196,54 @@ export class StructuredAgentSessionStatusFeed { journal: AgentSessionJournal ): AgentSessionStatusSummary { // An unreadable journal projects as "no turn": the chat itself shows the reset. - const snapshot = journal.isReadOnly ? null : journal.snapshot() - const items = snapshot?.items ?? [] - const submissions = snapshot?.submissions ?? [] + const cursor = journal.cursor() + const readOnly = journal.isReadOnly + const fence = session.fence + let projection = this.journalProjections.get(journal) + if ( + !projection || + projection.epoch !== cursor.epoch || + projection.sequence !== cursor.sequence || + projection.readOnly !== readOnly || + projection.fence !== fence + ) { + // A journalled submission bumps `lastSequence`, so the send-time working + // signal reaches the cache; the lease fence does not, hence the extra key. + const snapshot = readOnly ? null : journal.snapshot() + projection = { + ...cursor, + readOnly, + fence, + summary: projectStructuredAgentSessionStatusSummary( + snapshot?.items ?? [], + snapshot?.submissions ?? [], + fence + ) + } + this.journalProjections.set(journal, projection) + } const record = this.deps.getRecord(sessionId) const providerSession = structuredAgentSessionProviderSessionMetadata(record) // The journal has no model: the record's acknowledged options are where an owner // handoff or a mid-session switch lands, so the row follows whichever is in force. const model = normalizeOptionalField(record?.options?.model, AGENT_MODEL_MAX_LENGTH) + // Usage is dropped here on purpose: a `task_progress` tick would otherwise fail the + // equality check and re-broadcast a full summary to every remote subscriber for a + // number no session list renders. Tokens stay live on the background-task channel. + const backgroundTasks = this.deps + .readBackgroundTasks?.(sessionId) + ?.tasks?.map(({ totalTokens: _totalTokens, ...task }) => task) return { sessionId, workspaceId: session.params.location.workspaceId, agent: session.params.provider, ...(session.hasProviderChild ? { hostExecutionOwned: true as const } : {}), - ...projectStructuredAgentSessionStatusSummary(items, submissions, session.fence), + ...projection.summary, ...(record?.rewind?.phase === 'prepared' || record?.rewind?.phase === 'provider-succeeded' ? { rewindBlockedReason: 'outcome-unknown' as const } : {}), ...(model ? { model } : {}), + ...(backgroundTasks && backgroundTasks.length > 0 ? { backgroundTasks } : {}), ...(providerSession ? { providerSession } : {}), updatedAt: journal.lastActivityAt() || this.deps.now() } diff --git a/src/main/observability/local-file-sink-memory.test.ts b/src/main/observability/local-file-sink-memory.test.ts index fef5b8d00dd..2c6644694b9 100644 --- a/src/main/observability/local-file-sink-memory.test.ts +++ b/src/main/observability/local-file-sink-memory.test.ts @@ -2,11 +2,7 @@ import { mkdtempSync, readFileSync, rmSync, statSync } from 'node:fs' import { tmpdir } from 'node:os' import { join } from 'node:path' import { afterEach, beforeEach, describe, expect, it, vi } from 'vitest' -import { - createLocalFileSink, - DROPPED_RECORD_TYPE, - type LocalFileSink -} from './local-file-sink' +import { createLocalFileSink, DROPPED_RECORD_TYPE, type LocalFileSink } from './local-file-sink' function parseLine(raw: string): Record { return JSON.parse(raw) as Record diff --git a/src/main/runtime/orca-runtime-tests/worktree-ps-structured-host.spec.ts b/src/main/runtime/orca-runtime-tests/worktree-ps-structured-host.spec.ts index 58e7de1b674..5779c65edda 100644 --- a/src/main/runtime/orca-runtime-tests/worktree-ps-structured-host.spec.ts +++ b/src/main/runtime/orca-runtime-tests/worktree-ps-structured-host.spec.ts @@ -53,6 +53,8 @@ function journalWith(prompt: string): AgentSessionJournal { return { isReadOnly: false, lastActivityAt: () => OBSERVED_AT, + // This journal never changes, so a real one would hold its cursor steady. + cursor: () => ({ epoch: 1, sequence: 1 }), snapshot: () => ({ items: runningTurn(prompt) }) } as unknown as AgentSessionJournal } diff --git a/src/main/runtime/rpc/methods/structured-agent-session-rpc.test-fixture.ts b/src/main/runtime/rpc/methods/structured-agent-session-rpc.test-fixture.ts index a702bda5afc..f9f60ded887 100644 --- a/src/main/runtime/rpc/methods/structured-agent-session-rpc.test-fixture.ts +++ b/src/main/runtime/rpc/methods/structured-agent-session-rpc.test-fixture.ts @@ -97,6 +97,7 @@ function statusFeed(): StructuredAgentSessionStatusFeed { { journal: { isReadOnly: false, + cursor: () => ({ epoch: 'epoch-status', sequence: 2 }), lastActivityAt: () => 2, snapshot: () => ({ items: STATUS_ITEMS }) } as unknown as AgentSessionJournal, diff --git a/src/renderer/src/components/native-chat/NativeChatAutocompleteMenus.test.tsx b/src/renderer/src/components/native-chat/NativeChatAutocompleteMenus.test.tsx index d5dac1efab5..88506c51838 100644 --- a/src/renderer/src/components/native-chat/NativeChatAutocompleteMenus.test.tsx +++ b/src/renderer/src/components/native-chat/NativeChatAutocompleteMenus.test.tsx @@ -13,6 +13,7 @@ function autocomplete( query: '', triggerKey: '/:0', prefix: '/', + dispatchable: true, grouped: true, commandsEnabled: true, skillsEnabled: true, @@ -21,6 +22,7 @@ function autocomplete( kind: 'command', id: 'command:clear', name: 'clear', + token: '/clear', description: 'Clear history', skillCollision: false }, @@ -28,6 +30,7 @@ function autocomplete( kind: 'skill', id: 'skill:browser', name: 'browser', + token: '/browser', description: 'Use a browser', sources: [{ sourceKind: 'repo', skillFilePath: '/repo/browser/SKILL.md' }] } diff --git a/src/renderer/src/components/native-chat/NativeChatAutocompleteMenus.tsx b/src/renderer/src/components/native-chat/NativeChatAutocompleteMenus.tsx index 8357931fb49..7f1cd943c0f 100644 --- a/src/renderer/src/components/native-chat/NativeChatAutocompleteMenus.tsx +++ b/src/renderer/src/components/native-chat/NativeChatAutocompleteMenus.tsx @@ -12,7 +12,7 @@ export const NativeChatPickerMenu = memo(function NativeChatPickerMenu({ onChoose, onRetry }: { - autocomplete: Extract + autocomplete: Extract activeIndex: number listboxId: string onChoose: (item: NativeChatPickerItem) => void @@ -54,7 +54,6 @@ export const NativeChatPickerMenu = memo(function NativeChatPickerMenu({ + autocomplete: Extract ): string { - if (autocomplete.mode === 'skill' || !autocomplete.commandsEnabled) { + if (!autocomplete.commandsEnabled) { return translate('components.native-chat.composer.noSkills', 'No matching skills') } if (autocomplete.skillsEnabled) { @@ -174,7 +172,6 @@ function PickerStatus({ children }: { children: React.ReactNode }): React.JSX.El function PickerOption({ item, - prefix, index, activeIndex, listboxId, @@ -182,7 +179,6 @@ function PickerOption({ onChoose }: { item: NativeChatPickerItem - prefix: '/' | '$' index: number activeIndex: number listboxId: string @@ -213,7 +209,7 @@ function PickerOption({ ) : null} - {prefix + item.name} + {item.token} {item.description ? ( {item.description} ) : null} diff --git a/src/renderer/src/components/native-chat/NativeChatBackgroundTasksStatus.test.tsx b/src/renderer/src/components/native-chat/NativeChatBackgroundTasksStatus.test.tsx index e137220dd66..96be91820a3 100644 --- a/src/renderer/src/components/native-chat/NativeChatBackgroundTasksStatus.test.tsx +++ b/src/renderer/src/components/native-chat/NativeChatBackgroundTasksStatus.test.tsx @@ -2,12 +2,16 @@ import '@testing-library/jest-dom/vitest' -import { cleanup, fireEvent, render, screen } from '@testing-library/react' +import { act, cleanup, fireEvent, render, screen, within } from '@testing-library/react' +import { Profiler } from 'react' import { afterEach, describe, expect, it, vi } from 'vitest' import type { AgentSessionBackgroundTask } from '../../../../shared/agent-session-wire' import { NativeChatBackgroundTasksStatus } from './NativeChatBackgroundTasksStatus' -afterEach(cleanup) +afterEach(() => { + cleanup() + vi.useRealTimers() +}) const TASKS: AgentSessionBackgroundTask[] = [ { id: 'codex-agent:child-1', kind: 'agent', description: 'count_a' }, @@ -20,7 +24,10 @@ function renderStrip(props: { supportsTaskStop: boolean; supportsStopAll: boolea const onStop = vi.fn() render( { expect(screen.getByText('sleep 90')).toBeInTheDocument() }) }) + +describe('background-tasks strip header', () => { + function renderHeader(tasks: AgentSessionBackgroundTask[]): HTMLElement { + render( + {}} + /> + ) + return screen.getByRole('button', { expanded: false }) + } + + it('leads each kind segment with that kind icon and keeps the counts in the accessible name', () => { + const header = renderHeader([ + { id: 'a1', kind: 'agent' }, + { id: 'a2', kind: 'agent' }, + { id: 'a3', kind: 'agent' }, + { id: 'm1', kind: 'monitor' } + ]) + expect(header).toHaveAttribute('aria-label', '3 agents · 1 monitor') + expect(header.querySelector('.lucide-bot')).toBeInTheDocument() + // Heartbeat, the same glyph the agent sidebar shows for monitoring. + expect(header.querySelector('.lucide-activity')).toBeInTheDocument() + // Two kind icons and the chevron: the aggregate state dot is gone. + expect(header.querySelectorAll('svg')).toHaveLength(3) + for (const icon of header.querySelectorAll('svg')) { + expect(icon).toHaveAttribute('aria-hidden', 'true') + } + }) + + it('gives the monitor heartbeat the sidebar amber and leaves other kinds neutral', () => { + const header = renderHeader([ + { id: 'a1', kind: 'agent' }, + { id: 'm1', kind: 'monitor' } + ]) + // Same glyph AND same colour as AgentStateDot/StatusIndicator, or a monitor + // here does not read as the monitor there. + expect(header.querySelector('.lucide-activity')?.classList).toContain('text-yellow-500') + expect(header.querySelector('.lucide-bot')?.classList).toContain('text-muted-foreground') + expect(header.querySelector('.lucide-bot')?.classList).not.toContain('text-yellow-500') + }) + + it('dims the monitor amber while a turn owns the voice', () => { + render( + {}} + /> + ) + const header = screen.getByRole('button', { expanded: false }) + expect(header.querySelector('.lucide-activity')?.classList).toContain('text-yellow-500/40') + }) + + it('carries the monitor amber on the expanded row too', () => { + const header = renderHeader([ + { id: 'm1', kind: 'monitor', description: 'watcher' }, + { id: 'c1', kind: 'command', description: 'sleep 90' } + ]) + fireEvent.click(header) + // Each kind group is its own labelled list, so scope to the monitor one. + const monitors = screen.getByRole('list', { name: 'Monitors' }) + expect(monitors.querySelector('.lucide-activity')?.classList).toContain('text-yellow-500') + const shell = screen.getByRole('list', { name: 'Shell' }) + expect(shell.querySelector('.lucide-square-terminal')?.classList).toContain( + 'text-muted-foreground' + ) + }) + + it('draws the segment separator in a visible text tone, not the divider token', () => { + const header = renderHeader([ + { id: 'a1', kind: 'agent' }, + { id: 'c1', kind: 'command' } + ]) + const separators = [...header.querySelectorAll('span')].filter( + (element) => element.textContent === ' · ' + ) + expect(separators).toHaveLength(1) + // `--border` is a divider line (7% white in dark), an order of magnitude + // fainter than the counts it sits between. + expect(separators[0].classList).not.toContain('text-border') + expect(separators[0].classList).toContain('text-muted-foreground') + // One space either side; the icon's own margin is the icon-to-label gap. + expect(header.textContent).toBe('1 agent · 1 shell') + }) + + it('carries no icon on a collapsed total, which spans kinds', () => { + const header = renderHeader([ + { id: 'a1', kind: 'agent' }, + { id: 'c1', kind: 'command' }, + { id: 'm1', kind: 'monitor' }, + { id: 'w1', kind: 'workflow' } + ]) + expect(header).toHaveAttribute('aria-label', '4 background tasks') + expect(header.querySelectorAll('svg')).toHaveLength(1) + }) +}) + +describe('settled rows beside their live siblings', () => { + // Retention is the PR's headline: a finished child stays visible, keeps the + // usage it ended on, and stops claiming a clock or a stop control. + it('keeps a settled row with its final usage, no clock and no stop', () => { + render( + {}} + /> + ) + fireEvent.click(screen.getByRole('button', { expanded: false })) + const agents = screen.getByRole('list', { name: 'Agents' }) + const rows = within(agents).getAllByRole('listitem') + expect(rows).toHaveLength(2) + // First seen first: the settled sibling started earlier. + expect(rows[0].textContent).toBe('settled child18.1k') + expect(rows[1].textContent).toMatch(/^live child4\.1k · .+Stop$/) + expect(within(rows[1]).getByRole('button', { name: 'Stop live child' })).toBeInTheDocument() + expect(within(rows[0]).queryByRole('button')).toBeNull() + }) +}) + +describe('background-task row reasons', () => { + function expandedRows(tasks: AgentSessionBackgroundTask[]): HTMLElement[] { + render( + {}} + /> + ) + fireEvent.click(screen.getByRole('button', { expanded: false })) + return screen.getAllByRole('listitem') + } + + // `unverifiable` is the SSH verdict for "no contact"; a row that hides it reads + // like a working child. `blocked` is the same class of loss. + it('names the reason on every attention state, not only on waiting', () => { + const rows = expandedRows([ + { id: 'a1', kind: 'agent', description: 'ssh child', state: 'unverifiable' }, + { id: 'a2', kind: 'agent', description: 'flaky child', state: 'blocked' }, + { id: 'a3', kind: 'agent', description: 'approval child', state: 'waiting' }, + { id: 'a4', kind: 'agent', description: 'busy child', state: 'working' } + ]) + expect(rows).toHaveLength(4) + expect(rows[0].textContent).toContain('ssh child · no contact') + expect(rows[1].textContent).toContain('flaky child · failed') + expect(rows[2].textContent).toContain('approval child · needs approval') + // A running row has nothing to explain. + expect(rows[3].textContent).not.toContain('·') + }) +}) + +it('stops elapsed renders in a hidden pane and catches up on reveal', () => { + vi.useFakeTimers() + vi.setSystemTime(100_000) + const committed = vi.fn() + const view = (isVisible: boolean) => ( + + {}} + /> + + ) + const { rerender, unmount } = render(view(true)) + committed.mockClear() + act(() => vi.advanceTimersByTime(1_000)) + expect(committed).toHaveBeenCalled() + rerender(view(false)) + committed.mockClear() + act(() => vi.advanceTimersByTime(10_000)) + expect(committed).not.toHaveBeenCalled() + rerender(view(true)) + committed.mockClear() + act(() => vi.advanceTimersByTime(1_000)) + expect(committed).toHaveBeenCalled() + unmount() + expect(vi.getTimerCount()).toBe(0) +}) diff --git a/src/renderer/src/components/native-chat/NativeChatBackgroundTasksStatus.tsx b/src/renderer/src/components/native-chat/NativeChatBackgroundTasksStatus.tsx index c52b73842a9..21b8d73f258 100644 --- a/src/renderer/src/components/native-chat/NativeChatBackgroundTasksStatus.tsx +++ b/src/renderer/src/components/native-chat/NativeChatBackgroundTasksStatus.tsx @@ -1,62 +1,207 @@ -import { useId, useState } from 'react' -import { ChevronDown } from 'lucide-react' +import { useEffect, useId, useMemo, useRef, useState } from 'react' +import { Activity, Bot, ChevronDown, CircleHelp, SquareTerminal, Workflow } from 'lucide-react' import type { AgentSessionBackgroundTask } from '../../../../shared/agent-session-wire' import { AgentStateDot } from '@/components/AgentStateDot' import { Button } from '@/components/ui/button' +import { useNow } from '@/hooks/use-now' import { translate } from '@/i18n/i18n' +import { backgroundTasksHeaderContent } from './background-task-header-content' +import { + backgroundTaskElapsedLabel, + backgroundTaskGroupLabel, + backgroundTaskStateReason, + buildBackgroundTaskGroups, + formatBackgroundTaskTokens, + type BackgroundRosterTask +} from './background-task-roster' -function backgroundTaskLabel(task: AgentSessionBackgroundTask): string { - if (task.description) { - return task.description - } - switch (task.kind) { - case 'agent': - return translate('components.native-chat.backgroundTasks.agent', 'Background agent') - case 'workflow': - return translate('components.native-chat.backgroundTasks.workflow', 'Background workflow') - case 'command': - return translate('components.native-chat.backgroundTasks.command', 'Background command') - case 'monitor': - return translate('components.native-chat.backgroundTasks.monitor', 'Background monitor') - case 'unknown': - return translate('components.native-chat.backgroundTasks.task', 'Background task') - } +/** Below this strip width (border-box, live root font size) the header drops + * its per-kind breakdown for an honest total. A narrow split pane on a wide + * monitor must behave like a narrow window, so no viewport media query. */ +const NARROW_STRIP_REM = 24 + +function rootFontSizePx(): number { + const parsed = Number.parseFloat(getComputedStyle(document.documentElement).fontSize) + return Number.isFinite(parsed) && parsed > 0 ? parsed : 16 +} + +/** Observe the strip's own border-box width; the viewport is only the + * pre-measurement stand-in before the first observer callback. */ +function useNarrowStrip(ref: React.RefObject): boolean { + const [narrow, setNarrow] = useState(() => window.innerWidth < NARROW_STRIP_REM * 16) + useEffect(() => { + const element = ref.current + if (!element || typeof ResizeObserver === 'undefined') { + return + } + const observer = new ResizeObserver((observerEntries) => { + const width = + observerEntries[0]?.borderBoxSize?.[0]?.inlineSize ?? element.getBoundingClientRect().width + setNarrow(width < NARROW_STRIP_REM * rootFontSizePx()) + }) + observer.observe(element, { box: 'border-box' }) + return () => observer.disconnect() + }, [ref]) + return narrow +} + +const KIND_ICONS = { + agent: Bot, + command: SquareTerminal, + monitor: Activity, + workflow: Workflow, + unknown: CircleHelp +} as const + +/** Monitoring is a STATE the app colours the same on every surface — the agent + * sidebar and `AgentStateDot` both draw an amber heartbeat — so the strip must + * match it or the two stop reading as the same thing. The other four are plain + * kind markers and stay neutral. `dimmed` is the running-turn treatment. */ +function kindIconTone(kind: AgentSessionBackgroundTask['kind'], dimmed: boolean): string { + const tone = kind === 'monitor' ? 'text-yellow-500' : 'text-muted-foreground' + return dimmed ? `${tone}/40` : tone +} + +function BackgroundTaskRow(props: { + entry: BackgroundRosterTask + now: number + supportsTaskStop: boolean + stopping: boolean + onStop: (taskId: string) => void +}): React.JSX.Element { + const { entry, now } = props + const Icon = KIND_ICONS[entry.task.kind] + // Every attention state states its reason on the row, the same ones the collapsed + // header names; `unverifiable` ("no contact") must never be silently dropped. + const reason = backgroundTaskStateReason(entry.state) + // Settled rows keep their final usage but no elapsed — a still-growing clock + // on finished work would lie. + const meta = [ + entry.task.totalTokens !== undefined + ? formatBackgroundTaskTokens(entry.task.totalTokens) + : null, + entry.settled ? null : backgroundTaskElapsedLabel(entry.task, now) + ] + .filter((part): part is string => part !== null) + .join(' · ') + return ( +
  • +
    • + ) } export function NativeChatBackgroundTasksStatus(props: { tasks: readonly AgentSessionBackgroundTask[] + settledTasks: readonly AgentSessionBackgroundTask[] supportsTaskStop: boolean /** False when the provider exposes no honest stop at all; the fallback * control is hidden rather than offering a button that cannot act. */ supportsStopAll: boolean stoppingTaskIds: ReadonlySet stoppingAll: boolean + /** True while the session is idle: only then may the strip speak as the + * animated monitoring indicator. A running turn owns the voice. */ + indicatorActive: boolean + isVisible: boolean onStop: (taskId?: string) => void }): React.JSX.Element { const [expanded, setExpanded] = useState(false) const taskListId = useId() + const stripRef = useRef(null) + const narrow = useNarrowStrip(stripRef) + // The 1 Hz elapsed tick must not re-group, re-sort and re-translate the whole roster. + const groups = useMemo( + () => buildBackgroundTaskGroups(props.tasks, props.settledTasks), + [props.tasks, props.settledTasks] + ) + const singleLiveCommand = + groups.length === 1 && groups[0].kind === 'command' && groups[0].tasks.length === 1 + const hasElapsed = groups.some((group) => + group.tasks.some((entry) => !entry.settled && (entry.task.startedAt ?? 0) > 0) + ) + const now = useNow(1_000, props.isVisible && hasElapsed && (expanded || singleLiveCommand)) + const header = backgroundTasksHeaderContent(groups, { narrow, now }) + const headerText = `${header.segments.map((segment) => segment.text).join(' · ')}${header.detail ? `${header.segments.length > 0 ? ' — ' : ''}${header.detail}` : ''}` return (
    -
    +
    - ) : null} - - ) - })} - + ))} + +
    + )) ) : (

    {translate( @@ -119,7 +250,7 @@ export function NativeChatBackgroundTasksStatus(props: {

    )} {!props.supportsTaskStop && props.supportsStopAll ? ( -
    0 ? 'mt-2 border-t border-border pt-2' : 'mt-2'}> +
    0 ? 'mt-2 border-t border-border pt-2' : 'mt-2'}>